id
stringlengths 9
10
| submitter
stringlengths 1
64
⌀ | authors
stringlengths 4
20.7k
| title
stringlengths 4
246
| comments
stringlengths 1
523
⌀ | journal-ref
stringlengths 4
404
⌀ | doi
stringlengths 11
153
⌀ | report-no
stringlengths 2
254
⌀ | categories
stringlengths 5
98
| license
stringclasses 9
values | orig_abstract
stringlengths 14
3.35k
| versions
listlengths 1
60
| update_date
stringlengths 10
10
| authors_parsed
listlengths 1
1.35k
| abstract
stringlengths 11
3.34k
|
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
1407.7170
|
Amir Leshem
|
Amir Leshem, Maziyar Hamdi, Vikram Krishnamurthy
|
Boundary value problems in consensus networks
|
Submitted for publication, Feb. 2014
| null | null | null |
cs.SI cs.IT math.IT physics.soc-ph
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
This paper studies the effect of boundary value conditions on consensus
networks. Consider a network where some nodes keep their estimates constant
while other nodes average their estimates with that of their neighbors. We
analyze such networks and show that in contrast to standard consensus networks,
the network estimate converges to a general harmonic function on the graph.
Furthermore, the final value depends only on the value at the boundary nodes.
This has important implications in consensus networks -- for example, we show
that consensus networks are extremely sensitive to the existence of a single
malicious node or consistent errors in a single node. We also discuss
applications of this result in social and sensor networks. We investigate the
existence of boundary nodes in human social networks via an experimental study
involving human subjects. Finally, the paper is concluded with the numerical
studies of the boundary value problems in consensus networks.
|
[
{
"created": "Sun, 27 Jul 2014 00:27:42 GMT",
"version": "v1"
}
] |
2014-07-29
|
[
[
"Leshem",
"Amir",
""
],
[
"Hamdi",
"Maziyar",
""
],
[
"Krishnamurthy",
"Vikram",
""
]
] |
This paper studies the effect of boundary value conditions on consensus networks. Consider a network where some nodes keep their estimates constant while other nodes average their estimates with that of their neighbors. We analyze such networks and show that in contrast to standard consensus networks, the network estimate converges to a general harmonic function on the graph. Furthermore, the final value depends only on the value at the boundary nodes. This has important implications in consensus networks -- for example, we show that consensus networks are extremely sensitive to the existence of a single malicious node or consistent errors in a single node. We also discuss applications of this result in social and sensor networks. We investigate the existence of boundary nodes in human social networks via an experimental study involving human subjects. Finally, the paper is concluded with the numerical studies of the boundary value problems in consensus networks.
|
2210.00701
|
Dan Qiao
|
Dan Qiao, Yu-Xiang Wang
|
Near-Optimal Deployment Efficiency in Reward-Free Reinforcement Learning
with Linear Function Approximation
|
48 pages
| null | null | null |
cs.LG cs.AI stat.ML
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
We study the problem of deployment efficient reinforcement learning (RL) with
linear function approximation under the \emph{reward-free} exploration setting.
This is a well-motivated problem because deploying new policies is costly in
real-life RL applications. Under the linear MDP setting with feature dimension
$d$ and planning horizon $H$, we propose a new algorithm that collects at most
$\widetilde{O}(\frac{d^2H^5}{\epsilon^2})$ trajectories within $H$ deployments
to identify $\epsilon$-optimal policy for any (possibly data-dependent) choice
of reward functions. To the best of our knowledge, our approach is the first to
achieve optimal deployment complexity and optimal $d$ dependence in sample
complexity at the same time, even if the reward is known ahead of time. Our
novel techniques include an exploration-preserving policy discretization and a
generalized G-optimal experiment design, which could be of independent
interest. Lastly, we analyze the related problem of regret minimization in
low-adaptive RL and provide information-theoretic lower bounds for switching
cost and batch complexity.
|
[
{
"created": "Mon, 3 Oct 2022 03:48:26 GMT",
"version": "v1"
},
{
"created": "Wed, 22 Feb 2023 00:33:11 GMT",
"version": "v2"
}
] |
2023-02-23
|
[
[
"Qiao",
"Dan",
""
],
[
"Wang",
"Yu-Xiang",
""
]
] |
We study the problem of deployment efficient reinforcement learning (RL) with linear function approximation under the \emph{reward-free} exploration setting. This is a well-motivated problem because deploying new policies is costly in real-life RL applications. Under the linear MDP setting with feature dimension $d$ and planning horizon $H$, we propose a new algorithm that collects at most $\widetilde{O}(\frac{d^2H^5}{\epsilon^2})$ trajectories within $H$ deployments to identify $\epsilon$-optimal policy for any (possibly data-dependent) choice of reward functions. To the best of our knowledge, our approach is the first to achieve optimal deployment complexity and optimal $d$ dependence in sample complexity at the same time, even if the reward is known ahead of time. Our novel techniques include an exploration-preserving policy discretization and a generalized G-optimal experiment design, which could be of independent interest. Lastly, we analyze the related problem of regret minimization in low-adaptive RL and provide information-theoretic lower bounds for switching cost and batch complexity.
|
2406.00002
|
Manos Kamarianakis
|
Achilleas Filippidis, Nikolaos Marmaras, Michael Maravgakis, Alexandra
Plexousaki, Manos Kamarianakis, George Papagiannakis
|
VR Isle Academy: A VR Digital Twin Approach for Robotic Surgical Skill
Development
|
10 pages, 14 figures, Acknowledgement Section updated
| null | null | null |
cs.RO cs.GR
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
Contemporary progress in the field of robotics, marked by improved efficiency
and stability, has paved the way for the global adoption of surgical robotic
systems (SRS). While these systems enhance surgeons' skills by offering a more
accurate and less invasive approach to operations, they come at a considerable
cost. Moreover, SRS components often involve heavy machinery, making the
training process challenging due to limited access to such equipment. In this
paper we introduce a cost-effective way to facilitate training for a simulator
of a SRS via a portable, device-agnostic, ultra realistic simulation with hand
tracking and feet tracking support. Error assessment is accessible in both
real-time and offline, which enables the monitoring and tracking of users'
performance. The VR application has been objectively evaluated by several
untrained testers showcasing significant reduction in error metrics as the
number of training sessions increases. This indicates that the proposed VR
application denoted as VR Isle Academy operates efficiently, improving the
robot - controlling skills of the testers in an intuitive and immersive way
towards reducing the learning curve at minimal cost.
|
[
{
"created": "Sat, 4 May 2024 14:47:42 GMT",
"version": "v1"
},
{
"created": "Mon, 1 Jul 2024 14:41:04 GMT",
"version": "v2"
}
] |
2024-07-02
|
[
[
"Filippidis",
"Achilleas",
""
],
[
"Marmaras",
"Nikolaos",
""
],
[
"Maravgakis",
"Michael",
""
],
[
"Plexousaki",
"Alexandra",
""
],
[
"Kamarianakis",
"Manos",
""
],
[
"Papagiannakis",
"George",
""
]
] |
Contemporary progress in the field of robotics, marked by improved efficiency and stability, has paved the way for the global adoption of surgical robotic systems (SRS). While these systems enhance surgeons' skills by offering a more accurate and less invasive approach to operations, they come at a considerable cost. Moreover, SRS components often involve heavy machinery, making the training process challenging due to limited access to such equipment. In this paper we introduce a cost-effective way to facilitate training for a simulator of a SRS via a portable, device-agnostic, ultra realistic simulation with hand tracking and feet tracking support. Error assessment is accessible in both real-time and offline, which enables the monitoring and tracking of users' performance. The VR application has been objectively evaluated by several untrained testers showcasing significant reduction in error metrics as the number of training sessions increases. This indicates that the proposed VR application denoted as VR Isle Academy operates efficiently, improving the robot - controlling skills of the testers in an intuitive and immersive way towards reducing the learning curve at minimal cost.
|
2011.07069
|
Vigen Arakelyan
|
Jing Geng, Vigen Arakelian (LS2N, RoMas, INSA Rennes), Damien Chablat
(LS2N, ReV, CNRS)
|
Shaking Force Balancing of the Orthoglide
| null |
In: Zeghloul S., Laribi M., Sandoval Arevalo J. (eds) Advances in
Service and Industrial Robotics. RAAD 2020. Mechanisms and Machine Science,
vol 84. Springer, pp.227-234, 2020, 2211-0984
|
10.1007/978-3-030-48989-2_25
| null |
cs.RO
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
The shaking force balancing is a well-known problem in the design of
high-speed robotic systems because the variable dynamic loads cause noises,
wear and fatigue of mechanical structures. Different solutions, for full or
partial shaking force balancing, via internal mass redistribution or by adding
auxiliary links were developed. The paper deals with the shaking force
balancing of the Orthoglide. The suggested solution based on the optimal
acceleration control of the manipulator's common center of mass allows a
significant reduction of the shaking force. Compared with the balancing method
via adding counterweights or auxiliary substructures, the proposed method can
avoid some drawbacks: the increase of the total mass, the overall size and the
complexity of the mechanism, which become especially challenging for special
parallel manipulators. Using the proposed motion control method, the maximal
value of the total mass center acceleration is reduced, as a consequence, the
shaking force of the manipulator decreases. The efficiency of the suggested
method via numerical simulations carried out with ADAMS is demonstrated.
|
[
{
"created": "Fri, 13 Nov 2020 14:57:59 GMT",
"version": "v1"
}
] |
2020-11-17
|
[
[
"Geng",
"Jing",
"",
"LS2N, RoMas, INSA Rennes"
],
[
"Arakelian",
"Vigen",
"",
"LS2N, RoMas, INSA Rennes"
],
[
"Chablat",
"Damien",
"",
"LS2N, ReV, CNRS"
]
] |
The shaking force balancing is a well-known problem in the design of high-speed robotic systems because the variable dynamic loads cause noises, wear and fatigue of mechanical structures. Different solutions, for full or partial shaking force balancing, via internal mass redistribution or by adding auxiliary links were developed. The paper deals with the shaking force balancing of the Orthoglide. The suggested solution based on the optimal acceleration control of the manipulator's common center of mass allows a significant reduction of the shaking force. Compared with the balancing method via adding counterweights or auxiliary substructures, the proposed method can avoid some drawbacks: the increase of the total mass, the overall size and the complexity of the mechanism, which become especially challenging for special parallel manipulators. Using the proposed motion control method, the maximal value of the total mass center acceleration is reduced, as a consequence, the shaking force of the manipulator decreases. The efficiency of the suggested method via numerical simulations carried out with ADAMS is demonstrated.
|
2208.03526
|
Zhikang Wang
|
Zhikang Wang, Yue Bi, Tong Pan, Xiaoyu Wang, Chris Bain, Richard
Bassed, Seiya Imoto, Jianhua Yao, Jiangning Song
|
Multiplex-detection Based Multiple Instance Learning Network for Whole
Slide Image Classification
| null | null | null | null |
cs.CV
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
Multiple instance learning (MIL) is a powerful approach to classify whole
slide images (WSIs) for diagnostic pathology. A fundamental challenge of MIL on
WSI classification is to discover the \textit{critical instances} that trigger
the bag label. However, previous methods are primarily designed under the
independent and identical distribution hypothesis (\textit{i.i.d}), ignoring
either the correlations between instances or heterogeneity of tumours. In this
paper, we propose a novel multiplex-detection-based multiple instance learning
(MDMIL) to tackle the issues above. Specifically, MDMIL is constructed by the
internal query generation module (IQGM) and the multiplex detection module
(MDM) and assisted by the memory-based contrastive loss during training.
Firstly, IQGM gives the probability of instances and generates the internal
query (IQ) for the subsequent MDM by aggregating highly reliable features after
the distribution analysis. Secondly, the multiplex-detection cross-attention
(MDCA) and multi-head self-attention (MHSA) in MDM cooperate to generate the
final representations for the WSI. In this process, the IQ and trainable
variational query (VQ) successfully build up the connections between instances
and significantly improve the model's robustness toward heterogeneous tumours.
At last, to further enforce constraints in the feature space and stabilize the
training process, we adopt a memory-based contrastive loss, which is
practicable for WSI classification even with a single sample as input in each
iteration. We conduct experiments on three computational pathology datasets,
e.g., CAMELYON16, TCGA-NSCLC, and TCGA-RCC datasets. The superior accuracy and
AUC demonstrate the superiority of our proposed MDMIL over other
state-of-the-art methods.
|
[
{
"created": "Sat, 6 Aug 2022 14:36:48 GMT",
"version": "v1"
},
{
"created": "Wed, 31 Aug 2022 12:19:25 GMT",
"version": "v2"
},
{
"created": "Thu, 1 Sep 2022 03:55:04 GMT",
"version": "v3"
}
] |
2022-09-02
|
[
[
"Wang",
"Zhikang",
""
],
[
"Bi",
"Yue",
""
],
[
"Pan",
"Tong",
""
],
[
"Wang",
"Xiaoyu",
""
],
[
"Bain",
"Chris",
""
],
[
"Bassed",
"Richard",
""
],
[
"Imoto",
"Seiya",
""
],
[
"Yao",
"Jianhua",
""
],
[
"Song",
"Jiangning",
""
]
] |
Multiple instance learning (MIL) is a powerful approach to classify whole slide images (WSIs) for diagnostic pathology. A fundamental challenge of MIL on WSI classification is to discover the \textit{critical instances} that trigger the bag label. However, previous methods are primarily designed under the independent and identical distribution hypothesis (\textit{i.i.d}), ignoring either the correlations between instances or heterogeneity of tumours. In this paper, we propose a novel multiplex-detection-based multiple instance learning (MDMIL) to tackle the issues above. Specifically, MDMIL is constructed by the internal query generation module (IQGM) and the multiplex detection module (MDM) and assisted by the memory-based contrastive loss during training. Firstly, IQGM gives the probability of instances and generates the internal query (IQ) for the subsequent MDM by aggregating highly reliable features after the distribution analysis. Secondly, the multiplex-detection cross-attention (MDCA) and multi-head self-attention (MHSA) in MDM cooperate to generate the final representations for the WSI. In this process, the IQ and trainable variational query (VQ) successfully build up the connections between instances and significantly improve the model's robustness toward heterogeneous tumours. At last, to further enforce constraints in the feature space and stabilize the training process, we adopt a memory-based contrastive loss, which is practicable for WSI classification even with a single sample as input in each iteration. We conduct experiments on three computational pathology datasets, e.g., CAMELYON16, TCGA-NSCLC, and TCGA-RCC datasets. The superior accuracy and AUC demonstrate the superiority of our proposed MDMIL over other state-of-the-art methods.
|
2304.00367
|
Peter Du
|
Peter Du, Surya Murthy, Katherine Driggs-Campbell
|
Conveying Autonomous Robot Capabilities through Contrasting Behaviour
Summaries
| null | null | null | null |
cs.RO cs.AI cs.MA
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
As advances in artificial intelligence enable increasingly capable
learning-based autonomous agents, it becomes more challenging for human
observers to efficiently construct a mental model of the agent's behaviour. In
order to successfully deploy autonomous agents, humans should not only be able
to understand the individual limitations of the agents but also have insight on
how they compare against one another. To do so, we need effective methods for
generating human interpretable agent behaviour summaries. Single agent
behaviour summarization has been tackled in the past through methods that
generate explanations for why an agent chose to pick a particular action at a
single timestep. However, for complex tasks, a per-action explanation may not
be able to convey an agents global strategy. As a result, researchers have
looked towards multi-timestep summaries which can better help humans assess an
agents overall capability. More recently, multi-step summaries have also been
used for generating contrasting examples to evaluate multiple agents. However,
past approaches have largely relied on unstructured search methods to generate
summaries and require agents to have a discrete action space. In this paper we
present an adaptive search method for efficiently generating contrasting
behaviour summaries with support for continuous state and action spaces. We
perform a user study to evaluate the effectiveness of the summaries for helping
humans discern the superior autonomous agent for a given task. Our results
indicate that adaptive search can efficiently identify informative contrasting
scenarios that enable humans to accurately select the better performing agent
with a limited observation time budget.
|
[
{
"created": "Sat, 1 Apr 2023 18:20:59 GMT",
"version": "v1"
}
] |
2023-04-04
|
[
[
"Du",
"Peter",
""
],
[
"Murthy",
"Surya",
""
],
[
"Driggs-Campbell",
"Katherine",
""
]
] |
As advances in artificial intelligence enable increasingly capable learning-based autonomous agents, it becomes more challenging for human observers to efficiently construct a mental model of the agent's behaviour. In order to successfully deploy autonomous agents, humans should not only be able to understand the individual limitations of the agents but also have insight on how they compare against one another. To do so, we need effective methods for generating human interpretable agent behaviour summaries. Single agent behaviour summarization has been tackled in the past through methods that generate explanations for why an agent chose to pick a particular action at a single timestep. However, for complex tasks, a per-action explanation may not be able to convey an agents global strategy. As a result, researchers have looked towards multi-timestep summaries which can better help humans assess an agents overall capability. More recently, multi-step summaries have also been used for generating contrasting examples to evaluate multiple agents. However, past approaches have largely relied on unstructured search methods to generate summaries and require agents to have a discrete action space. In this paper we present an adaptive search method for efficiently generating contrasting behaviour summaries with support for continuous state and action spaces. We perform a user study to evaluate the effectiveness of the summaries for helping humans discern the superior autonomous agent for a given task. Our results indicate that adaptive search can efficiently identify informative contrasting scenarios that enable humans to accurately select the better performing agent with a limited observation time budget.
|
1908.09701
|
Huan Zhao
|
Huan Zhao, Yingqi Zhou, Yangqiu Song, Dik Lun Lee
|
Motif Enhanced Recommendation over Heterogeneous Information Network
|
CIKM 2019 camera ready version
| null | null | null |
cs.SI cs.IR
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
Heterogeneous Information Networks (HIN) has been widely used in recommender
systems (RSs). In previous HIN-based RSs, meta-path is used to compute the
similarity between users and items. However, existing meta-path based methods
only consider first-order relations, ignoring higher-order relations among the
nodes of \textit{same} type, captured by \textit{motifs}. In this paper, we
propose to use motifs to capture higher-order relations among nodes of same
type in a HIN and develop the motif-enhanced meta-path (MEMP) to combine
motif-based higher-order relations with edge-based first-order relations. With
MEMP-based similarities between users and items, we design a recommending model
MoHINRec, and experimental results on two real-world datasets, Epinions and
CiaoDVD, demonstrate its superiority over existing HIN-based RS methods.
|
[
{
"created": "Mon, 26 Aug 2019 14:21:14 GMT",
"version": "v1"
}
] |
2019-08-27
|
[
[
"Zhao",
"Huan",
""
],
[
"Zhou",
"Yingqi",
""
],
[
"Song",
"Yangqiu",
""
],
[
"Lee",
"Dik Lun",
""
]
] |
Heterogeneous Information Networks (HIN) has been widely used in recommender systems (RSs). In previous HIN-based RSs, meta-path is used to compute the similarity between users and items. However, existing meta-path based methods only consider first-order relations, ignoring higher-order relations among the nodes of \textit{same} type, captured by \textit{motifs}. In this paper, we propose to use motifs to capture higher-order relations among nodes of same type in a HIN and develop the motif-enhanced meta-path (MEMP) to combine motif-based higher-order relations with edge-based first-order relations. With MEMP-based similarities between users and items, we design a recommending model MoHINRec, and experimental results on two real-world datasets, Epinions and CiaoDVD, demonstrate its superiority over existing HIN-based RS methods.
|
1705.08339
|
Carlos Mosquera
|
Carlos Mosquera, Roberto Lopez-Valcarce, Vahid Joroughi
|
Distributed Precoding Systems in Multi-Gateway Multibeam Satellites:
Regularization and Coarse Beamforming
|
Submitted to IEEE Transactions on Wireless Communications
| null | null | null |
cs.IT math.IT
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
This paper deals with the problem of beamforming design in a multibeam
satellite, which is shared by different groups of terminals -clusters-, each
served by an Earth station or gateway. Each gateway precodes the symbols
addressed to its respective users; the design follows an MMSE criterion, and a
regularization factor judiciously chosen allows to account for the presence of
mutually interfering clusters, extending more classical results applicable to
one centralized station. More importantly, channel statistics can be used
instead of instantaneous channel state information, avoiding the exchange of
information among gateways through backhaul links. The on-board satellite
beamforming weights are designed to exploit the degrees of freedom of the
satellite antennas to minimize the noise impact and the interference to some
specific users. On-ground beamforming results are provided as a reference to
compare the joint performance of MMSE precoders and on-board beamforming
network. A non-adaptive design complements the results and makes them more
amenable to practical use by designing a coarse beamforming network.
|
[
{
"created": "Tue, 23 May 2017 14:57:48 GMT",
"version": "v1"
}
] |
2017-05-24
|
[
[
"Mosquera",
"Carlos",
""
],
[
"Lopez-Valcarce",
"Roberto",
""
],
[
"Joroughi",
"Vahid",
""
]
] |
This paper deals with the problem of beamforming design in a multibeam satellite, which is shared by different groups of terminals -clusters-, each served by an Earth station or gateway. Each gateway precodes the symbols addressed to its respective users; the design follows an MMSE criterion, and a regularization factor judiciously chosen allows to account for the presence of mutually interfering clusters, extending more classical results applicable to one centralized station. More importantly, channel statistics can be used instead of instantaneous channel state information, avoiding the exchange of information among gateways through backhaul links. The on-board satellite beamforming weights are designed to exploit the degrees of freedom of the satellite antennas to minimize the noise impact and the interference to some specific users. On-ground beamforming results are provided as a reference to compare the joint performance of MMSE precoders and on-board beamforming network. A non-adaptive design complements the results and makes them more amenable to practical use by designing a coarse beamforming network.
|
1303.2175
|
Keivan Navi
|
Samira Shirinabadi Farahani, Ronak Zarhoun, Mohammad Hossein Moaiyeri
and Keivan Navi
|
An efficient cntfet-based 7-input minority gate
| null | null | null | null |
cs.AR
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
Complementary metal oxide semiconductor technology (CMOS) has been faced
critical challenges in nano-scale regime. CNTFET (Carbon Nanotube Field effect
transistor) technology is a promising alternative for CMOS technology. In this
paper, we proposed a novel 7-input minority gate in CNTFET technology that has
only 9 CNTFETs. Minority function is utilized in the voting systems for
decision making and also it is used in data mining. This proposed 7-input
minority gate is utilized less fewer transistors than the conventional CMOS
method which utilizes many transistors for implementing sum of products. By
means of this proposed 7-input minority gate, a 4-input NAND gate can be
implemented, which gets better the conventional design in terms of delay and
energy efficiency and has much more deriving power at its output.
|
[
{
"created": "Sat, 9 Mar 2013 06:57:21 GMT",
"version": "v1"
}
] |
2013-03-12
|
[
[
"Farahani",
"Samira Shirinabadi",
""
],
[
"Zarhoun",
"Ronak",
""
],
[
"Moaiyeri",
"Mohammad Hossein",
""
],
[
"Navi",
"Keivan",
""
]
] |
Complementary metal oxide semiconductor technology (CMOS) has been faced critical challenges in nano-scale regime. CNTFET (Carbon Nanotube Field effect transistor) technology is a promising alternative for CMOS technology. In this paper, we proposed a novel 7-input minority gate in CNTFET technology that has only 9 CNTFETs. Minority function is utilized in the voting systems for decision making and also it is used in data mining. This proposed 7-input minority gate is utilized less fewer transistors than the conventional CMOS method which utilizes many transistors for implementing sum of products. By means of this proposed 7-input minority gate, a 4-input NAND gate can be implemented, which gets better the conventional design in terms of delay and energy efficiency and has much more deriving power at its output.
|
1909.11764
|
Chen Zhu
|
Chen Zhu, Yu Cheng, Zhe Gan, Siqi Sun, Tom Goldstein, Jingjing Liu
|
FreeLB: Enhanced Adversarial Training for Natural Language Understanding
|
Adding results with ALBERT
| null | null | null |
cs.CL cs.LG
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
Adversarial training, which minimizes the maximal risk for label-preserving
input perturbations, has proved to be effective for improving the
generalization of language models. In this work, we propose a novel adversarial
training algorithm, FreeLB, that promotes higher invariance in the embedding
space, by adding adversarial perturbations to word embeddings and minimizing
the resultant adversarial risk inside different regions around input samples.
To validate the effectiveness of the proposed approach, we apply it to
Transformer-based models for natural language understanding and commonsense
reasoning tasks. Experiments on the GLUE benchmark show that when applied only
to the finetuning stage, it is able to improve the overall test scores of
BERT-base model from 78.3 to 79.4, and RoBERTa-large model from 88.5 to 88.8.
In addition, the proposed approach achieves state-of-the-art single-model test
accuracies of 85.44\% and 67.75\% on ARC-Easy and ARC-Challenge. Experiments on
CommonsenseQA benchmark further demonstrate that FreeLB can be generalized and
boost the performance of RoBERTa-large model on other tasks as well. Code is
available at \url{https://github.com/zhuchen03/FreeLB .
|
[
{
"created": "Wed, 25 Sep 2019 20:50:32 GMT",
"version": "v1"
},
{
"created": "Mon, 30 Sep 2019 18:53:21 GMT",
"version": "v2"
},
{
"created": "Sat, 5 Oct 2019 04:05:46 GMT",
"version": "v3"
},
{
"created": "Wed, 19 Feb 2020 01:57:24 GMT",
"version": "v4"
},
{
"created": "Thu, 23 Apr 2020 07:19:00 GMT",
"version": "v5"
}
] |
2020-04-24
|
[
[
"Zhu",
"Chen",
""
],
[
"Cheng",
"Yu",
""
],
[
"Gan",
"Zhe",
""
],
[
"Sun",
"Siqi",
""
],
[
"Goldstein",
"Tom",
""
],
[
"Liu",
"Jingjing",
""
]
] |
Adversarial training, which minimizes the maximal risk for label-preserving input perturbations, has proved to be effective for improving the generalization of language models. In this work, we propose a novel adversarial training algorithm, FreeLB, that promotes higher invariance in the embedding space, by adding adversarial perturbations to word embeddings and minimizing the resultant adversarial risk inside different regions around input samples. To validate the effectiveness of the proposed approach, we apply it to Transformer-based models for natural language understanding and commonsense reasoning tasks. Experiments on the GLUE benchmark show that when applied only to the finetuning stage, it is able to improve the overall test scores of BERT-base model from 78.3 to 79.4, and RoBERTa-large model from 88.5 to 88.8. In addition, the proposed approach achieves state-of-the-art single-model test accuracies of 85.44\% and 67.75\% on ARC-Easy and ARC-Challenge. Experiments on CommonsenseQA benchmark further demonstrate that FreeLB can be generalized and boost the performance of RoBERTa-large model on other tasks as well. Code is available at \url{https://github.com/zhuchen03/FreeLB .
|
2404.02565
|
Sreela Kodali
|
Sreela Kodali, Cihualpilli Camino Cruz, Thomas C. Bulea, Kevin S. Rao
Diana Bharucha-Goebel, Alexander T. Chesler, Carsten G. Bonnemann, Allison M.
Okamura
|
Spatial Summation of Localized Pressure for Haptic Sensory Prostheses
|
2 pages, 2 figures, 2024 IEEE Haptics Symposium Work-in-Progress
Paper
| null | null | null |
cs.HC
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
A host of medical conditions, including amputations, diabetes, stroke, and
genetic disease, result in loss of touch sensation. Because most types of
sensory loss have no pharmacological treatment or rehabilitative therapy, we
propose a haptic sensory prosthesis that provides substitutive feedback. The
wrist and forearm are compelling locations for feedback due to available skin
area and not occluding the hands, but have reduced mechanoreceptor density
compared to the fingertips. Focusing on localized pressure as the feedback
modality, we hypothesize that we can improve on prior devices by invoking a
wider range of stimulus intensity using multiple points of pressure to evoke
spatial summation, which is the cumulative perceptual experience from multiple
points of stimuli. We conducted a preliminary perceptual test to investigate
this idea and found that just noticeable difference is reduced with two points
of pressure compared to one, motivating future work using spatial summation in
sensory prostheses.
|
[
{
"created": "Wed, 3 Apr 2024 08:37:05 GMT",
"version": "v1"
}
] |
2024-04-04
|
[
[
"Kodali",
"Sreela",
""
],
[
"Cruz",
"Cihualpilli Camino",
""
],
[
"Bulea",
"Thomas C.",
""
],
[
"Bharucha-Goebel",
"Kevin S. Rao Diana",
""
],
[
"Chesler",
"Alexander T.",
""
],
[
"Bonnemann",
"Carsten G.",
""
],
[
"Okamura",
"Allison M.",
""
]
] |
A host of medical conditions, including amputations, diabetes, stroke, and genetic disease, result in loss of touch sensation. Because most types of sensory loss have no pharmacological treatment or rehabilitative therapy, we propose a haptic sensory prosthesis that provides substitutive feedback. The wrist and forearm are compelling locations for feedback due to available skin area and not occluding the hands, but have reduced mechanoreceptor density compared to the fingertips. Focusing on localized pressure as the feedback modality, we hypothesize that we can improve on prior devices by invoking a wider range of stimulus intensity using multiple points of pressure to evoke spatial summation, which is the cumulative perceptual experience from multiple points of stimuli. We conducted a preliminary perceptual test to investigate this idea and found that just noticeable difference is reduced with two points of pressure compared to one, motivating future work using spatial summation in sensory prostheses.
|
1909.07166
|
Smriti Prathapan
|
Kaushik Velusamy, Smriti Prathapan, Milton Halem
|
Exploring the Behavior of Coherent Accelerator Processor Interface
(CAPI) on IBM Power8+ Architecture and FlashSystem 900
|
18 pages, 7 figures, 3 tables, Accepted for publication at 2019
International Workshop on OpenPOWER for HPC (IWOPH19) International
Supercomputing Conference HPC Frankfurt, Germany
| null | null | null |
cs.DC
|
http://creativecommons.org/licenses/by/4.0/
|
The Coherent Accelerator Processor Interface (CAPI) is a general term for the
infrastructure that provides high throughput and low latency path to the flash
storage connected to the IBM POWER 8+ System. CAPI accelerator card is attached
coherently as a peer to the Power8+ processor. This removes the overhead and
complexity of the IO subsystem and allows the accelerator to operate as part of
an application. In this paper, we present the results of experiments on IBM
FlashSystem900 (FS900) with CAPI accelerator card using the "CAPI-Flash IBM
Data Engine for NoSQL Software" Library. This library provides the application,
a direct access to the underlying flash storage through user space APIs, to
manage and access the data in flash. This offloads kernel IO driver
functionality to dedicated CAPI FPGA accelerator hardware. We conducted
experiments to analyze the performance of FS900 with CAPI accelerator card,
using the Key Value Layer APIs, employing NASA's MODIS Land Surface Reflectance
dataset as a large dataset use case. We performed Read and Write operations on
datasets of size ranging from 1MB to 3TB by varying the number of threads. We
then compared this performance with other heterogeneous storage and memory
devices such as NVM, SSD and RAM, without using the CAPI Accelerator in
synchronous and asynchronous file IO modes of operations. The results indicate
that FS900 & CAPI, together with the metadata cache in RAM, delivers the
highest IO/s and OP/s for read operations. This was higher than just using RAM,
along with utilizing lesser CPU resources. Among FS900, SSD and NVM, FS900 had
the highest write IO/s. Another important observation is that, when the size of
the input dataset exceeds the capacity of RAM, and when the data access is
non-uniform and sparse, FS900 with CAPI would be a cost-effective alternative.
|
[
{
"created": "Thu, 12 Sep 2019 15:45:37 GMT",
"version": "v1"
}
] |
2019-09-17
|
[
[
"Velusamy",
"Kaushik",
""
],
[
"Prathapan",
"Smriti",
""
],
[
"Halem",
"Milton",
""
]
] |
The Coherent Accelerator Processor Interface (CAPI) is a general term for the infrastructure that provides high throughput and low latency path to the flash storage connected to the IBM POWER 8+ System. CAPI accelerator card is attached coherently as a peer to the Power8+ processor. This removes the overhead and complexity of the IO subsystem and allows the accelerator to operate as part of an application. In this paper, we present the results of experiments on IBM FlashSystem900 (FS900) with CAPI accelerator card using the "CAPI-Flash IBM Data Engine for NoSQL Software" Library. This library provides the application, a direct access to the underlying flash storage through user space APIs, to manage and access the data in flash. This offloads kernel IO driver functionality to dedicated CAPI FPGA accelerator hardware. We conducted experiments to analyze the performance of FS900 with CAPI accelerator card, using the Key Value Layer APIs, employing NASA's MODIS Land Surface Reflectance dataset as a large dataset use case. We performed Read and Write operations on datasets of size ranging from 1MB to 3TB by varying the number of threads. We then compared this performance with other heterogeneous storage and memory devices such as NVM, SSD and RAM, without using the CAPI Accelerator in synchronous and asynchronous file IO modes of operations. The results indicate that FS900 & CAPI, together with the metadata cache in RAM, delivers the highest IO/s and OP/s for read operations. This was higher than just using RAM, along with utilizing lesser CPU resources. Among FS900, SSD and NVM, FS900 had the highest write IO/s. Another important observation is that, when the size of the input dataset exceeds the capacity of RAM, and when the data access is non-uniform and sparse, FS900 with CAPI would be a cost-effective alternative.
|
1801.04473
|
Regis Perrier
|
Iulia Tunaru and Beno\^it Denis and R\'egis Perrier and Bernard Uguen
|
Channel Whispering: a Protocol for Physical Layer Group Key Generation.
Application to IR-UWB through Deconvolution
|
21 pages
| null | null | null |
cs.CR cs.IT math.IT
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
As wireless ad hoc and mobile networks are emerging and the transferred data
become more sensitive, information security measures should make use of all the
available contextual resources to secure information flows. The physical layer
security framework provides models, algorithms, and proofs of concept for
generating pairwise symmetric keys over single links between two nodes within
communication range. In this study, we focus on cooperative group key
generation over multiple Impulse Radio - Ultra Wideband (IR-UWB) channels
according to the source model. The main idea, proposed in previous work,
consists in generating receiver-specific signals, also called s-signals, so
that only the intended receiver has access to the non-observable channels
corresponding to its non-adjacent links. Herein, we complete the analysis of
the proposed protocol and investigate several signal processing algorithms to
generate the s-signal expressed as a solution to a deconvolution problem in the
case of IR-UWB. Our findings indicate that it is compulsory to add a
parameterizable constraint to the searched s-signal and that the
Expectation-Maximization algorithm can provide a stable self-parameterizable
solution. Compared to physical layer key distribution methods, the proposed key
generation protocol requires less traffic overhead for small cooperative groups
while being robust at medium and high signal-to-noise ratios.
|
[
{
"created": "Sat, 13 Jan 2018 18:26:45 GMT",
"version": "v1"
}
] |
2018-01-16
|
[
[
"Tunaru",
"Iulia",
""
],
[
"Denis",
"Benoît",
""
],
[
"Perrier",
"Régis",
""
],
[
"Uguen",
"Bernard",
""
]
] |
As wireless ad hoc and mobile networks are emerging and the transferred data become more sensitive, information security measures should make use of all the available contextual resources to secure information flows. The physical layer security framework provides models, algorithms, and proofs of concept for generating pairwise symmetric keys over single links between two nodes within communication range. In this study, we focus on cooperative group key generation over multiple Impulse Radio - Ultra Wideband (IR-UWB) channels according to the source model. The main idea, proposed in previous work, consists in generating receiver-specific signals, also called s-signals, so that only the intended receiver has access to the non-observable channels corresponding to its non-adjacent links. Herein, we complete the analysis of the proposed protocol and investigate several signal processing algorithms to generate the s-signal expressed as a solution to a deconvolution problem in the case of IR-UWB. Our findings indicate that it is compulsory to add a parameterizable constraint to the searched s-signal and that the Expectation-Maximization algorithm can provide a stable self-parameterizable solution. Compared to physical layer key distribution methods, the proposed key generation protocol requires less traffic overhead for small cooperative groups while being robust at medium and high signal-to-noise ratios.
|
2207.04647
|
Zvi Schreiber
|
Zvi Schreiber
|
Epi-constructivism: Decidable sets of computable numbers as foundational
objects for mathematics
| null | null | null | null |
cs.LO math.LO
|
http://creativecommons.org/licenses/by-nc-nd/4.0/
|
It is well known that the R, the set of real numbers, is an abstract set,
where almost all its elements cannot be described in any finite language.
We investigate possible approaches to what might be called an
epi-constructionist approach to mathematics. While most constructive
mathematics is concerned with constructive proofs, the agenda here is that the
objects that we study, specifically the class of numbers that we study, should
be an enumerable set of finite symbol strings. These might also be called
decidable constructive real numbers, that is our class of numbers should be a
computable sets of explicitly represented computable numbers.
There have been various investigations of the computable numbers going back
to Turing. Most are however not expressed constructively, rather computable is
a property assigned to some of the abstract real numbers. Other definitions
define constructive real numbers without reference to the abstract R, but the
construction is undecidable, i.e., we cannot determine if a given construction
represents a computable real number or not. For example, we may define a real
as a computable convergent sequence of rationals, but cannot in general decide
if a given computable sequence is convergent.
This paper explores several specific classes of decidable constructive real
numbers that could form foundational objects for what we might call an
epi-constructionist mathematics.
|
[
{
"created": "Mon, 11 Jul 2022 06:18:24 GMT",
"version": "v1"
}
] |
2022-07-12
|
[
[
"Schreiber",
"Zvi",
""
]
] |
It is well known that the R, the set of real numbers, is an abstract set, where almost all its elements cannot be described in any finite language. We investigate possible approaches to what might be called an epi-constructionist approach to mathematics. While most constructive mathematics is concerned with constructive proofs, the agenda here is that the objects that we study, specifically the class of numbers that we study, should be an enumerable set of finite symbol strings. These might also be called decidable constructive real numbers, that is our class of numbers should be a computable sets of explicitly represented computable numbers. There have been various investigations of the computable numbers going back to Turing. Most are however not expressed constructively, rather computable is a property assigned to some of the abstract real numbers. Other definitions define constructive real numbers without reference to the abstract R, but the construction is undecidable, i.e., we cannot determine if a given construction represents a computable real number or not. For example, we may define a real as a computable convergent sequence of rationals, but cannot in general decide if a given computable sequence is convergent. This paper explores several specific classes of decidable constructive real numbers that could form foundational objects for what we might call an epi-constructionist mathematics.
|
2105.00173
|
Daniel Szelogowski
|
Daniel Szelogowski
|
Emotion Recognition of the Singing Voice: Toward a Real-Time Analysis
Tool for Singers
|
26 pages, 10 figures, 6 tables
| null | null | null |
cs.SD cs.AI cs.CY cs.LG cs.NE eess.AS
|
http://creativecommons.org/licenses/by/4.0/
|
Current computational-emotion research has focused on applying acoustic
properties to analyze how emotions are perceived mathematically or used in
natural language processing machine learning models. While recent interest has
focused on analyzing emotions from the spoken voice, little experimentation has
been performed to discover how emotions are recognized in the singing voice --
both in noiseless and noisy data (i.e., data that is either inaccurate,
difficult to interpret, has corrupted/distorted/nonsense information like
actual noise sounds in this case, or has a low ratio of usable/unusable
information). Not only does this ignore the challenges of training machine
learning models on more subjective data and testing them with much noisier
data, but there is also a clear disconnect in progress between advancing the
development of convolutional neural networks and the goal of emotionally
cognizant artificial intelligence. By training a new model to include this type
of information with a rich comprehension of psycho-acoustic properties, not
only can models be trained to recognize information within extremely noisy
data, but advancement can be made toward more complex biofeedback applications
-- including creating a model which could recognize emotions given any human
information (language, breath, voice, body, posture) and be used in any
performance medium (music, speech, acting) or psychological assistance for
patients with disorders such as BPD, alexithymia, autism, among others. This
paper seeks to reflect and expand upon the findings of related research and
present a stepping-stone toward this end goal.
|
[
{
"created": "Sat, 1 May 2021 05:47:15 GMT",
"version": "v1"
},
{
"created": "Sun, 4 Jul 2021 07:34:14 GMT",
"version": "v2"
}
] |
2021-07-06
|
[
[
"Szelogowski",
"Daniel",
""
]
] |
Current computational-emotion research has focused on applying acoustic properties to analyze how emotions are perceived mathematically or used in natural language processing machine learning models. While recent interest has focused on analyzing emotions from the spoken voice, little experimentation has been performed to discover how emotions are recognized in the singing voice -- both in noiseless and noisy data (i.e., data that is either inaccurate, difficult to interpret, has corrupted/distorted/nonsense information like actual noise sounds in this case, or has a low ratio of usable/unusable information). Not only does this ignore the challenges of training machine learning models on more subjective data and testing them with much noisier data, but there is also a clear disconnect in progress between advancing the development of convolutional neural networks and the goal of emotionally cognizant artificial intelligence. By training a new model to include this type of information with a rich comprehension of psycho-acoustic properties, not only can models be trained to recognize information within extremely noisy data, but advancement can be made toward more complex biofeedback applications -- including creating a model which could recognize emotions given any human information (language, breath, voice, body, posture) and be used in any performance medium (music, speech, acting) or psychological assistance for patients with disorders such as BPD, alexithymia, autism, among others. This paper seeks to reflect and expand upon the findings of related research and present a stepping-stone toward this end goal.
|
2407.13648
|
Skyler Grandel
|
Skyler Grandel (1), Scott Thomas Andersen (2), Yu Huang (1), Kevin
Leach (1) ((1) Vanderbilt University, (2) Universidad Nacional Aut\`onoma de
M\`exico)
|
COMCAT: Leveraging Human Judgment to Improve Automatic Documentation and
Summarization
|
12 pages, 6 figures
| null | null | null |
cs.SE
|
http://creativecommons.org/licenses/by/4.0/
|
Software maintenance constitutes a substantial portion of the total lifetime
costs of software, with a significant portion attributed to code comprehension.
Software comprehension is eased by documentation such as comments that
summarize and explain code. We present COMCAT, an approach to automate comment
generation by augmenting Large Language Models (LLMs) with expertise-guided
context to target the annotation of source code with comments that improve
comprehension. Our approach enables the selection of the most relevant and
informative comments for a given snippet or file containing source code. We
develop the COMCAT pipeline to comment C/C++ files by (1) automatically
identifying suitable locations in which to place comments, (2) predicting the
most helpful type of comment for each location, and (3) generating a comment
based on the selected location and comment type. In a human subject evaluation,
we demonstrate that COMCAT-generated comments significantly improve developer
code comprehension across three indicative software engineering tasks by up to
12% for 87% of participants. In addition, we demonstrate that COMCAT-generated
comments are at least as accurate and readable as human-generated comments and
are preferred over standard ChatGPT-generated comments for up to 92% of
snippets of code. Furthermore, we develop and release a dataset containing
source code snippets, human-written comments, and human-annotated comment
categories. COMCAT leverages LLMs to offer a significant improvement in code
comprehension across a variety of human software engineering tasks.
|
[
{
"created": "Thu, 18 Jul 2024 16:26:31 GMT",
"version": "v1"
}
] |
2024-07-19
|
[
[
"Grandel",
"Skyler",
""
],
[
"Andersen",
"Scott Thomas",
""
],
[
"Huang",
"Yu",
""
],
[
"Leach",
"Kevin",
""
]
] |
Software maintenance constitutes a substantial portion of the total lifetime costs of software, with a significant portion attributed to code comprehension. Software comprehension is eased by documentation such as comments that summarize and explain code. We present COMCAT, an approach to automate comment generation by augmenting Large Language Models (LLMs) with expertise-guided context to target the annotation of source code with comments that improve comprehension. Our approach enables the selection of the most relevant and informative comments for a given snippet or file containing source code. We develop the COMCAT pipeline to comment C/C++ files by (1) automatically identifying suitable locations in which to place comments, (2) predicting the most helpful type of comment for each location, and (3) generating a comment based on the selected location and comment type. In a human subject evaluation, we demonstrate that COMCAT-generated comments significantly improve developer code comprehension across three indicative software engineering tasks by up to 12% for 87% of participants. In addition, we demonstrate that COMCAT-generated comments are at least as accurate and readable as human-generated comments and are preferred over standard ChatGPT-generated comments for up to 92% of snippets of code. Furthermore, we develop and release a dataset containing source code snippets, human-written comments, and human-annotated comment categories. COMCAT leverages LLMs to offer a significant improvement in code comprehension across a variety of human software engineering tasks.
|
2007.02984
|
Nitzan Zamir
|
Nitzan Zamir and Yoram Moses
|
Probably Approximately Knowing
|
23 pages, 2 figures, a full version of a paper whose extended
abstract appears in the proceeding of PODC 2020
| null | null | null |
cs.DC
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
Whereas deterministic protocols are typically guaranteed to obtain particular
goals of interest, probabilistic protocols typically provide only probabilistic
guarantees. This paper initiates an investigation of the interdependence
between actions and subjective beliefs of agents in a probabilistic setting. In
particular, we study what probabilistic beliefs an agent should have when
performing actions, in a protocol that satisfies a probabilistic constraint of
the form: 'Condition C should hold with probability at least p when action a is
performed'. Our main result is that the expected degree of an agent's belief in
C when it performs a equals the probability that C holds when a is performed.
Indeed, if the threshold of the probabilistic constraint should hold with
probaility p=1-x^2 for some small value of x then, with probability 1-x, when
the agent acts it will assign a probabilistic belief no smaller than 1-x to the
possibility that C holds. In other words, viewing strong belief as,
intuitively, approximate knowledge, the agent must probably approximately know
(PAK-know) that C is true when it acts.
|
[
{
"created": "Mon, 6 Jul 2020 18:12:41 GMT",
"version": "v1"
}
] |
2020-07-08
|
[
[
"Zamir",
"Nitzan",
""
],
[
"Moses",
"Yoram",
""
]
] |
Whereas deterministic protocols are typically guaranteed to obtain particular goals of interest, probabilistic protocols typically provide only probabilistic guarantees. This paper initiates an investigation of the interdependence between actions and subjective beliefs of agents in a probabilistic setting. In particular, we study what probabilistic beliefs an agent should have when performing actions, in a protocol that satisfies a probabilistic constraint of the form: 'Condition C should hold with probability at least p when action a is performed'. Our main result is that the expected degree of an agent's belief in C when it performs a equals the probability that C holds when a is performed. Indeed, if the threshold of the probabilistic constraint should hold with probaility p=1-x^2 for some small value of x then, with probability 1-x, when the agent acts it will assign a probabilistic belief no smaller than 1-x to the possibility that C holds. In other words, viewing strong belief as, intuitively, approximate knowledge, the agent must probably approximately know (PAK-know) that C is true when it acts.
|
1510.04817
|
Javier Alvez
|
Javier \'Alvez and Paqui Lucio and German Rigau
|
Improving the Competency of First-Order Ontologies
|
8 pages, 2 tables
|
Proceedings of the 8th International Conference on Knowledge
Capture (K-CAP 2015). Palisades, NY. 2015
|
10.1145/2815833.2815841
| null |
cs.AI cs.LO
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
We introduce a new framework to evaluate and improve first-order (FO)
ontologies using automated theorem provers (ATPs) on the basis of competency
questions (CQs). Our framework includes both the adaptation of a methodology
for evaluating ontologies to the framework of first-order logic and a new set
of non-trivial CQs designed to evaluate FO versions of SUMO, which
significantly extends the very small set of CQs proposed in the literature.
Most of these new CQs have been automatically generated from a small set of
patterns and the mapping of WordNet to SUMO. Applying our framework, we
demonstrate that Adimen-SUMO v2.2 outperforms TPTP-SUMO. In addition, using the
feedback provided by ATPs we have set an improved version of Adimen-SUMO
(v2.4). This new version outperforms the previous ones in terms of competency.
For instance, "Humans can reason" is automatically inferred from Adimen-SUMO
v2.4, while it is neither deducible from TPTP-SUMO nor Adimen-SUMO v2.2.
|
[
{
"created": "Fri, 16 Oct 2015 09:01:35 GMT",
"version": "v1"
}
] |
2015-10-19
|
[
[
"Álvez",
"Javier",
""
],
[
"Lucio",
"Paqui",
""
],
[
"Rigau",
"German",
""
]
] |
We introduce a new framework to evaluate and improve first-order (FO) ontologies using automated theorem provers (ATPs) on the basis of competency questions (CQs). Our framework includes both the adaptation of a methodology for evaluating ontologies to the framework of first-order logic and a new set of non-trivial CQs designed to evaluate FO versions of SUMO, which significantly extends the very small set of CQs proposed in the literature. Most of these new CQs have been automatically generated from a small set of patterns and the mapping of WordNet to SUMO. Applying our framework, we demonstrate that Adimen-SUMO v2.2 outperforms TPTP-SUMO. In addition, using the feedback provided by ATPs we have set an improved version of Adimen-SUMO (v2.4). This new version outperforms the previous ones in terms of competency. For instance, "Humans can reason" is automatically inferred from Adimen-SUMO v2.4, while it is neither deducible from TPTP-SUMO nor Adimen-SUMO v2.2.
|
2304.01950
|
Yu Qiao
|
Yu Qiao, Md. Shirajum Munir, Apurba Adhikary, Huy Q. Le, Avi Deb Raha,
Chaoning Zhang, Choong Seon Hong
|
MP-FedCL: Multiprototype Federated Contrastive Learning for Edge
Intelligence
|
Accepted by IEEE Internet of Things
| null | null | null |
cs.LG cs.AI cs.CV cs.DC
|
http://creativecommons.org/licenses/by-nc-nd/4.0/
|
Federated learning-assisted edge intelligence enables privacy protection in
modern intelligent services. However, not independent and identically
distributed (non-IID) distribution among edge clients can impair the local
model performance. The existing single prototype-based strategy represents a
class by using the mean of the feature space. However, feature spaces are
usually not clustered, and a single prototype may not represent a class well.
Motivated by this, this paper proposes a multi-prototype federated contrastive
learning approach (MP-FedCL) which demonstrates the effectiveness of using a
multi-prototype strategy over a single-prototype under non-IID settings,
including both label and feature skewness. Specifically, a multi-prototype
computation strategy based on \textit{k-means} is first proposed to capture
different embedding representations for each class space, using multiple
prototypes ($k$ centroids) to represent a class in the embedding space. In each
global round, the computed multiple prototypes and their respective model
parameters are sent to the edge server for aggregation into a global prototype
pool, which is then sent back to all clients to guide their local training.
Finally, local training for each client minimizes their own supervised learning
tasks and learns from shared prototypes in the global prototype pool through
supervised contrastive learning, which encourages them to learn knowledge
related to their own class from others and reduces the absorption of unrelated
knowledge in each global iteration. Experimental results on MNIST, Digit-5,
Office-10, and DomainNet show that our method outperforms multiple baselines,
with an average test accuracy improvement of about 4.6\% and 10.4\% under
feature and label non-IID distributions, respectively.
|
[
{
"created": "Sat, 1 Apr 2023 09:16:40 GMT",
"version": "v1"
},
{
"created": "Wed, 11 Oct 2023 14:21:29 GMT",
"version": "v2"
}
] |
2023-10-12
|
[
[
"Qiao",
"Yu",
""
],
[
"Munir",
"Md. Shirajum",
""
],
[
"Adhikary",
"Apurba",
""
],
[
"Le",
"Huy Q.",
""
],
[
"Raha",
"Avi Deb",
""
],
[
"Zhang",
"Chaoning",
""
],
[
"Hong",
"Choong Seon",
""
]
] |
Federated learning-assisted edge intelligence enables privacy protection in modern intelligent services. However, not independent and identically distributed (non-IID) distribution among edge clients can impair the local model performance. The existing single prototype-based strategy represents a class by using the mean of the feature space. However, feature spaces are usually not clustered, and a single prototype may not represent a class well. Motivated by this, this paper proposes a multi-prototype federated contrastive learning approach (MP-FedCL) which demonstrates the effectiveness of using a multi-prototype strategy over a single-prototype under non-IID settings, including both label and feature skewness. Specifically, a multi-prototype computation strategy based on \textit{k-means} is first proposed to capture different embedding representations for each class space, using multiple prototypes ($k$ centroids) to represent a class in the embedding space. In each global round, the computed multiple prototypes and their respective model parameters are sent to the edge server for aggregation into a global prototype pool, which is then sent back to all clients to guide their local training. Finally, local training for each client minimizes their own supervised learning tasks and learns from shared prototypes in the global prototype pool through supervised contrastive learning, which encourages them to learn knowledge related to their own class from others and reduces the absorption of unrelated knowledge in each global iteration. Experimental results on MNIST, Digit-5, Office-10, and DomainNet show that our method outperforms multiple baselines, with an average test accuracy improvement of about 4.6\% and 10.4\% under feature and label non-IID distributions, respectively.
|
1811.08521
|
Scott Wisdom
|
Scott Wisdom, John R. Hershey, Kevin Wilson, Jeremy Thorpe, Michael
Chinen, Brian Patton, Rif A. Saurous
|
Differentiable Consistency Constraints for Improved Deep Speech
Enhancement
| null | null | null | null |
cs.SD eess.AS
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
In recent years, deep networks have led to dramatic improvements in speech
enhancement by framing it as a data-driven pattern recognition problem. In many
modern enhancement systems, large amounts of data are used to train a deep
network to estimate masks for complex-valued short-time Fourier transforms
(STFTs) to suppress noise and preserve speech. However, current masking
approaches often neglect two important constraints: STFT consistency and
mixture consistency. Without STFT consistency, the system's output is not
necessarily the STFT of a time-domain signal, and without mixture consistency,
the sum of the estimated sources does not necessarily equal the input mixture.
Furthermore, the only previous approaches that apply mixture consistency use
real-valued masks; mixture consistency has been ignored for complex-valued
masks. In this paper, we show that STFT consistency and mixture consistency can
be jointly imposed by adding simple differentiable projection layers to the
enhancement network. These layers are compatible with real or complex-valued
masks. Using both of these constraints with complex-valued masks provides a 0.7
dB increase in scale-invariant signal-to-distortion ratio (SI-SDR) on a large
dataset of speech corrupted by a wide variety of nonstationary noise across a
range of input SNRs.
|
[
{
"created": "Tue, 20 Nov 2018 22:44:12 GMT",
"version": "v1"
}
] |
2018-11-22
|
[
[
"Wisdom",
"Scott",
""
],
[
"Hershey",
"John R.",
""
],
[
"Wilson",
"Kevin",
""
],
[
"Thorpe",
"Jeremy",
""
],
[
"Chinen",
"Michael",
""
],
[
"Patton",
"Brian",
""
],
[
"Saurous",
"Rif A.",
""
]
] |
In recent years, deep networks have led to dramatic improvements in speech enhancement by framing it as a data-driven pattern recognition problem. In many modern enhancement systems, large amounts of data are used to train a deep network to estimate masks for complex-valued short-time Fourier transforms (STFTs) to suppress noise and preserve speech. However, current masking approaches often neglect two important constraints: STFT consistency and mixture consistency. Without STFT consistency, the system's output is not necessarily the STFT of a time-domain signal, and without mixture consistency, the sum of the estimated sources does not necessarily equal the input mixture. Furthermore, the only previous approaches that apply mixture consistency use real-valued masks; mixture consistency has been ignored for complex-valued masks. In this paper, we show that STFT consistency and mixture consistency can be jointly imposed by adding simple differentiable projection layers to the enhancement network. These layers are compatible with real or complex-valued masks. Using both of these constraints with complex-valued masks provides a 0.7 dB increase in scale-invariant signal-to-distortion ratio (SI-SDR) on a large dataset of speech corrupted by a wide variety of nonstationary noise across a range of input SNRs.
|
2404.08181
|
Sina Hajimiri
|
Sina Hajimiri, Ismail Ben Ayed, Jose Dolz
|
Pay Attention to Your Neighbours: Training-Free Open-Vocabulary Semantic
Segmentation
| null | null | null | null |
cs.CV
|
http://creativecommons.org/licenses/by/4.0/
|
Despite the significant progress in deep learning for dense visual
recognition problems, such as semantic segmentation, traditional methods are
constrained by fixed class sets. Meanwhile, vision-language foundation models,
such as CLIP, have showcased remarkable effectiveness in numerous zero-shot
image-level tasks, owing to their robust generalizability. Recently, a body of
work has investigated utilizing these models in open-vocabulary semantic
segmentation (OVSS). However, existing approaches often rely on impractical
supervised pre-training or access to additional pre-trained networks. In this
work, we propose a strong baseline for training-free OVSS, termed
Neighbour-Aware CLIP (NACLIP), representing a straightforward adaptation of
CLIP tailored for this scenario. Our method enforces localization of patches in
the self-attention of CLIP's vision transformer which, despite being crucial
for dense prediction tasks, has been overlooked in the OVSS literature. By
incorporating design choices favouring segmentation, our approach significantly
improves performance without requiring additional data, auxiliary pre-trained
networks, or extensive hyperparameter tuning, making it highly practical for
real-world applications. Experiments are performed on 8 popular semantic
segmentation benchmarks, yielding state-of-the-art performance on most
scenarios. Our code is publicly available at https://github.com/sinahmr/NACLIP .
|
[
{
"created": "Fri, 12 Apr 2024 01:08:04 GMT",
"version": "v1"
}
] |
2024-04-15
|
[
[
"Hajimiri",
"Sina",
""
],
[
"Ayed",
"Ismail Ben",
""
],
[
"Dolz",
"Jose",
""
]
] |
Despite the significant progress in deep learning for dense visual recognition problems, such as semantic segmentation, traditional methods are constrained by fixed class sets. Meanwhile, vision-language foundation models, such as CLIP, have showcased remarkable effectiveness in numerous zero-shot image-level tasks, owing to their robust generalizability. Recently, a body of work has investigated utilizing these models in open-vocabulary semantic segmentation (OVSS). However, existing approaches often rely on impractical supervised pre-training or access to additional pre-trained networks. In this work, we propose a strong baseline for training-free OVSS, termed Neighbour-Aware CLIP (NACLIP), representing a straightforward adaptation of CLIP tailored for this scenario. Our method enforces localization of patches in the self-attention of CLIP's vision transformer which, despite being crucial for dense prediction tasks, has been overlooked in the OVSS literature. By incorporating design choices favouring segmentation, our approach significantly improves performance without requiring additional data, auxiliary pre-trained networks, or extensive hyperparameter tuning, making it highly practical for real-world applications. Experiments are performed on 8 popular semantic segmentation benchmarks, yielding state-of-the-art performance on most scenarios. Our code is publicly available at https://github.com/sinahmr/NACLIP .
|
1708.07549
|
Moi Hoon Yap
|
Adrian K. Davison, Walied Merghani and Moi Hoon Yap
|
Objective Classes for Micro-Facial Expression Recognition
|
11 pages, 4 figures and 5 tables. This paper will be submitted for
journal review
| null | null | null |
cs.CV
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
Micro-expressions are brief spontaneous facial expressions that appear on a
face when a person conceals an emotion, making them different to normal facial
expressions in subtlety and duration. Currently, emotion classes within the
CASME II dataset are based on Action Units and self-reports, creating conflicts
during machine learning training. We will show that classifying expressions
using Action Units, instead of predicted emotion, removes the potential bias of
human reporting. The proposed classes are tested using LBP-TOP, HOOF and HOG 3D
feature descriptors. The experiments are evaluated on two benchmark FACS coded
datasets: CASME II and SAMM. The best result achieves 86.35\% accuracy when
classifying the proposed 5 classes on CASME II using HOG 3D, outperforming the
result of the state-of-the-art 5-class emotional-based classification in CASME
II. Results indicate that classification based on Action Units provides an
objective method to improve micro-expression recognition.
|
[
{
"created": "Thu, 24 Aug 2017 20:37:10 GMT",
"version": "v1"
},
{
"created": "Sun, 3 Dec 2017 06:12:57 GMT",
"version": "v2"
}
] |
2017-12-05
|
[
[
"Davison",
"Adrian K.",
""
],
[
"Merghani",
"Walied",
""
],
[
"Yap",
"Moi Hoon",
""
]
] |
Micro-expressions are brief spontaneous facial expressions that appear on a face when a person conceals an emotion, making them different to normal facial expressions in subtlety and duration. Currently, emotion classes within the CASME II dataset are based on Action Units and self-reports, creating conflicts during machine learning training. We will show that classifying expressions using Action Units, instead of predicted emotion, removes the potential bias of human reporting. The proposed classes are tested using LBP-TOP, HOOF and HOG 3D feature descriptors. The experiments are evaluated on two benchmark FACS coded datasets: CASME II and SAMM. The best result achieves 86.35\% accuracy when classifying the proposed 5 classes on CASME II using HOG 3D, outperforming the result of the state-of-the-art 5-class emotional-based classification in CASME II. Results indicate that classification based on Action Units provides an objective method to improve micro-expression recognition.
|
2001.04552
|
Luca Puglia
|
Luca Puglia and Cormac Brick
|
Deep Learning Stereo Vision at the edge
| null | null | null | null |
cs.CV cs.LG
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
We present an overview of the methodology used to build a new stereo vision
solution that is suitable for System on Chip. This new solution was developed
to bring computer vision capability to embedded devices that live in a power
constrained environment. The solution is constructured as a hybrid between
classical Stereo Vision techniques and deep learning approaches. The
stereoscopic module is composed of two separate modules: one that accelerates
the neural network we trained and one that accelerates the front-end part. The
system is completely passive and does not require any structured light to
obtain very compelling accuracy. With respect to the previous Stereo Vision
solutions offered by the industries we offer a major improvement is robustness
to noise. This is mainly possible due to the deep learning part of the chosen
architecture. We submitted our result to Middlebury dataset challenge. It
currently ranks as the best System on Chip solution. The system has been
developed for low latency applications which require better than real time
performance on high definition videos.
|
[
{
"created": "Mon, 13 Jan 2020 22:30:41 GMT",
"version": "v1"
}
] |
2020-01-15
|
[
[
"Puglia",
"Luca",
""
],
[
"Brick",
"Cormac",
""
]
] |
We present an overview of the methodology used to build a new stereo vision solution that is suitable for System on Chip. This new solution was developed to bring computer vision capability to embedded devices that live in a power constrained environment. The solution is constructured as a hybrid between classical Stereo Vision techniques and deep learning approaches. The stereoscopic module is composed of two separate modules: one that accelerates the neural network we trained and one that accelerates the front-end part. The system is completely passive and does not require any structured light to obtain very compelling accuracy. With respect to the previous Stereo Vision solutions offered by the industries we offer a major improvement is robustness to noise. This is mainly possible due to the deep learning part of the chosen architecture. We submitted our result to Middlebury dataset challenge. It currently ranks as the best System on Chip solution. The system has been developed for low latency applications which require better than real time performance on high definition videos.
|
1411.0710
|
Eric Bax
|
Valeria Stourm and Eric Bax
|
Incorporating Hidden Costs of Annoying Ads in Display Auctions
| null | null | null | null |
cs.GT
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
Media publisher platforms often face an effectiveness-nuisance tradeoff: more
annoying ads can be more effective for some advertisers because of their
ability to attract attention, but after attracting viewers' attention, their
nuisance to viewers can decrease engagement with the platform over time. With
the rise of mobile technology and ad blockers, many platforms are becoming
increasingly concerned about how to improve monetization through digital ads
while improving viewer experience.
We study an online ad auction mechanism that incorporates a charge for ad
impact on user experience as a criterion for ad selection and pricing. Like a
Pigovian tax, the charge causes advertisers to internalize the hidden cost of
foregone future platform revenue due to ad impact on user experience. Over
time, the mechanism provides an incentive for advertisers to develop ads that
are effective while offering viewers a more pleasant experience. We show that
adopting the mechanism can simultaneously benefit the publisher, advertisers,
and viewers, even in the short term.
Incorporating a charge for ad impact can increase expected advertiser profits
if enough advertisers compete. A stronger effectiveness-nuisance tradeoff,
meaning that ad effectiveness is more strongly associated with negative impact
on user experience, increases the amount of competition required for the
mechanism to benefit advertisers. The findings suggest that the mechanism can
benefit the marketplace for ad slots that consistently attract many
advertisers.
|
[
{
"created": "Mon, 3 Nov 2014 21:38:55 GMT",
"version": "v1"
},
{
"created": "Thu, 26 Jan 2017 06:44:28 GMT",
"version": "v2"
}
] |
2017-01-27
|
[
[
"Stourm",
"Valeria",
""
],
[
"Bax",
"Eric",
""
]
] |
Media publisher platforms often face an effectiveness-nuisance tradeoff: more annoying ads can be more effective for some advertisers because of their ability to attract attention, but after attracting viewers' attention, their nuisance to viewers can decrease engagement with the platform over time. With the rise of mobile technology and ad blockers, many platforms are becoming increasingly concerned about how to improve monetization through digital ads while improving viewer experience. We study an online ad auction mechanism that incorporates a charge for ad impact on user experience as a criterion for ad selection and pricing. Like a Pigovian tax, the charge causes advertisers to internalize the hidden cost of foregone future platform revenue due to ad impact on user experience. Over time, the mechanism provides an incentive for advertisers to develop ads that are effective while offering viewers a more pleasant experience. We show that adopting the mechanism can simultaneously benefit the publisher, advertisers, and viewers, even in the short term. Incorporating a charge for ad impact can increase expected advertiser profits if enough advertisers compete. A stronger effectiveness-nuisance tradeoff, meaning that ad effectiveness is more strongly associated with negative impact on user experience, increases the amount of competition required for the mechanism to benefit advertisers. The findings suggest that the mechanism can benefit the marketplace for ad slots that consistently attract many advertisers.
|
2312.15869
|
Piji Li
|
Ruoqing Zhao, Xi Wang, Hongliang Dai, Pan Gao, Piji Li
|
Medical Report Generation based on Segment-Enhanced Contrastive
Representation Learning
|
NLPCC 2023
| null | null | null |
cs.CL cs.AI
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
Automated radiology report generation has the potential to improve radiology
reporting and alleviate the workload of radiologists. However, the medical
report generation task poses unique challenges due to the limited availability
of medical data and the presence of data bias. To maximize the utility of
available data and reduce data bias, we propose MSCL (Medical image
Segmentation with Contrastive Learning), a framework that utilizes the Segment
Anything Model (SAM) to segment organs, abnormalities, bones, etc., and can pay
more attention to the meaningful ROIs in the image to get better visual
representations. Then we introduce a supervised contrastive loss that assigns
more weight to reports that are semantically similar to the target while
training. The design of this loss function aims to mitigate the impact of data
bias and encourage the model to capture the essential features of a medical
image and generate high-quality reports. Experimental results demonstrate the
effectiveness of our proposed model, where we achieve state-of-the-art
performance on the IU X-Ray public dataset.
|
[
{
"created": "Tue, 26 Dec 2023 03:33:48 GMT",
"version": "v1"
}
] |
2023-12-27
|
[
[
"Zhao",
"Ruoqing",
""
],
[
"Wang",
"Xi",
""
],
[
"Dai",
"Hongliang",
""
],
[
"Gao",
"Pan",
""
],
[
"Li",
"Piji",
""
]
] |
Automated radiology report generation has the potential to improve radiology reporting and alleviate the workload of radiologists. However, the medical report generation task poses unique challenges due to the limited availability of medical data and the presence of data bias. To maximize the utility of available data and reduce data bias, we propose MSCL (Medical image Segmentation with Contrastive Learning), a framework that utilizes the Segment Anything Model (SAM) to segment organs, abnormalities, bones, etc., and can pay more attention to the meaningful ROIs in the image to get better visual representations. Then we introduce a supervised contrastive loss that assigns more weight to reports that are semantically similar to the target while training. The design of this loss function aims to mitigate the impact of data bias and encourage the model to capture the essential features of a medical image and generate high-quality reports. Experimental results demonstrate the effectiveness of our proposed model, where we achieve state-of-the-art performance on the IU X-Ray public dataset.
|
1601.06704
|
Nikita Polyanskii
|
A. G. D'yachkov, I.V. Vorobyev, N.A. Polyanskii and V.Yu. Shchukin
|
On a Hypergraph Approach to Multistage Group Testing Problems
|
5 pages, IEEE conference
| null |
10.1109/ISIT.2016.7541486
| null |
cs.IT math.CO math.IT
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
Group testing is a well known search problem that consists in detecting up to
$s$ defective elements of the set $[t]=\{1,\ldots,t\}$ by carrying out tests on
properly chosen subsets of $[t]$. In classical group testing the goal is to
find all defective elements by using the minimal possible number of tests. In
this paper we consider multistage group testing. We propose a general idea how
to use a hypergraph approach to searching defects. For the case $s=2$, we
design an explicit construction, which makes use of $2\log_2t(1+o(1))$ tests in
the worst case and consists of $4$ stages. For the general case $s>2$, we
provide an explicit construction, which uses $(2s-1)\log_2t(1+o(1))$ tests and
consists of $2s-1$ rounds.
|
[
{
"created": "Mon, 25 Jan 2016 18:20:45 GMT",
"version": "v1"
}
] |
2016-11-18
|
[
[
"D'yachkov",
"A. G.",
""
],
[
"Vorobyev",
"I. V.",
""
],
[
"Polyanskii",
"N. A.",
""
],
[
"Shchukin",
"V. Yu.",
""
]
] |
Group testing is a well known search problem that consists in detecting up to $s$ defective elements of the set $[t]=\{1,\ldots,t\}$ by carrying out tests on properly chosen subsets of $[t]$. In classical group testing the goal is to find all defective elements by using the minimal possible number of tests. In this paper we consider multistage group testing. We propose a general idea how to use a hypergraph approach to searching defects. For the case $s=2$, we design an explicit construction, which makes use of $2\log_2t(1+o(1))$ tests in the worst case and consists of $4$ stages. For the general case $s>2$, we provide an explicit construction, which uses $(2s-1)\log_2t(1+o(1))$ tests and consists of $2s-1$ rounds.
|
1506.05527
|
Kim-Kwang Raymond Choo
|
Ben Martini, Quang Do, Kim-Kwang Raymond Choo
|
Conceptual evidence collection and analysis methodology for Android
devices
|
in Cloud Security Ecosystem (Syngress, an Imprint of Elsevier), 2015
| null |
10.1016/B978-0-12-801595-7.00014-8
| null |
cs.CY
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
Android devices continue to grow in popularity and capability meaning the
need for a forensically sound evidence collection methodology for these devices
also increases. This chapter proposes a methodology for evidence collection and
analysis for Android devices that is, as far as practical, device agnostic.
Android devices may contain a significant amount of evidential data that could
be essential to a forensic practitioner in their investigations. However, the
retrieval of this data requires that the practitioner understand and utilize
techniques to analyze information collected from the device. The major
contribution of this research is an in-depth evidence collection and analysis
methodology for forensic practitioners.
|
[
{
"created": "Thu, 18 Jun 2015 01:25:00 GMT",
"version": "v1"
}
] |
2015-06-19
|
[
[
"Martini",
"Ben",
""
],
[
"Do",
"Quang",
""
],
[
"Choo",
"Kim-Kwang Raymond",
""
]
] |
Android devices continue to grow in popularity and capability meaning the need for a forensically sound evidence collection methodology for these devices also increases. This chapter proposes a methodology for evidence collection and analysis for Android devices that is, as far as practical, device agnostic. Android devices may contain a significant amount of evidential data that could be essential to a forensic practitioner in their investigations. However, the retrieval of this data requires that the practitioner understand and utilize techniques to analyze information collected from the device. The major contribution of this research is an in-depth evidence collection and analysis methodology for forensic practitioners.
|
2203.10131
|
Patrick Schnell
|
Patrick Schnell, Philipp Holl, Nils Thuerey
|
Half-Inverse Gradients for Physical Deep Learning
|
ICLR 2022 spotlight, code available at
https://github.com/tum-pbs/half-inverse-gradients
| null | null | null |
cs.LG physics.comp-ph
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
Recent works in deep learning have shown that integrating differentiable
physics simulators into the training process can greatly improve the quality of
results. Although this combination represents a more complex optimization task
than supervised neural network training, the same gradient-based optimizers are
typically employed to minimize the loss function. However, the integrated
physics solvers have a profound effect on the gradient flow as manipulating
scales in magnitude and direction is an inherent property of many physical
processes. Consequently, the gradient flow is often highly unbalanced and
creates an environment in which existing gradient-based optimizers perform
poorly. In this work, we analyze the characteristics of both physical and
neural network optimizations to derive a new method that does not suffer from
this phenomenon. Our method is based on a half-inversion of the Jacobian and
combines principles of both classical network and physics optimizers to solve
the combined optimization task. Compared to state-of-the-art neural network
optimizers, our method converges more quickly and yields better solutions,
which we demonstrate on three complex learning problems involving nonlinear
oscillators, the Schroedinger equation and the Poisson problem.
|
[
{
"created": "Fri, 18 Mar 2022 19:11:04 GMT",
"version": "v1"
}
] |
2022-03-22
|
[
[
"Schnell",
"Patrick",
""
],
[
"Holl",
"Philipp",
""
],
[
"Thuerey",
"Nils",
""
]
] |
Recent works in deep learning have shown that integrating differentiable physics simulators into the training process can greatly improve the quality of results. Although this combination represents a more complex optimization task than supervised neural network training, the same gradient-based optimizers are typically employed to minimize the loss function. However, the integrated physics solvers have a profound effect on the gradient flow as manipulating scales in magnitude and direction is an inherent property of many physical processes. Consequently, the gradient flow is often highly unbalanced and creates an environment in which existing gradient-based optimizers perform poorly. In this work, we analyze the characteristics of both physical and neural network optimizations to derive a new method that does not suffer from this phenomenon. Our method is based on a half-inversion of the Jacobian and combines principles of both classical network and physics optimizers to solve the combined optimization task. Compared to state-of-the-art neural network optimizers, our method converges more quickly and yields better solutions, which we demonstrate on three complex learning problems involving nonlinear oscillators, the Schroedinger equation and the Poisson problem.
|
1903.12468
|
Niveditha Manjunath
|
Ezio Bartocci, Niveditha Manjunath, Leonardo Mariani, Cristinel
Mateis, Dejan Ni\v{c}kovi\'c
|
Automatic Failure Explanation in CPS Models
| null | null | null | null |
cs.SE cs.LO
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
Debugging Cyber-Physical System (CPS) models can be extremely complex.
Indeed, only the detection of a failure is insuffcient to know how to correct a
faulty model. Faults can propagate in time and in space producing observable
misbehaviours in locations completely different from the location of the fault.
Understanding the reason of an observed failure is typically a challenging and
laborious task left to the experience and domain knowledge of the designer. \n
In this paper, we propose CPSDebug, a novel approach that by combining testing,
specification mining, and failure analysis, can automatically explain failures
in Simulink/Stateflow models. We evaluate CPSDebug on two case studies,
involving two use scenarios and several classes of faults, demonstrating the
potential value of our approach.
|
[
{
"created": "Fri, 29 Mar 2019 12:26:42 GMT",
"version": "v1"
}
] |
2020-10-14
|
[
[
"Bartocci",
"Ezio",
""
],
[
"Manjunath",
"Niveditha",
""
],
[
"Mariani",
"Leonardo",
""
],
[
"Mateis",
"Cristinel",
""
],
[
"Ničković",
"Dejan",
""
]
] |
Debugging Cyber-Physical System (CPS) models can be extremely complex. Indeed, only the detection of a failure is insuffcient to know how to correct a faulty model. Faults can propagate in time and in space producing observable misbehaviours in locations completely different from the location of the fault. Understanding the reason of an observed failure is typically a challenging and laborious task left to the experience and domain knowledge of the designer. \n In this paper, we propose CPSDebug, a novel approach that by combining testing, specification mining, and failure analysis, can automatically explain failures in Simulink/Stateflow models. We evaluate CPSDebug on two case studies, involving two use scenarios and several classes of faults, demonstrating the potential value of our approach.
|
1410.7460
|
Ahmed Ewaisha
|
Ahmed Ewaisha and Cihan Tepedelenlio\u{g}lu
|
Throughput Optimization in Multi-Channel Cognitive Radios with Hard
Deadline Constraints
|
Keywords: Delay Constraint, Optimal Stopping Rule, Water Filling,
Stochastic Optimization, Optimal Channel Selection
| null |
10.1109/TVT.2015.2425951
| null |
cs.IT math.IT
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
In a cognitive radio scenario we consider a single secondary user (SU)
accessing a multi-channel system. The SU senses the channels sequentially to
detect if a primary user (PU) is occupying the channels, and stops its search
to access a channel if it offers a significantly high throughput. The optimal
stopping rule and power control problem is considered. The problem is
formulated as a SU's throughput-maximization problem under a power,
interference and packet delay constraints. We first show the effect of the
optimal stopping rule on the packet delay, then solve this optimization problem
for both the overlay system where the SU transmits only at the spectrum holes
as well as the underlay system where tolerable interference (or tolerable
collision probability) is allowed. We provide closed-form expressions for the
optimal stopping rule, and show that the optimal power control strategy for
this multi-channel problem is a modified water-filling approach. We extend the
work to multiple SU scenario and show that when the number of SUs is large the
complexity of the solution becomes smaller than that of the single SU case. We
discuss the application of this problem in typical networks where packets
arrive simultaneously and have the same departure deadline. We further propose
an online adaptation policy to the optimal stopping rule that meets the
packets' hard-deadline constraint and, at the same time, gives higher
throughput than the offline policy.
|
[
{
"created": "Mon, 27 Oct 2014 23:45:23 GMT",
"version": "v1"
},
{
"created": "Fri, 11 Dec 2015 03:20:38 GMT",
"version": "v2"
},
{
"created": "Tue, 29 Dec 2015 14:30:13 GMT",
"version": "v3"
}
] |
2015-12-31
|
[
[
"Ewaisha",
"Ahmed",
""
],
[
"Tepedelenlioğlu",
"Cihan",
""
]
] |
In a cognitive radio scenario we consider a single secondary user (SU) accessing a multi-channel system. The SU senses the channels sequentially to detect if a primary user (PU) is occupying the channels, and stops its search to access a channel if it offers a significantly high throughput. The optimal stopping rule and power control problem is considered. The problem is formulated as a SU's throughput-maximization problem under a power, interference and packet delay constraints. We first show the effect of the optimal stopping rule on the packet delay, then solve this optimization problem for both the overlay system where the SU transmits only at the spectrum holes as well as the underlay system where tolerable interference (or tolerable collision probability) is allowed. We provide closed-form expressions for the optimal stopping rule, and show that the optimal power control strategy for this multi-channel problem is a modified water-filling approach. We extend the work to multiple SU scenario and show that when the number of SUs is large the complexity of the solution becomes smaller than that of the single SU case. We discuss the application of this problem in typical networks where packets arrive simultaneously and have the same departure deadline. We further propose an online adaptation policy to the optimal stopping rule that meets the packets' hard-deadline constraint and, at the same time, gives higher throughput than the offline policy.
|
1809.05369
|
Chris Norval
|
Chris Norval, Jennifer Cobbe, Heleen Janssen, Jatinder Singh
|
Reclaiming Data: Overcoming app identification barriers for exercising
data protection rights
|
Author preprint (accepted 20-Aug-18) To appear in the proceedings of
the 4th Workshop on Legal and Technical Issues in Cloud and Pervasive
Computing (IoT) [CLaw-18], UbiComp/ISWC'18 Adjunct,
https://doi.org/10.1145/3267305.3274153
| null |
10.1145/3267305.3274153
| null |
cs.CY cs.CR cs.HC
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
Data protection regulations generally afford individuals certain rights over
their personal data, including the rights to access, rectify, and delete the
data held on them. Exercising such rights naturally requires those with data
management obligations (service providers) to be able to match an individual
with their data. However, many mobile apps collect personal data, without
requiring user registration or collecting details of a user's identity (email
address, names, phone number, and so forth). As a result, a user's ability to
exercise their rights will be hindered without means for an individual to link
themselves with this 'nameless' data. Current approaches often involve those
seeking to exercise their legal rights having to give the app's provider more
personal information, or even to register for a service; both of which seem
contrary to the spirit of data protection law. This paper explores these
concerns, and indicates simple means for facilitating data subject rights
through both application and mobile platform (OS) design.
|
[
{
"created": "Fri, 14 Sep 2018 12:05:05 GMT",
"version": "v1"
}
] |
2018-09-17
|
[
[
"Norval",
"Chris",
""
],
[
"Cobbe",
"Jennifer",
""
],
[
"Janssen",
"Heleen",
""
],
[
"Singh",
"Jatinder",
""
]
] |
Data protection regulations generally afford individuals certain rights over their personal data, including the rights to access, rectify, and delete the data held on them. Exercising such rights naturally requires those with data management obligations (service providers) to be able to match an individual with their data. However, many mobile apps collect personal data, without requiring user registration or collecting details of a user's identity (email address, names, phone number, and so forth). As a result, a user's ability to exercise their rights will be hindered without means for an individual to link themselves with this 'nameless' data. Current approaches often involve those seeking to exercise their legal rights having to give the app's provider more personal information, or even to register for a service; both of which seem contrary to the spirit of data protection law. This paper explores these concerns, and indicates simple means for facilitating data subject rights through both application and mobile platform (OS) design.
|
2206.04397
|
Rafael Menezes
|
Rafael Menezes, Daniel Moura, Helena Cavalcante, Rosiane de Freitas
and Lucas C. Cordeiro
|
ESBMC-Jimple: Verifying Kotlin Programs via Jimple Intermediate
Representation
|
ACM SIGSOFT International Symposium on Software Testing and Analysis
2022
| null |
10.1145/3533767.3543294
| null |
cs.SE
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
In this work, we describe and evaluate the first model checker for verifying
Kotlin programs through the Jimple intermediate representation. The verifier,
named ESBMC-Jimple, is built on top of the Efficient SMT-based Context-Bounded
Model Checker (ESBMC). It uses the Soot framework to obtain the Jimple IR,
representing a simplified version of the Kotlin source code, containing a
maximum of three operands per instruction. ESBMC-Jimple processes Kotlin source
code together with a model of the standard Kotlin libraries and checks a set of
safety properties. Experimental results show that ESBMC-Jimple can correctly
verify a set of Kotlin benchmarks from the literature and that it is
competitive with state-of-the-art Java bytecode verifiers. A demonstration is
available at https://youtu.be/J6WhNfXvJNc.
|
[
{
"created": "Thu, 9 Jun 2022 10:18:53 GMT",
"version": "v1"
},
{
"created": "Wed, 20 Jul 2022 13:26:30 GMT",
"version": "v2"
}
] |
2022-07-21
|
[
[
"Menezes",
"Rafael",
""
],
[
"Moura",
"Daniel",
""
],
[
"Cavalcante",
"Helena",
""
],
[
"de Freitas",
"Rosiane",
""
],
[
"Cordeiro",
"Lucas C.",
""
]
] |
In this work, we describe and evaluate the first model checker for verifying Kotlin programs through the Jimple intermediate representation. The verifier, named ESBMC-Jimple, is built on top of the Efficient SMT-based Context-Bounded Model Checker (ESBMC). It uses the Soot framework to obtain the Jimple IR, representing a simplified version of the Kotlin source code, containing a maximum of three operands per instruction. ESBMC-Jimple processes Kotlin source code together with a model of the standard Kotlin libraries and checks a set of safety properties. Experimental results show that ESBMC-Jimple can correctly verify a set of Kotlin benchmarks from the literature and that it is competitive with state-of-the-art Java bytecode verifiers. A demonstration is available at https://youtu.be/J6WhNfXvJNc.
|
1211.5520
|
Ashish Tendulkar Dr
|
Vivekanand Samant, Arvind Hulgeri, Alfonso Valencia, Ashish V.
Tendulkar
|
Accurate Demarcation of Protein Domain Linkers based on Structural
Analysis of Linker Probable Region
|
18 pages, 2 figures
|
International Journal of Computational Biology, 0001:01-19, 2012
| null | null |
cs.CE q-bio.BM
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
In multi-domain proteins, the domains are connected by a flexible
unstructured region called as protein domain linker. The accurate demarcation
of these linkers holds a key to understanding of their biochemical and
evolutionary attributes. This knowledge helps in designing a suitable linker
for engineering stable multi-domain chimeric proteins. Here we propose a novel
method for the demarcation of the linker based on a three-dimensional protein
structure and a domain definition. The proposed method is based on biological
knowledge about structural flexibility of the linkers. We performed structural
analysis on a linker probable region (LPR) around domain boundary points of
known SCOP domains. The LPR was described using a set of overlapping peptide
fragments of fixed size. Each peptide fragment was then described by geometric
invariants (GIs) and subjected to clustering process where the fragments
corresponding to actual linker come up as outliers. We then discover the actual
linkers by finding the longest continuous stretch of outlier fragments from
LPRs. This method was evaluated on a benchmark dataset of 51 continuous
multi-domain proteins, where it achieves F1 score of 0.745 (0.83 precision and
0.66 recall). When the method was applied on 725 continuous multi-domain
proteins, it was able to identify novel linkers that were not reported
previously. This method can be used in combination with supervised / sequence
based linker prediction methods for accurate linker demarcation.
|
[
{
"created": "Fri, 23 Nov 2012 14:53:54 GMT",
"version": "v1"
}
] |
2012-11-26
|
[
[
"Samant",
"Vivekanand",
""
],
[
"Hulgeri",
"Arvind",
""
],
[
"Valencia",
"Alfonso",
""
],
[
"Tendulkar",
"Ashish V.",
""
]
] |
In multi-domain proteins, the domains are connected by a flexible unstructured region called as protein domain linker. The accurate demarcation of these linkers holds a key to understanding of their biochemical and evolutionary attributes. This knowledge helps in designing a suitable linker for engineering stable multi-domain chimeric proteins. Here we propose a novel method for the demarcation of the linker based on a three-dimensional protein structure and a domain definition. The proposed method is based on biological knowledge about structural flexibility of the linkers. We performed structural analysis on a linker probable region (LPR) around domain boundary points of known SCOP domains. The LPR was described using a set of overlapping peptide fragments of fixed size. Each peptide fragment was then described by geometric invariants (GIs) and subjected to clustering process where the fragments corresponding to actual linker come up as outliers. We then discover the actual linkers by finding the longest continuous stretch of outlier fragments from LPRs. This method was evaluated on a benchmark dataset of 51 continuous multi-domain proteins, where it achieves F1 score of 0.745 (0.83 precision and 0.66 recall). When the method was applied on 725 continuous multi-domain proteins, it was able to identify novel linkers that were not reported previously. This method can be used in combination with supervised / sequence based linker prediction methods for accurate linker demarcation.
|
2402.04231
|
Subhamoy Maitra
|
Ajeet Kumar and Subhamoy Maitra
|
Further Constructions of AMUBs for Non-prime power Composite Dimensions
| null | null | null | null |
cs.DM
|
http://creativecommons.org/licenses/by/4.0/
|
Construction of a large class of Mutually Unbiased Bases (MUBs) for non-prime
power composite dimensions ($d = k\times s$) is a long standing open problem,
which leads to different construction methods for the class Approximate MUBs
(AMUBs) by relaxing the criterion that the absolute value of the dot product
between two vectors chosen from different bases should be $\leq
\frac{\beta}{\sqrt{d}}$. In this chapter, we consider a more general class of
AMUBs (ARMUBs, considering the real ones too), compared to our earlier work in
[Cryptography and Communications, 14(3): 527--549, 2022]. We note that the
quality of AMUBs (ARMUBs) constructed using RBD$(X,A)$ with $|X|= d$,
critically depends on the parameters, $|s-k|$, $\mu$ (maximum number of
elements common between any pair of blocks), and the set of block sizes. We
present the construction of $\mathcal{O}(\sqrt{d})$ many $\beta$-AMUBs for
composite $d$ when $|s-k|< \sqrt{d}$, using RBDs having block sizes
approximately $\sqrt{d}$, such that $|\braket{\psi^l_i|\psi^m_j}| \leq
\frac{\beta}{\sqrt{d}}$ where $\beta = 1 + \frac{|s-k|}{2\sqrt{d}}+
\mathcal{O}(d^{-1}) \leq 2$. Moreover, if real Hadamard matrix of order $k$ or
$s$ exists, then one can construct at least $N(k)+1$ (or $N(s)+1$) many
$\beta$-ARMUBs for dimension $d$, with $\beta \leq 2 - \frac{|s-k|}{2\sqrt{d}}+
\mathcal{O}(d^{-1})< 2$, where $N(w)$ is the number of MOLS$(w)$. This improves
and generalizes some of our previous results for ARMUBs from two points, viz.,
the real cases are now extended to complex ones too. The earlier efforts use
some existing RBDs, whereas here we consider new instances of RBDs that provide
better results. Similar to the earlier cases, the AMUBs (ARMUBs) constructed
using RBDs are in general very sparse, where the sparsity $(\epsilon)$ is $1 -
\mathcal{O}(d^{-\frac{1}{2}})$.
|
[
{
"created": "Tue, 6 Feb 2024 18:39:25 GMT",
"version": "v1"
}
] |
2024-02-07
|
[
[
"Kumar",
"Ajeet",
""
],
[
"Maitra",
"Subhamoy",
""
]
] |
Construction of a large class of Mutually Unbiased Bases (MUBs) for non-prime power composite dimensions ($d = k\times s$) is a long standing open problem, which leads to different construction methods for the class Approximate MUBs (AMUBs) by relaxing the criterion that the absolute value of the dot product between two vectors chosen from different bases should be $\leq \frac{\beta}{\sqrt{d}}$. In this chapter, we consider a more general class of AMUBs (ARMUBs, considering the real ones too), compared to our earlier work in [Cryptography and Communications, 14(3): 527--549, 2022]. We note that the quality of AMUBs (ARMUBs) constructed using RBD$(X,A)$ with $|X|= d$, critically depends on the parameters, $|s-k|$, $\mu$ (maximum number of elements common between any pair of blocks), and the set of block sizes. We present the construction of $\mathcal{O}(\sqrt{d})$ many $\beta$-AMUBs for composite $d$ when $|s-k|< \sqrt{d}$, using RBDs having block sizes approximately $\sqrt{d}$, such that $|\braket{\psi^l_i|\psi^m_j}| \leq \frac{\beta}{\sqrt{d}}$ where $\beta = 1 + \frac{|s-k|}{2\sqrt{d}}+ \mathcal{O}(d^{-1}) \leq 2$. Moreover, if real Hadamard matrix of order $k$ or $s$ exists, then one can construct at least $N(k)+1$ (or $N(s)+1$) many $\beta$-ARMUBs for dimension $d$, with $\beta \leq 2 - \frac{|s-k|}{2\sqrt{d}}+ \mathcal{O}(d^{-1})< 2$, where $N(w)$ is the number of MOLS$(w)$. This improves and generalizes some of our previous results for ARMUBs from two points, viz., the real cases are now extended to complex ones too. The earlier efforts use some existing RBDs, whereas here we consider new instances of RBDs that provide better results. Similar to the earlier cases, the AMUBs (ARMUBs) constructed using RBDs are in general very sparse, where the sparsity $(\epsilon)$ is $1 - \mathcal{O}(d^{-\frac{1}{2}})$.
|
2003.11456
|
Ralf M\"oller
|
Ralf M\"oller
|
Derivation of Coupled PCA and SVD Learning Rules from a Newton
Zero-Finding Framework
| null | null | null | null |
cs.NE
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
In coupled learning rules for PCA (principal component analysis) and SVD
(singular value decomposition), the update of the estimates of eigenvectors or
singular vectors is influenced by the estimates of eigenvalues or singular
values, respectively. This coupled update mitigates the speed-stability problem
since the update equations converge from all directions with approximately the
same speed. A method to derive coupled learning rules from information criteria
by Newton optimization is known. However, these information criteria have to be
designed, offer no explanatory value, and can only impose Euclidean constraints
on the vector estimates. Here we describe an alternative approach where coupled
PCA and SVD learning rules can systematically be derived from a Newton
zero-finding framework. The derivation starts from an objective function,
combines the equations for its extrema with arbitrary constraints on the vector
estimates, and solves the resulting vector zero-point equation using Newton's
zero-finding method. To demonstrate the framework, we derive PCA and SVD
learning rules with constant Euclidean length or constant sum of the vector
estimates.
|
[
{
"created": "Wed, 25 Mar 2020 15:49:55 GMT",
"version": "v1"
}
] |
2020-03-26
|
[
[
"Möller",
"Ralf",
""
]
] |
In coupled learning rules for PCA (principal component analysis) and SVD (singular value decomposition), the update of the estimates of eigenvectors or singular vectors is influenced by the estimates of eigenvalues or singular values, respectively. This coupled update mitigates the speed-stability problem since the update equations converge from all directions with approximately the same speed. A method to derive coupled learning rules from information criteria by Newton optimization is known. However, these information criteria have to be designed, offer no explanatory value, and can only impose Euclidean constraints on the vector estimates. Here we describe an alternative approach where coupled PCA and SVD learning rules can systematically be derived from a Newton zero-finding framework. The derivation starts from an objective function, combines the equations for its extrema with arbitrary constraints on the vector estimates, and solves the resulting vector zero-point equation using Newton's zero-finding method. To demonstrate the framework, we derive PCA and SVD learning rules with constant Euclidean length or constant sum of the vector estimates.
|
2403.06999
|
Nan Liu
|
Ziwen Wang, Jin Wee Lee, Tanujit Chakraborty, Yilin Ning, Mingxuan
Liu, Feng Xie, Marcus Eng Hock Ong, Nan Liu
|
Survival modeling using deep learning, machine learning and statistical
methods: A comparative analysis for predicting mortality after hospital
admission
| null | null | null | null |
cs.LG cs.AI cs.CY
|
http://creativecommons.org/licenses/by/4.0/
|
Survival analysis is essential for studying time-to-event outcomes and
providing a dynamic understanding of the probability of an event occurring over
time. Various survival analysis techniques, from traditional statistical models
to state-of-the-art machine learning algorithms, support healthcare
intervention and policy decisions. However, there remains ongoing discussion
about their comparative performance. We conducted a comparative study of
several survival analysis methods, including Cox proportional hazards (CoxPH),
stepwise CoxPH, elastic net penalized Cox model, Random Survival Forests (RSF),
Gradient Boosting machine (GBM) learning, AutoScore-Survival, DeepSurv,
time-dependent Cox model based on neural network (CoxTime), and DeepHit
survival neural network. We applied the concordance index (C-index) for model
goodness-of-fit, and integral Brier scores (IBS) for calibration, and
considered the model interpretability. As a case study, we performed a
retrospective analysis of patients admitted through the emergency department of
a tertiary hospital from 2017 to 2019, predicting 90-day all-cause mortality
based on patient demographics, clinicopathological features, and historical
data. The results of the C-index indicate that deep learning achieved
comparable performance, with DeepSurv producing the best discrimination
(DeepSurv: 0.893; CoxTime: 0.892; DeepHit: 0.891). The calibration of DeepSurv
(IBS: 0.041) performed the best, followed by RSF (IBS: 0.042) and GBM (IBS:
0.0421), all using the full variables. Moreover, AutoScore-Survival, using a
minimal variable subset, is easy to interpret, and can achieve good
discrimination and calibration (C-index: 0.867; IBS: 0.044). While all models
were satisfactory, DeepSurv exhibited the best discrimination and calibration.
In addition, AutoScore-Survival offers a more parsimonious model and excellent
interpretability.
|
[
{
"created": "Mon, 4 Mar 2024 10:46:02 GMT",
"version": "v1"
}
] |
2024-03-13
|
[
[
"Wang",
"Ziwen",
""
],
[
"Lee",
"Jin Wee",
""
],
[
"Chakraborty",
"Tanujit",
""
],
[
"Ning",
"Yilin",
""
],
[
"Liu",
"Mingxuan",
""
],
[
"Xie",
"Feng",
""
],
[
"Ong",
"Marcus Eng Hock",
""
],
[
"Liu",
"Nan",
""
]
] |
Survival analysis is essential for studying time-to-event outcomes and providing a dynamic understanding of the probability of an event occurring over time. Various survival analysis techniques, from traditional statistical models to state-of-the-art machine learning algorithms, support healthcare intervention and policy decisions. However, there remains ongoing discussion about their comparative performance. We conducted a comparative study of several survival analysis methods, including Cox proportional hazards (CoxPH), stepwise CoxPH, elastic net penalized Cox model, Random Survival Forests (RSF), Gradient Boosting machine (GBM) learning, AutoScore-Survival, DeepSurv, time-dependent Cox model based on neural network (CoxTime), and DeepHit survival neural network. We applied the concordance index (C-index) for model goodness-of-fit, and integral Brier scores (IBS) for calibration, and considered the model interpretability. As a case study, we performed a retrospective analysis of patients admitted through the emergency department of a tertiary hospital from 2017 to 2019, predicting 90-day all-cause mortality based on patient demographics, clinicopathological features, and historical data. The results of the C-index indicate that deep learning achieved comparable performance, with DeepSurv producing the best discrimination (DeepSurv: 0.893; CoxTime: 0.892; DeepHit: 0.891). The calibration of DeepSurv (IBS: 0.041) performed the best, followed by RSF (IBS: 0.042) and GBM (IBS: 0.0421), all using the full variables. Moreover, AutoScore-Survival, using a minimal variable subset, is easy to interpret, and can achieve good discrimination and calibration (C-index: 0.867; IBS: 0.044). While all models were satisfactory, DeepSurv exhibited the best discrimination and calibration. In addition, AutoScore-Survival offers a more parsimonious model and excellent interpretability.
|
1809.08566
|
Faegheh Hasibi
|
Arash Dargahi Nobari, Arian Askari, Faegheh Hasibi and Mahmood Neshati
|
Query Understanding via Entity Attribute Identification
|
Proceedings of the 27th International Conference on Information and
Knowledge Management (CIKM '18), 2018
| null |
10.1145/3269206.3269245
| null |
cs.IR
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
Understanding searchers' queries is an essential component of semantic search
systems. In many cases, search queries involve specific attributes of an entity
in a knowledge base (KB), which can be further used to find query answers. In
this study, we aim to move forward the understanding of queries by identifying
their related entity attributes from a knowledge base. To this end, we
introduce the task of entity attribute identification and propose two methods
to address it: (i) a model based on Markov Random Field, and (ii) a learning to
rank model. We develop a human annotated test collection and show that our
proposed methods can bring significant improvements over the baseline methods.
|
[
{
"created": "Sun, 23 Sep 2018 09:49:19 GMT",
"version": "v1"
}
] |
2018-09-25
|
[
[
"Nobari",
"Arash Dargahi",
""
],
[
"Askari",
"Arian",
""
],
[
"Hasibi",
"Faegheh",
""
],
[
"Neshati",
"Mahmood",
""
]
] |
Understanding searchers' queries is an essential component of semantic search systems. In many cases, search queries involve specific attributes of an entity in a knowledge base (KB), which can be further used to find query answers. In this study, we aim to move forward the understanding of queries by identifying their related entity attributes from a knowledge base. To this end, we introduce the task of entity attribute identification and propose two methods to address it: (i) a model based on Markov Random Field, and (ii) a learning to rank model. We develop a human annotated test collection and show that our proposed methods can bring significant improvements over the baseline methods.
|
2107.13718
|
Luchuan Song
|
Kun Zhao, Luchuan Song, Bin Liu, Qi Chu, Nenghai Yu
|
Cascaded Residual Density Network for Crowd Counting
| null | null | null | null |
cs.CV
|
http://creativecommons.org/licenses/by/4.0/
|
Crowd counting is a challenging task due to the issues such as scale
variation and perspective variation in real crowd scenes. In this paper, we
propose a novel Cascaded Residual Density Network (CRDNet) in a coarse-to-fine
approach to generate the high-quality density map for crowd counting more
accurately. (1) We estimate the residual density maps by multi-scale pyramidal
features through cascaded residual density modules. It can improve the quality
of density map layer by layer effectively. (2) A novel additional local count
loss is presented to refine the accuracy of crowd counting, which reduces the
errors of pixel-wise Euclidean loss by restricting the number of people in the
local crowd areas. Experiments on two public benchmark datasets show that the
proposed method achieves effective improvement compared with the
state-of-the-art methods.
|
[
{
"created": "Thu, 29 Jul 2021 03:07:11 GMT",
"version": "v1"
}
] |
2021-07-30
|
[
[
"Zhao",
"Kun",
""
],
[
"Song",
"Luchuan",
""
],
[
"Liu",
"Bin",
""
],
[
"Chu",
"Qi",
""
],
[
"Yu",
"Nenghai",
""
]
] |
Crowd counting is a challenging task due to the issues such as scale variation and perspective variation in real crowd scenes. In this paper, we propose a novel Cascaded Residual Density Network (CRDNet) in a coarse-to-fine approach to generate the high-quality density map for crowd counting more accurately. (1) We estimate the residual density maps by multi-scale pyramidal features through cascaded residual density modules. It can improve the quality of density map layer by layer effectively. (2) A novel additional local count loss is presented to refine the accuracy of crowd counting, which reduces the errors of pixel-wise Euclidean loss by restricting the number of people in the local crowd areas. Experiments on two public benchmark datasets show that the proposed method achieves effective improvement compared with the state-of-the-art methods.
|
2306.05442
|
Zhaoyang Huang
|
Zhaoyang Huang, Xiaoyu Shi, Chao Zhang, Qiang Wang, Yijin Li, Hongwei
Qin, Jifeng Dai, Xiaogang Wang, and Hongsheng Li
|
FlowFormer: A Transformer Architecture and Its Masked Cost Volume
Autoencoding for Optical Flow
|
arXiv admin note: substantial text overlap with arXiv:2203.16194,
arXiv:2303.01237
| null | null | null |
cs.CV
|
http://creativecommons.org/licenses/by-sa/4.0/
|
This paper introduces a novel transformer-based network architecture,
FlowFormer, along with the Masked Cost Volume AutoEncoding (MCVA) for
pretraining it to tackle the problem of optical flow estimation. FlowFormer
tokenizes the 4D cost-volume built from the source-target image pair and
iteratively refines flow estimation with a cost-volume encoder-decoder
architecture. The cost-volume encoder derives a cost memory with
alternate-group transformer~(AGT) layers in a latent space and the decoder
recurrently decodes flow from the cost memory with dynamic positional cost
queries. On the Sintel benchmark, FlowFormer architecture achieves 1.16 and
2.09 average end-point-error~(AEPE) on the clean and final pass, a 16.5\% and
15.5\% error reduction from the GMA~(1.388 and 2.47). MCVA enhances FlowFormer
by pretraining the cost-volume encoder with a masked autoencoding scheme, which
further unleashes the capability of FlowFormer with unlabeled data. This is
especially critical in optical flow estimation because ground truth flows are
more expensive to acquire than labels in other vision tasks. MCVA improves
FlowFormer all-sided and FlowFormer+MCVA ranks 1st among all published methods
on both Sintel and KITTI-2015 benchmarks and achieves the best generalization
performance. Specifically, FlowFormer+MCVA achieves 1.07 and 1.94 AEPE on the
Sintel benchmark, leading to 7.76\% and 7.18\% error reductions from
FlowFormer.
|
[
{
"created": "Thu, 8 Jun 2023 12:24:04 GMT",
"version": "v1"
}
] |
2023-06-12
|
[
[
"Huang",
"Zhaoyang",
""
],
[
"Shi",
"Xiaoyu",
""
],
[
"Zhang",
"Chao",
""
],
[
"Wang",
"Qiang",
""
],
[
"Li",
"Yijin",
""
],
[
"Qin",
"Hongwei",
""
],
[
"Dai",
"Jifeng",
""
],
[
"Wang",
"Xiaogang",
""
],
[
"Li",
"Hongsheng",
""
]
] |
This paper introduces a novel transformer-based network architecture, FlowFormer, along with the Masked Cost Volume AutoEncoding (MCVA) for pretraining it to tackle the problem of optical flow estimation. FlowFormer tokenizes the 4D cost-volume built from the source-target image pair and iteratively refines flow estimation with a cost-volume encoder-decoder architecture. The cost-volume encoder derives a cost memory with alternate-group transformer~(AGT) layers in a latent space and the decoder recurrently decodes flow from the cost memory with dynamic positional cost queries. On the Sintel benchmark, FlowFormer architecture achieves 1.16 and 2.09 average end-point-error~(AEPE) on the clean and final pass, a 16.5\% and 15.5\% error reduction from the GMA~(1.388 and 2.47). MCVA enhances FlowFormer by pretraining the cost-volume encoder with a masked autoencoding scheme, which further unleashes the capability of FlowFormer with unlabeled data. This is especially critical in optical flow estimation because ground truth flows are more expensive to acquire than labels in other vision tasks. MCVA improves FlowFormer all-sided and FlowFormer+MCVA ranks 1st among all published methods on both Sintel and KITTI-2015 benchmarks and achieves the best generalization performance. Specifically, FlowFormer+MCVA achieves 1.07 and 1.94 AEPE on the Sintel benchmark, leading to 7.76\% and 7.18\% error reductions from FlowFormer.
|
2206.01781
|
Jaskaran Grover
|
Jaskaran Grover and Changliu Liu and Katia Sycara
|
The Before, During, and After of Multi-Robot Deadlock
|
Accepted to International Journal of Robotics Research 2022, WAFR
2020 Special Issue
| null | null | null |
cs.RO cs.MA math.OC
|
http://creativecommons.org/licenses/by/4.0/
|
Collision avoidance for multirobot systems is a well-studied problem.
Recently, control barrier functions (CBFs) have been proposed for synthesizing
controllers that guarantee collision avoidance and goal stabilization for
multiple robots. However, it has been noted that reactive control synthesis
methods (such as CBFs) are prone to \textit{deadlock}, an equilibrium of system
dynamics that causes the robots to stall before reaching their goals. In this
paper, we analyze the closed-loop dynamics of robots using CBFs, to
characterize controller parameters, initial conditions, and goal locations that
invariably lead the system to deadlock. Using tools from duality theory, we
derive geometric properties of robot configurations of an $N$ robot system once
it is in deadlock and we justify them using the mechanics interpretation of KKT
conditions. Our key deductions are that 1) system deadlock is characterized by
a force-equilibrium on robots and 2) deadlock occurs to ensure safety when
safety is on the brink of being violated. These deductions allow us to
interpret deadlock as a subset of the state space, and we show that this set is
non-empty and located on the boundary of the safe set. By exploiting these
properties, we analyze the number of admissible robot configurations in
deadlock and develop a provably-correct decentralized algorithm for deadlock
resolution to safely deliver the robots to their goals. This algorithm is
validated in simulations as well as experimentally on Khepera-IV robots.
|
[
{
"created": "Fri, 3 Jun 2022 18:48:55 GMT",
"version": "v1"
}
] |
2022-06-07
|
[
[
"Grover",
"Jaskaran",
""
],
[
"Liu",
"Changliu",
""
],
[
"Sycara",
"Katia",
""
]
] |
Collision avoidance for multirobot systems is a well-studied problem. Recently, control barrier functions (CBFs) have been proposed for synthesizing controllers that guarantee collision avoidance and goal stabilization for multiple robots. However, it has been noted that reactive control synthesis methods (such as CBFs) are prone to \textit{deadlock}, an equilibrium of system dynamics that causes the robots to stall before reaching their goals. In this paper, we analyze the closed-loop dynamics of robots using CBFs, to characterize controller parameters, initial conditions, and goal locations that invariably lead the system to deadlock. Using tools from duality theory, we derive geometric properties of robot configurations of an $N$ robot system once it is in deadlock and we justify them using the mechanics interpretation of KKT conditions. Our key deductions are that 1) system deadlock is characterized by a force-equilibrium on robots and 2) deadlock occurs to ensure safety when safety is on the brink of being violated. These deductions allow us to interpret deadlock as a subset of the state space, and we show that this set is non-empty and located on the boundary of the safe set. By exploiting these properties, we analyze the number of admissible robot configurations in deadlock and develop a provably-correct decentralized algorithm for deadlock resolution to safely deliver the robots to their goals. This algorithm is validated in simulations as well as experimentally on Khepera-IV robots.
|
1910.00511
|
Yang Zhang
|
Yang Zhang, Shiyu Chang, Mo Yu, Kaizhi Qian
|
An Efficient and Margin-Approaching Zero-Confidence Adversarial Attack
| null | null | null | null |
cs.LG stat.ML
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
There are two major paradigms of white-box adversarial attacks that attempt
to impose input perturbations. The first paradigm, called the fix-perturbation
attack, crafts adversarial samples within a given perturbation level. The
second paradigm, called the zero-confidence attack, finds the smallest
perturbation needed to cause mis-classification, also known as the margin of an
input feature. While the former paradigm is well-resolved, the latter is not.
Existing zero-confidence attacks either introduce significant ap-proximation
errors, or are too time-consuming. We therefore propose MARGINATTACK, a
zero-confidence attack framework that is able to compute the margin with
improved accuracy and efficiency. Our experiments show that MARGINATTACK is
able to compute a smaller margin than the state-of-the-art zero-confidence
attacks, and matches the state-of-the-art fix-perturbation at-tacks. In
addition, it runs significantly faster than the Carlini-Wagner attack,
currently the most ac-curate zero-confidence attack algorithm.
|
[
{
"created": "Tue, 1 Oct 2019 15:59:52 GMT",
"version": "v1"
}
] |
2019-10-02
|
[
[
"Zhang",
"Yang",
""
],
[
"Chang",
"Shiyu",
""
],
[
"Yu",
"Mo",
""
],
[
"Qian",
"Kaizhi",
""
]
] |
There are two major paradigms of white-box adversarial attacks that attempt to impose input perturbations. The first paradigm, called the fix-perturbation attack, crafts adversarial samples within a given perturbation level. The second paradigm, called the zero-confidence attack, finds the smallest perturbation needed to cause mis-classification, also known as the margin of an input feature. While the former paradigm is well-resolved, the latter is not. Existing zero-confidence attacks either introduce significant ap-proximation errors, or are too time-consuming. We therefore propose MARGINATTACK, a zero-confidence attack framework that is able to compute the margin with improved accuracy and efficiency. Our experiments show that MARGINATTACK is able to compute a smaller margin than the state-of-the-art zero-confidence attacks, and matches the state-of-the-art fix-perturbation at-tacks. In addition, it runs significantly faster than the Carlini-Wagner attack, currently the most ac-curate zero-confidence attack algorithm.
|
2307.15198
|
Yao Su
|
Yao Su and Zhentian Qian and Lei Ma and Lifang He and Xiangnan Kong
|
One-shot Joint Extraction, Registration and Segmentation of Neuroimaging
Data
|
Published as a research track paper at KDD 2023. Code:
https://github.com/Anonymous4545/JERS
| null |
10.1145/3580305.3599452
| null |
cs.CV cs.AI
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
Brain extraction, registration and segmentation are indispensable
preprocessing steps in neuroimaging studies. The aim is to extract the brain
from raw imaging scans (i.e., extraction step), align it with a target brain
image (i.e., registration step) and label the anatomical brain regions (i.e.,
segmentation step). Conventional studies typically focus on developing separate
methods for the extraction, registration and segmentation tasks in a supervised
setting. The performance of these methods is largely contingent on the quantity
of training samples and the extent of visual inspections carried out by experts
for error correction. Nevertheless, collecting voxel-level labels and
performing manual quality control on high-dimensional neuroimages (e.g., 3D
MRI) are expensive and time-consuming in many medical studies. In this paper,
we study the problem of one-shot joint extraction, registration and
segmentation in neuroimaging data, which exploits only one labeled template
image (a.k.a. atlas) and a few unlabeled raw images for training. We propose a
unified end-to-end framework, called JERS, to jointly optimize the extraction,
registration and segmentation tasks, allowing feedback among them.
Specifically, we use a group of extraction, registration and segmentation
modules to learn the extraction mask, transformation and segmentation mask,
where modules are interconnected and mutually reinforced by self-supervision.
Empirical results on real-world datasets demonstrate that our proposed method
performs exceptionally in the extraction, registration and segmentation tasks.
Our code and data can be found at https://github.com/Anonymous4545/JERS
|
[
{
"created": "Thu, 27 Jul 2023 21:14:40 GMT",
"version": "v1"
}
] |
2023-07-31
|
[
[
"Su",
"Yao",
""
],
[
"Qian",
"Zhentian",
""
],
[
"Ma",
"Lei",
""
],
[
"He",
"Lifang",
""
],
[
"Kong",
"Xiangnan",
""
]
] |
Brain extraction, registration and segmentation are indispensable preprocessing steps in neuroimaging studies. The aim is to extract the brain from raw imaging scans (i.e., extraction step), align it with a target brain image (i.e., registration step) and label the anatomical brain regions (i.e., segmentation step). Conventional studies typically focus on developing separate methods for the extraction, registration and segmentation tasks in a supervised setting. The performance of these methods is largely contingent on the quantity of training samples and the extent of visual inspections carried out by experts for error correction. Nevertheless, collecting voxel-level labels and performing manual quality control on high-dimensional neuroimages (e.g., 3D MRI) are expensive and time-consuming in many medical studies. In this paper, we study the problem of one-shot joint extraction, registration and segmentation in neuroimaging data, which exploits only one labeled template image (a.k.a. atlas) and a few unlabeled raw images for training. We propose a unified end-to-end framework, called JERS, to jointly optimize the extraction, registration and segmentation tasks, allowing feedback among them. Specifically, we use a group of extraction, registration and segmentation modules to learn the extraction mask, transformation and segmentation mask, where modules are interconnected and mutually reinforced by self-supervision. Empirical results on real-world datasets demonstrate that our proposed method performs exceptionally in the extraction, registration and segmentation tasks. Our code and data can be found at https://github.com/Anonymous4545/JERS
|
2310.10910
|
Davut Emre Ta\c{s}ar
|
Davut Emre Tasar, Kutan Koruyan, Ceren Ocal Tasar
|
Machine Learning in the Quantum Age: Quantum vs. Classical Support
Vector Machines
|
6 Pages, in Turkish language
| null | null | null |
cs.LG
|
http://creativecommons.org/licenses/by/4.0/
|
This work endeavors to juxtapose the efficacy of machine learning algorithms
within classical and quantum computational paradigms. Particularly, by
emphasizing on Support Vector Machines (SVM), we scrutinize the classification
prowess of classical SVM and Quantum Support Vector Machines (QSVM) operational
on quantum hardware over the Iris dataset. The methodology embraced
encapsulates an extensive array of experiments orchestrated through the Qiskit
library, alongside hyperparameter optimization. The findings unveil that in
particular scenarios, QSVMs extend a level of accuracy that can vie with
classical SVMs, albeit the execution times are presently protracted. Moreover,
we underscore that augmenting quantum computational capacity and the magnitude
of parallelism can markedly ameliorate the performance of quantum machine
learning algorithms. This inquiry furnishes invaluable insights regarding the
extant scenario and future potentiality of machine learning applications in the
quantum epoch. Colab: https://t.ly/QKuz0
|
[
{
"created": "Tue, 17 Oct 2023 01:06:59 GMT",
"version": "v1"
}
] |
2023-10-18
|
[
[
"Tasar",
"Davut Emre",
""
],
[
"Koruyan",
"Kutan",
""
],
[
"Tasar",
"Ceren Ocal",
""
]
] |
This work endeavors to juxtapose the efficacy of machine learning algorithms within classical and quantum computational paradigms. Particularly, by emphasizing on Support Vector Machines (SVM), we scrutinize the classification prowess of classical SVM and Quantum Support Vector Machines (QSVM) operational on quantum hardware over the Iris dataset. The methodology embraced encapsulates an extensive array of experiments orchestrated through the Qiskit library, alongside hyperparameter optimization. The findings unveil that in particular scenarios, QSVMs extend a level of accuracy that can vie with classical SVMs, albeit the execution times are presently protracted. Moreover, we underscore that augmenting quantum computational capacity and the magnitude of parallelism can markedly ameliorate the performance of quantum machine learning algorithms. This inquiry furnishes invaluable insights regarding the extant scenario and future potentiality of machine learning applications in the quantum epoch. Colab: https://t.ly/QKuz0
|
1304.0357
|
Arkadiusz Stopczynski Mr.
|
Arkadiusz Stopczynski, Carsten Stahlhut, Jakob Eg Larsen, Michael Kai
Petersen, and Lars Kai Hansen
|
The Smartphone Brain Scanner: A Mobile Real-time Neuroimaging System
| null | null |
10.1371/journal.pone.0086733
| null |
cs.HC
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
Combining low cost wireless EEG sensors with smartphones offers novel
opportunities for mobile brain imaging in an everyday context. We present a
framework for building multi-platform, portable EEG applications with real-time
3D source reconstruction. The system - Smartphone Brain Scanner - combines an
off-the-shelf neuroheadset or EEG cap with a smartphone or tablet, and as such
represents the first fully mobile system for real-time 3D EEG imaging. We
discuss the benefits and challenges of a fully portable system, including
technical limitations as well as real-time reconstruction of 3D images of brain
activity. We present examples of the brain activity captured in a simple
experiment involving imagined finger tapping, showing that the acquired signal
in a relevant brain region is similar to that obtained with standard EEG lab
equipment. Although the quality of the signal in a mobile solution using a
off-the-shelf consumer neuroheadset is lower compared to that obtained using
high density standard EEG equipment, we propose that mobile application
development may offset the disadvantages and provide completely new
opportunities for neuroimaging in natural settings.
|
[
{
"created": "Mon, 1 Apr 2013 13:51:52 GMT",
"version": "v1"
}
] |
2014-03-05
|
[
[
"Stopczynski",
"Arkadiusz",
""
],
[
"Stahlhut",
"Carsten",
""
],
[
"Larsen",
"Jakob Eg",
""
],
[
"Petersen",
"Michael Kai",
""
],
[
"Hansen",
"Lars Kai",
""
]
] |
Combining low cost wireless EEG sensors with smartphones offers novel opportunities for mobile brain imaging in an everyday context. We present a framework for building multi-platform, portable EEG applications with real-time 3D source reconstruction. The system - Smartphone Brain Scanner - combines an off-the-shelf neuroheadset or EEG cap with a smartphone or tablet, and as such represents the first fully mobile system for real-time 3D EEG imaging. We discuss the benefits and challenges of a fully portable system, including technical limitations as well as real-time reconstruction of 3D images of brain activity. We present examples of the brain activity captured in a simple experiment involving imagined finger tapping, showing that the acquired signal in a relevant brain region is similar to that obtained with standard EEG lab equipment. Although the quality of the signal in a mobile solution using a off-the-shelf consumer neuroheadset is lower compared to that obtained using high density standard EEG equipment, we propose that mobile application development may offset the disadvantages and provide completely new opportunities for neuroimaging in natural settings.
|
2207.06590
|
Jun Yang
|
Jun Yang, Yuehan Wang, Yiling Lou, Ming Wen and Lingming Zhang
|
Attention: Not Just Another Dataset for Patch-Correctness Checking
| null | null | null | null |
cs.SE
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
Automated Program Repair (APR) techniques have drawn wide attention from both
academia and industry. Meanwhile, one main limitation with the current
state-of-the-art APR tools is that patches passing all the original tests are
not necessarily the correct ones wanted by developers, i.e., the plausible
patch problem. To date, various Patch-Correctness Checking (PCC) techniques
have been proposed to address this important issue. However, they are only
evaluated on very limited datasets as the APR tools used for generating such
patches can only explore a small subset of the search space of possible
patches, posing serious threats to external validity to existing PCC studies.
In this paper, we construct an extensive PCC dataset (the largest manually
labeled PCC dataset to our knowledge) to revisit all state-of-the-art PCC
techniques. More specifically, our PCC dataset includes 1,988 patches generated
from the recent PraPR APR tool, which leverages highly-optimized bytecode-level
patch executions and can exhaustively explore all possible plausible patches
within its large predefined search space (including well-known fixing patterns
from various prior APR tools). Our extensive study of representative PCC
techniques on the new dataset has revealed various surprising findings and
provided guidelines for future PCC research.
|
[
{
"created": "Thu, 14 Jul 2022 01:07:17 GMT",
"version": "v1"
},
{
"created": "Wed, 8 Feb 2023 23:10:09 GMT",
"version": "v2"
}
] |
2023-02-10
|
[
[
"Yang",
"Jun",
""
],
[
"Wang",
"Yuehan",
""
],
[
"Lou",
"Yiling",
""
],
[
"Wen",
"Ming",
""
],
[
"Zhang",
"Lingming",
""
]
] |
Automated Program Repair (APR) techniques have drawn wide attention from both academia and industry. Meanwhile, one main limitation with the current state-of-the-art APR tools is that patches passing all the original tests are not necessarily the correct ones wanted by developers, i.e., the plausible patch problem. To date, various Patch-Correctness Checking (PCC) techniques have been proposed to address this important issue. However, they are only evaluated on very limited datasets as the APR tools used for generating such patches can only explore a small subset of the search space of possible patches, posing serious threats to external validity to existing PCC studies. In this paper, we construct an extensive PCC dataset (the largest manually labeled PCC dataset to our knowledge) to revisit all state-of-the-art PCC techniques. More specifically, our PCC dataset includes 1,988 patches generated from the recent PraPR APR tool, which leverages highly-optimized bytecode-level patch executions and can exhaustively explore all possible plausible patches within its large predefined search space (including well-known fixing patterns from various prior APR tools). Our extensive study of representative PCC techniques on the new dataset has revealed various surprising findings and provided guidelines for future PCC research.
|
2001.04325
|
Eli Chen
|
Oren Haik, Oded Perry, Eli Chen, Peter Klammer
|
A Novel Inspection System For Variable Data Printing Using Deep Learning
|
Accepted for publication in: Winter Applications of Computer Vision
(WACV) 2020
| null | null | null |
cs.CV
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
We present a novel approach for inspecting variable data prints (VDP) with an
ultra-low false alarm rate (0.005%) and potential applicability to other
real-world problems. The system is based on a comparison between two images: a
reference image and an image captured by low-cost scanners. The comparison task
is challenging as low-cost imaging systems create artifacts that may
erroneously be classified as true (genuine) defects. To address this challenge
we introduce two new fusion methods, for change detection applications, which
are both fast and efficient. The first is an early fusion method that combines
the two input images into a single pseudo-color image. The second, called
Change-Detection Single Shot Detector (CD-SSD) leverages the SSD by fusing
features in the middle of the network. We demonstrate the effectiveness of the
proposed deep learning-based approach with a large dataset from real-world
printing scenarios. Finally, we evaluate our models on a different domain of
aerial imagery change detection (AICD). Our best method clearly outperforms the
state-of-the-art baseline on this dataset.
|
[
{
"created": "Mon, 13 Jan 2020 15:07:13 GMT",
"version": "v1"
}
] |
2020-01-14
|
[
[
"Haik",
"Oren",
""
],
[
"Perry",
"Oded",
""
],
[
"Chen",
"Eli",
""
],
[
"Klammer",
"Peter",
""
]
] |
We present a novel approach for inspecting variable data prints (VDP) with an ultra-low false alarm rate (0.005%) and potential applicability to other real-world problems. The system is based on a comparison between two images: a reference image and an image captured by low-cost scanners. The comparison task is challenging as low-cost imaging systems create artifacts that may erroneously be classified as true (genuine) defects. To address this challenge we introduce two new fusion methods, for change detection applications, which are both fast and efficient. The first is an early fusion method that combines the two input images into a single pseudo-color image. The second, called Change-Detection Single Shot Detector (CD-SSD) leverages the SSD by fusing features in the middle of the network. We demonstrate the effectiveness of the proposed deep learning-based approach with a large dataset from real-world printing scenarios. Finally, we evaluate our models on a different domain of aerial imagery change detection (AICD). Our best method clearly outperforms the state-of-the-art baseline on this dataset.
|
2402.18117
|
Changqi Wang
|
Haoyu Xie, Changqi Wang, Jian Zhao, Yang Liu, Jun Dan, Chong Fu,
Baigui Sun
|
PRCL: Probabilistic Representation Contrastive Learning for
Semi-Supervised Semantic Segmentation
|
19 pages, 11 figures
| null | null | null |
cs.CV cs.LG
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
Tremendous breakthroughs have been developed in Semi-Supervised Semantic
Segmentation (S4) through contrastive learning. However, due to limited
annotations, the guidance on unlabeled images is generated by the model itself,
which inevitably exists noise and disturbs the unsupervised training process.
To address this issue, we propose a robust contrastive-based S4 framework,
termed the Probabilistic Representation Contrastive Learning (PRCL) framework
to enhance the robustness of the unsupervised training process. We model the
pixel-wise representation as Probabilistic Representations (PR) via
multivariate Gaussian distribution and tune the contribution of the ambiguous
representations to tolerate the risk of inaccurate guidance in contrastive
learning. Furthermore, we introduce Global Distribution Prototypes (GDP) by
gathering all PRs throughout the whole training process. Since the GDP contains
the information of all representations with the same class, it is robust from
the instant noise in representations and bears the intra-class variance of
representations. In addition, we generate Virtual Negatives (VNs) based on GDP
to involve the contrastive learning process. Extensive experiments on two
public benchmarks demonstrate the superiority of our PRCL framework.
|
[
{
"created": "Wed, 28 Feb 2024 07:10:37 GMT",
"version": "v1"
}
] |
2024-02-29
|
[
[
"Xie",
"Haoyu",
""
],
[
"Wang",
"Changqi",
""
],
[
"Zhao",
"Jian",
""
],
[
"Liu",
"Yang",
""
],
[
"Dan",
"Jun",
""
],
[
"Fu",
"Chong",
""
],
[
"Sun",
"Baigui",
""
]
] |
Tremendous breakthroughs have been developed in Semi-Supervised Semantic Segmentation (S4) through contrastive learning. However, due to limited annotations, the guidance on unlabeled images is generated by the model itself, which inevitably exists noise and disturbs the unsupervised training process. To address this issue, we propose a robust contrastive-based S4 framework, termed the Probabilistic Representation Contrastive Learning (PRCL) framework to enhance the robustness of the unsupervised training process. We model the pixel-wise representation as Probabilistic Representations (PR) via multivariate Gaussian distribution and tune the contribution of the ambiguous representations to tolerate the risk of inaccurate guidance in contrastive learning. Furthermore, we introduce Global Distribution Prototypes (GDP) by gathering all PRs throughout the whole training process. Since the GDP contains the information of all representations with the same class, it is robust from the instant noise in representations and bears the intra-class variance of representations. In addition, we generate Virtual Negatives (VNs) based on GDP to involve the contrastive learning process. Extensive experiments on two public benchmarks demonstrate the superiority of our PRCL framework.
|
2109.15200
|
Zhao Hengling
|
Hengling Zhao, Yipeng Liu, Xiaolin Huang and Ce Zhu
|
Semi-tensor Product-based TensorDecomposition for Neural Network
Compression
| null | null | null | null |
cs.LG
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
The existing tensor networks adopt conventional matrix product for
connection. The classical matrix product requires strict dimensionality
consistency between factors, which can result in redundancy in data
representation. In this paper, the semi-tensor product is used to generalize
classical matrix product-based mode product to semi-tensor mode product. As it
permits the connection of two factors with different dimensionality, more
flexible and compact tensor decompositions can be obtained with smaller sizes
of factors. Tucker decomposition, Tensor Train (TT) and Tensor Ring (TR) are
common decomposition for low rank compression of deep neural networks. The
semi-tensor product is applied to these tensor decompositions to obtained their
generalized versions, i.e., semi-tensor Tucker decomposition (STTu),
semi-tensor train(STT) and semi-tensor ring (STR). Experimental results show
the STTu, STT and STR achieve higher compression factors than the conventional
tensor decompositions with the same accuracy but less training times in ResNet
and WideResNetcompression. With 2% accuracy degradation, the TT-RN (rank = 14)
and the TR-WRN (rank = 16) only obtain 3 times and99t times compression factors
while the STT-RN (rank = 14) and the STR-WRN (rank = 16) achieve 9 times and
179 times compression factors, respectively.
|
[
{
"created": "Thu, 30 Sep 2021 15:18:14 GMT",
"version": "v1"
}
] |
2021-10-01
|
[
[
"Zhao",
"Hengling",
""
],
[
"Liu",
"Yipeng",
""
],
[
"Huang",
"Xiaolin",
""
],
[
"Zhu",
"Ce",
""
]
] |
The existing tensor networks adopt conventional matrix product for connection. The classical matrix product requires strict dimensionality consistency between factors, which can result in redundancy in data representation. In this paper, the semi-tensor product is used to generalize classical matrix product-based mode product to semi-tensor mode product. As it permits the connection of two factors with different dimensionality, more flexible and compact tensor decompositions can be obtained with smaller sizes of factors. Tucker decomposition, Tensor Train (TT) and Tensor Ring (TR) are common decomposition for low rank compression of deep neural networks. The semi-tensor product is applied to these tensor decompositions to obtained their generalized versions, i.e., semi-tensor Tucker decomposition (STTu), semi-tensor train(STT) and semi-tensor ring (STR). Experimental results show the STTu, STT and STR achieve higher compression factors than the conventional tensor decompositions with the same accuracy but less training times in ResNet and WideResNetcompression. With 2% accuracy degradation, the TT-RN (rank = 14) and the TR-WRN (rank = 16) only obtain 3 times and99t times compression factors while the STT-RN (rank = 14) and the STR-WRN (rank = 16) achieve 9 times and 179 times compression factors, respectively.
|
2002.03704
|
Sebastian Farquhar
|
Sebastian Farquhar, Lewis Smith, Yarin Gal
|
Liberty or Depth: Deep Bayesian Neural Nets Do Not Need Complex Weight
Posterior Approximations
|
Advances In Neural Information Processing Systems. 2020
| null | null | null |
cs.LG stat.ML
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
We challenge the longstanding assumption that the mean-field approximation
for variational inference in Bayesian neural networks is severely restrictive,
and show this is not the case in deep networks. We prove several results
indicating that deep mean-field variational weight posteriors can induce
similar distributions in function-space to those induced by shallower networks
with complex weight posteriors. We validate our theoretical contributions
empirically, both through examination of the weight posterior using Hamiltonian
Monte Carlo in small models and by comparing diagonal- to structured-covariance
in large settings. Since complex variational posteriors are often expensive and
cumbersome to implement, our results suggest that using mean-field variational
inference in a deeper model is both a practical and theoretically justified
alternative to structured approximations.
|
[
{
"created": "Mon, 10 Feb 2020 13:11:45 GMT",
"version": "v1"
},
{
"created": "Wed, 8 Jul 2020 10:39:50 GMT",
"version": "v2"
},
{
"created": "Mon, 2 Nov 2020 11:55:29 GMT",
"version": "v3"
},
{
"created": "Wed, 10 Mar 2021 09:19:13 GMT",
"version": "v4"
}
] |
2021-03-11
|
[
[
"Farquhar",
"Sebastian",
""
],
[
"Smith",
"Lewis",
""
],
[
"Gal",
"Yarin",
""
]
] |
We challenge the longstanding assumption that the mean-field approximation for variational inference in Bayesian neural networks is severely restrictive, and show this is not the case in deep networks. We prove several results indicating that deep mean-field variational weight posteriors can induce similar distributions in function-space to those induced by shallower networks with complex weight posteriors. We validate our theoretical contributions empirically, both through examination of the weight posterior using Hamiltonian Monte Carlo in small models and by comparing diagonal- to structured-covariance in large settings. Since complex variational posteriors are often expensive and cumbersome to implement, our results suggest that using mean-field variational inference in a deeper model is both a practical and theoretically justified alternative to structured approximations.
|
1303.2223
|
C. Titus Brown
|
Eric McDonald and C. Titus Brown
|
khmer: Working with Big Data in Bioinformatics
|
Invited chapter for forthcoming book on Performance of Open Source
Applications
| null | null | null |
cs.CE q-bio.GN
|
http://creativecommons.org/licenses/by/3.0/
|
We introduce design and optimization considerations for the 'khmer' package.
|
[
{
"created": "Sat, 9 Mar 2013 15:34:25 GMT",
"version": "v1"
}
] |
2013-03-12
|
[
[
"McDonald",
"Eric",
""
],
[
"Brown",
"C. Titus",
""
]
] |
We introduce design and optimization considerations for the 'khmer' package.
|
1009.3145
|
Petros Boufounos
|
Petros T. Boufounos
|
Universal Rate-Efficient Scalar Quantization
| null | null | null | null |
cs.IT math.IT
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
Scalar quantization is the most practical and straightforward approach to
signal quantization. However, it has been shown that scalar quantization of
oversampled or Compressively Sensed signals can be inefficient in terms of the
rate-distortion trade-off, especially as the oversampling rate or the sparsity
of the signal increases. In this paper, we modify the scalar quantizer to have
discontinuous quantization regions. We demonstrate that with this modification
it is possible to achieve exponential decay of the quantization error as a
function of the oversampling rate instead of the quadratic decay exhibited by
current approaches. Our approach is universal in the sense that prior knowledge
of the signal model is not necessary in the quantizer design, only in the
reconstruction. Thus, we demonstrate that it is possible to reduce the
quantization error by incorporating side information on the acquired signal,
such as sparse signal models or signal similarity with known signals. In doing
so, we establish a relationship between quantization performance and the
Kolmogorov entropy of the signal model.
|
[
{
"created": "Thu, 16 Sep 2010 11:14:09 GMT",
"version": "v1"
},
{
"created": "Mon, 4 Oct 2010 17:56:24 GMT",
"version": "v2"
},
{
"created": "Thu, 14 Jul 2011 23:53:44 GMT",
"version": "v3"
}
] |
2011-07-18
|
[
[
"Boufounos",
"Petros T.",
""
]
] |
Scalar quantization is the most practical and straightforward approach to signal quantization. However, it has been shown that scalar quantization of oversampled or Compressively Sensed signals can be inefficient in terms of the rate-distortion trade-off, especially as the oversampling rate or the sparsity of the signal increases. In this paper, we modify the scalar quantizer to have discontinuous quantization regions. We demonstrate that with this modification it is possible to achieve exponential decay of the quantization error as a function of the oversampling rate instead of the quadratic decay exhibited by current approaches. Our approach is universal in the sense that prior knowledge of the signal model is not necessary in the quantizer design, only in the reconstruction. Thus, we demonstrate that it is possible to reduce the quantization error by incorporating side information on the acquired signal, such as sparse signal models or signal similarity with known signals. In doing so, we establish a relationship between quantization performance and the Kolmogorov entropy of the signal model.
|
2002.05153
|
Andrew Bennett
|
Andrew Bennett and Nathan Kallus
|
Efficient Policy Learning from Surrogate-Loss Classification Reductions
| null | null | null | null |
cs.LG econ.EM math.ST stat.ML stat.TH
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
Recent work on policy learning from observational data has highlighted the
importance of efficient policy evaluation and has proposed reductions to
weighted (cost-sensitive) classification. But, efficient policy evaluation need
not yield efficient estimation of policy parameters. We consider the estimation
problem given by a weighted surrogate-loss classification reduction of policy
learning with any score function, either direct, inverse-propensity weighted,
or doubly robust. We show that, under a correct specification assumption, the
weighted classification formulation need not be efficient for policy
parameters. We draw a contrast to actual (possibly weighted) binary
classification, where correct specification implies a parametric model, while
for policy learning it only implies a semiparametric model. In light of this,
we instead propose an estimation approach based on generalized method of
moments, which is efficient for the policy parameters. We propose a particular
method based on recent developments on solving moment problems using neural
networks and demonstrate the efficiency and regret benefits of this method
empirically.
|
[
{
"created": "Wed, 12 Feb 2020 18:54:41 GMT",
"version": "v1"
}
] |
2020-02-13
|
[
[
"Bennett",
"Andrew",
""
],
[
"Kallus",
"Nathan",
""
]
] |
Recent work on policy learning from observational data has highlighted the importance of efficient policy evaluation and has proposed reductions to weighted (cost-sensitive) classification. But, efficient policy evaluation need not yield efficient estimation of policy parameters. We consider the estimation problem given by a weighted surrogate-loss classification reduction of policy learning with any score function, either direct, inverse-propensity weighted, or doubly robust. We show that, under a correct specification assumption, the weighted classification formulation need not be efficient for policy parameters. We draw a contrast to actual (possibly weighted) binary classification, where correct specification implies a parametric model, while for policy learning it only implies a semiparametric model. In light of this, we instead propose an estimation approach based on generalized method of moments, which is efficient for the policy parameters. We propose a particular method based on recent developments on solving moment problems using neural networks and demonstrate the efficiency and regret benefits of this method empirically.
|
2011.09967
|
Bin Wang
|
Wanshi Hong, Cong Zhang, Cy Chan, Bin Wang
|
Electric Vehicle Charging Infrastructure Planning: A Scalable
Computational Framework
| null | null | null | null |
cs.AI cs.NE eess.SP math.OC
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
The optimal charging infrastructure planning problem over a large geospatial
area is challenging due to the increasing network sizes of the transportation
system and the electric grid. The coupling between the electric vehicle travel
behaviors and charging events is therefore complex. This paper focuses on the
demonstration of a scalable computational framework for the electric vehicle
charging infrastructure planning over the tightly integrated transportation and
electric grid networks. On the transportation side, a charging profile
generation strategy is proposed leveraging the EV energy consumption model,
trip routing, and charger selection methods. On the grid side, a genetic
algorithm is utilized within the optimal power flow program to solve the
optimal charger placement problem with integer variables by adaptively
evaluating candidate solutions in the current iteration and generating new
solutions for the next iterations.
|
[
{
"created": "Tue, 17 Nov 2020 16:48:07 GMT",
"version": "v1"
}
] |
2020-11-20
|
[
[
"Hong",
"Wanshi",
""
],
[
"Zhang",
"Cong",
""
],
[
"Chan",
"Cy",
""
],
[
"Wang",
"Bin",
""
]
] |
The optimal charging infrastructure planning problem over a large geospatial area is challenging due to the increasing network sizes of the transportation system and the electric grid. The coupling between the electric vehicle travel behaviors and charging events is therefore complex. This paper focuses on the demonstration of a scalable computational framework for the electric vehicle charging infrastructure planning over the tightly integrated transportation and electric grid networks. On the transportation side, a charging profile generation strategy is proposed leveraging the EV energy consumption model, trip routing, and charger selection methods. On the grid side, a genetic algorithm is utilized within the optimal power flow program to solve the optimal charger placement problem with integer variables by adaptively evaluating candidate solutions in the current iteration and generating new solutions for the next iterations.
|
1502.06108
|
Xiao Lin
|
Xiao Lin, Devi Parikh
|
Don't Just Listen, Use Your Imagination: Leveraging Visual Common Sense
for Non-Visual Tasks
| null | null | null | null |
cs.CV
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
Artificial agents today can answer factual questions. But they fall short on
questions that require common sense reasoning. Perhaps this is because most
existing common sense databases rely on text to learn and represent knowledge.
But much of common sense knowledge is unwritten - partly because it tends not
to be interesting enough to talk about, and partly because some common sense is
unnatural to articulate in text. While unwritten, it is not unseen. In this
paper we leverage semantic common sense knowledge learned from images - i.e.
visual common sense - in two textual tasks: fill-in-the-blank and visual
paraphrasing. We propose to "imagine" the scene behind the text, and leverage
visual cues from the "imagined" scenes in addition to textual cues while
answering these questions. We imagine the scenes as a visual abstraction. Our
approach outperforms a strong text-only baseline on these tasks. Our proposed
tasks can serve as benchmarks to quantitatively evaluate progress in solving
tasks that go "beyond recognition". Our code and datasets are publicly
available.
|
[
{
"created": "Sat, 21 Feb 2015 15:25:40 GMT",
"version": "v1"
},
{
"created": "Tue, 5 May 2015 18:54:05 GMT",
"version": "v2"
},
{
"created": "Wed, 29 Jul 2015 03:04:19 GMT",
"version": "v3"
}
] |
2015-07-30
|
[
[
"Lin",
"Xiao",
""
],
[
"Parikh",
"Devi",
""
]
] |
Artificial agents today can answer factual questions. But they fall short on questions that require common sense reasoning. Perhaps this is because most existing common sense databases rely on text to learn and represent knowledge. But much of common sense knowledge is unwritten - partly because it tends not to be interesting enough to talk about, and partly because some common sense is unnatural to articulate in text. While unwritten, it is not unseen. In this paper we leverage semantic common sense knowledge learned from images - i.e. visual common sense - in two textual tasks: fill-in-the-blank and visual paraphrasing. We propose to "imagine" the scene behind the text, and leverage visual cues from the "imagined" scenes in addition to textual cues while answering these questions. We imagine the scenes as a visual abstraction. Our approach outperforms a strong text-only baseline on these tasks. Our proposed tasks can serve as benchmarks to quantitatively evaluate progress in solving tasks that go "beyond recognition". Our code and datasets are publicly available.
|
1612.07117
|
Nam Khanh Tran
|
Nam Khanh Tran
|
Classification and Learning-to-rank Approaches for Cross-Device Matching
at CIKM Cup 2016
|
CIKM Cup 2016
| null | null | null |
cs.IR cs.LG
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
In this paper, we propose two methods for tackling the problem of
cross-device matching for online advertising at CIKM Cup 2016. The first method
considers the matching problem as a binary classification task and solve it by
utilizing ensemble learning techniques. The second method defines the matching
problem as a ranking task and effectively solve it with using learning-to-rank
algorithms. The results show that the proposed methods obtain promising
results, in which the ranking-based method outperforms the classification-based
method for the task.
|
[
{
"created": "Tue, 20 Dec 2016 15:02:41 GMT",
"version": "v1"
}
] |
2016-12-22
|
[
[
"Tran",
"Nam Khanh",
""
]
] |
In this paper, we propose two methods for tackling the problem of cross-device matching for online advertising at CIKM Cup 2016. The first method considers the matching problem as a binary classification task and solve it by utilizing ensemble learning techniques. The second method defines the matching problem as a ranking task and effectively solve it with using learning-to-rank algorithms. The results show that the proposed methods obtain promising results, in which the ranking-based method outperforms the classification-based method for the task.
|
2010.09810
|
Congyu Wu
|
Congyu Wu
|
Connections between Relational Event Model and Inverse Reinforcement
Learning for Characterizing Group Interaction Sequences
| null | null | null | null |
cs.LG cs.SI
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
In this paper we explore previously unidentified connections between
relational event model (REM) from the field of network science and inverse
reinforcement learning (IRL) from the field of machine learning with respect to
their ability to characterize sequences of directed social interaction events
in group settings. REM is a conventional approach to tackle such a problem
whereas the application of IRL is a largely unbeaten path. We begin by
examining the mathematical components of both REM and IRL and find
straightforward analogies between the two methods as well as unique
characteristics of the IRL approach. We demonstrate the special utility of IRL
in characterizing group social interactions with an empirical experiment, in
which we use IRL to infer individual behavioral preferences based on a sequence
of directed communication events from a group of virtual-reality game players
interacting and cooperating to accomplish a shared goal. Our comparison and
experiment introduce fresh perspectives for social behavior analytics and help
inspire new research opportunities at the nexus of social network analysis and
machine learning.
|
[
{
"created": "Mon, 19 Oct 2020 19:40:29 GMT",
"version": "v1"
}
] |
2020-10-21
|
[
[
"Wu",
"Congyu",
""
]
] |
In this paper we explore previously unidentified connections between relational event model (REM) from the field of network science and inverse reinforcement learning (IRL) from the field of machine learning with respect to their ability to characterize sequences of directed social interaction events in group settings. REM is a conventional approach to tackle such a problem whereas the application of IRL is a largely unbeaten path. We begin by examining the mathematical components of both REM and IRL and find straightforward analogies between the two methods as well as unique characteristics of the IRL approach. We demonstrate the special utility of IRL in characterizing group social interactions with an empirical experiment, in which we use IRL to infer individual behavioral preferences based on a sequence of directed communication events from a group of virtual-reality game players interacting and cooperating to accomplish a shared goal. Our comparison and experiment introduce fresh perspectives for social behavior analytics and help inspire new research opportunities at the nexus of social network analysis and machine learning.
|
2306.00858
|
Simon Keizer
|
Simon Keizer, Caroline Dockes, Norbert Braunschweiler, Svetlana
Stoyanchev, Rama Doddipatla
|
Adversarial learning of neural user simulators for dialogue policy
optimisation
|
UK Speech 2023
| null | null | null |
cs.CL
|
http://creativecommons.org/licenses/by-nc-nd/4.0/
|
Reinforcement learning based dialogue policies are typically trained in
interaction with a user simulator. To obtain an effective and robust policy,
this simulator should generate user behaviour that is both realistic and
varied. Current data-driven simulators are trained to accurately model the user
behaviour in a dialogue corpus. We propose an alternative method using
adversarial learning, with the aim to simulate realistic user behaviour with
more variation. We train and evaluate several simulators on a corpus of
restaurant search dialogues, and then use them to train dialogue system
policies. In policy cross-evaluation experiments we demonstrate that an
adversarially trained simulator produces policies with 8.3% higher success rate
than those trained with a maximum likelihood simulator. Subjective results from
a crowd-sourced dialogue system user evaluation confirm the effectiveness of
adversarially training user simulators.
|
[
{
"created": "Thu, 1 Jun 2023 16:17:16 GMT",
"version": "v1"
}
] |
2023-06-02
|
[
[
"Keizer",
"Simon",
""
],
[
"Dockes",
"Caroline",
""
],
[
"Braunschweiler",
"Norbert",
""
],
[
"Stoyanchev",
"Svetlana",
""
],
[
"Doddipatla",
"Rama",
""
]
] |
Reinforcement learning based dialogue policies are typically trained in interaction with a user simulator. To obtain an effective and robust policy, this simulator should generate user behaviour that is both realistic and varied. Current data-driven simulators are trained to accurately model the user behaviour in a dialogue corpus. We propose an alternative method using adversarial learning, with the aim to simulate realistic user behaviour with more variation. We train and evaluate several simulators on a corpus of restaurant search dialogues, and then use them to train dialogue system policies. In policy cross-evaluation experiments we demonstrate that an adversarially trained simulator produces policies with 8.3% higher success rate than those trained with a maximum likelihood simulator. Subjective results from a crowd-sourced dialogue system user evaluation confirm the effectiveness of adversarially training user simulators.
|
2408.02320
|
Yuxin Chen
|
Gen Li and Yuting Wei and Yuejie Chi and Yuxin Chen
|
A Sharp Convergence Theory for The Probability Flow ODEs of Diffusion
Models
|
This manuscript presents improved theory for probability flow ODEs
compared to its earlier version arXiv:2306.09251
| null | null | null |
cs.LG cs.NA eess.SP math.NA math.ST stat.ML stat.TH
|
http://creativecommons.org/licenses/by/4.0/
|
Diffusion models, which convert noise into new data instances by learning to
reverse a diffusion process, have become a cornerstone in contemporary
generative modeling. In this work, we develop non-asymptotic convergence theory
for a popular diffusion-based sampler (i.e., the probability flow ODE sampler)
in discrete time, assuming access to $\ell_2$-accurate estimates of the (Stein)
score functions. For distributions in $\mathbb{R}^d$, we prove that
$d/\varepsilon$ iterations -- modulo some logarithmic and lower-order terms --
are sufficient to approximate the target distribution to within $\varepsilon$
total-variation distance. This is the first result establishing nearly linear
dimension-dependency (in $d$) for the probability flow ODE sampler. Imposing
only minimal assumptions on the target data distribution (e.g., no smoothness
assumption is imposed), our results also characterize how $\ell_2$ score
estimation errors affect the quality of the data generation processes. In
contrast to prior works, our theory is developed based on an elementary yet
versatile non-asymptotic approach without the need of resorting to SDE and ODE
toolboxes.
|
[
{
"created": "Mon, 5 Aug 2024 09:02:24 GMT",
"version": "v1"
}
] |
2024-08-06
|
[
[
"Li",
"Gen",
""
],
[
"Wei",
"Yuting",
""
],
[
"Chi",
"Yuejie",
""
],
[
"Chen",
"Yuxin",
""
]
] |
Diffusion models, which convert noise into new data instances by learning to reverse a diffusion process, have become a cornerstone in contemporary generative modeling. In this work, we develop non-asymptotic convergence theory for a popular diffusion-based sampler (i.e., the probability flow ODE sampler) in discrete time, assuming access to $\ell_2$-accurate estimates of the (Stein) score functions. For distributions in $\mathbb{R}^d$, we prove that $d/\varepsilon$ iterations -- modulo some logarithmic and lower-order terms -- are sufficient to approximate the target distribution to within $\varepsilon$ total-variation distance. This is the first result establishing nearly linear dimension-dependency (in $d$) for the probability flow ODE sampler. Imposing only minimal assumptions on the target data distribution (e.g., no smoothness assumption is imposed), our results also characterize how $\ell_2$ score estimation errors affect the quality of the data generation processes. In contrast to prior works, our theory is developed based on an elementary yet versatile non-asymptotic approach without the need of resorting to SDE and ODE toolboxes.
|
2105.04683
|
Mattia Rigotti
|
Rong Zhu, Mattia Rigotti
|
Deep Bandits Show-Off: Simple and Efficient Exploration with Deep
Networks
| null |
35th Conference on Neural Information Processing Systems (NeurIPS
2021), Sydney, Australia
| null | null |
cs.LG stat.ML
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
Designing efficient exploration is central to Reinforcement Learning due to
the fundamental problem posed by the exploration-exploitation dilemma. Bayesian
exploration strategies like Thompson Sampling resolve this trade-off in a
principled way by modeling and updating the distribution of the parameters of
the action-value function, the outcome model of the environment. However, this
technique becomes infeasible for complex environments due to the computational
intractability of maintaining probability distributions over parameters of
outcome models of corresponding complexity. Moreover, the approximation
techniques introduced to mitigate this issue typically result in poor
exploration-exploitation trade-offs, as observed in the case of deep neural
network models with approximate posterior methods that have been shown to
underperform in the deep bandit scenario. In this paper we introduce Sample
Average Uncertainty (SAU), a simple and efficient uncertainty measure for
contextual bandits. While Bayesian approaches like Thompson Sampling estimate
outcomes uncertainty indirectly by first quantifying the variability over the
parameters of the outcome model, SAU is a frequentist approach that directly
estimates the uncertainty of the outcomes based on the value predictions.
Importantly, we show theoretically that the uncertainty measure estimated by
SAU asymptotically matches the uncertainty provided by Thompson Sampling, as
well as its regret bounds. Because of its simplicity SAU can be seamlessly
applied to deep contextual bandits as a very scalable drop-in replacement for
epsilon-greedy exploration. We confirm empirically our theory by showing that
SAU-based exploration outperforms current state-of-the-art deep Bayesian bandit
methods on several real-world datasets at modest computation cost. Code is
available at \url{https://github.com/ibm/sau-explore}.
|
[
{
"created": "Mon, 10 May 2021 21:45:01 GMT",
"version": "v1"
},
{
"created": "Tue, 26 Oct 2021 09:28:25 GMT",
"version": "v2"
}
] |
2021-10-27
|
[
[
"Zhu",
"Rong",
""
],
[
"Rigotti",
"Mattia",
""
]
] |
Designing efficient exploration is central to Reinforcement Learning due to the fundamental problem posed by the exploration-exploitation dilemma. Bayesian exploration strategies like Thompson Sampling resolve this trade-off in a principled way by modeling and updating the distribution of the parameters of the action-value function, the outcome model of the environment. However, this technique becomes infeasible for complex environments due to the computational intractability of maintaining probability distributions over parameters of outcome models of corresponding complexity. Moreover, the approximation techniques introduced to mitigate this issue typically result in poor exploration-exploitation trade-offs, as observed in the case of deep neural network models with approximate posterior methods that have been shown to underperform in the deep bandit scenario. In this paper we introduce Sample Average Uncertainty (SAU), a simple and efficient uncertainty measure for contextual bandits. While Bayesian approaches like Thompson Sampling estimate outcomes uncertainty indirectly by first quantifying the variability over the parameters of the outcome model, SAU is a frequentist approach that directly estimates the uncertainty of the outcomes based on the value predictions. Importantly, we show theoretically that the uncertainty measure estimated by SAU asymptotically matches the uncertainty provided by Thompson Sampling, as well as its regret bounds. Because of its simplicity SAU can be seamlessly applied to deep contextual bandits as a very scalable drop-in replacement for epsilon-greedy exploration. We confirm empirically our theory by showing that SAU-based exploration outperforms current state-of-the-art deep Bayesian bandit methods on several real-world datasets at modest computation cost. Code is available at \url{https://github.com/ibm/sau-explore}.
|
2106.00846
|
Babatunji Omoniwa
|
Babatunji Omoniwa, Riaz Hussain, Muhammad Adil, Atif Shakeel, Ahmed
Kamal Tahir, Qadeer Ul Hasan, and Shahzad A. Malik
|
An Optimal Relay Scheme for Outage Minimization in Fog-based
Internet-of-Things (IoT) Networks
|
Accepted and Published in IEEE Internet of Things Journal
| null | null | null |
cs.NI
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
Fog devices are beginning to play a key role in relaying data and services
within the Internet-of-Things (IoT) ecosystem. These relays may be static or
mobile, with the latter offering a new degree of freedom for performance
improvement via careful relay mobility design. Besides that, power conservation
has been a prevalent issue in IoT networks with devices being
power-constrained, requiring optimal power-control mechanisms. In this paper,
we consider a multi-tier fog-based IoT architecture where a mobile/static fog
node acts as an amplify and forward relay that transmits received information
from a sensor node to a higher hierarchically-placed static fog device, which
offers some localized services. The outage probability of the presented
scenario was efficiently minimized by jointly optimizing the mobility pattern
and the transmit power of the fog relay. A closed-form analytical expression
for the outage probability was derived. Furthermore, due to the intractability
and non-convexity of the formulated problem, we applied an iterative algorithm
based on the steepest descent method to arrive at a desirable objective.
Simulations reveal that the outage probability was improved by 62.7% in the
optimized-location fixed-power (OLFP) scheme, 79.3% in the optimized-power
fixed-location (OPFL) scheme, and 94.2% in the optimized-location
optimized-power (OLOP) scheme, as against the fixed-location and fixed-power
(FLFP) scheme (i.e., without optimization). Lastly, we present an optimal relay
selection strategy that chooses an appropriate relay node from randomly
distributed relaying candidates.
|
[
{
"created": "Tue, 1 Jun 2021 22:57:51 GMT",
"version": "v1"
}
] |
2021-06-03
|
[
[
"Omoniwa",
"Babatunji",
""
],
[
"Hussain",
"Riaz",
""
],
[
"Adil",
"Muhammad",
""
],
[
"Shakeel",
"Atif",
""
],
[
"Tahir",
"Ahmed Kamal",
""
],
[
"Hasan",
"Qadeer Ul",
""
],
[
"Malik",
"Shahzad A.",
""
]
] |
Fog devices are beginning to play a key role in relaying data and services within the Internet-of-Things (IoT) ecosystem. These relays may be static or mobile, with the latter offering a new degree of freedom for performance improvement via careful relay mobility design. Besides that, power conservation has been a prevalent issue in IoT networks with devices being power-constrained, requiring optimal power-control mechanisms. In this paper, we consider a multi-tier fog-based IoT architecture where a mobile/static fog node acts as an amplify and forward relay that transmits received information from a sensor node to a higher hierarchically-placed static fog device, which offers some localized services. The outage probability of the presented scenario was efficiently minimized by jointly optimizing the mobility pattern and the transmit power of the fog relay. A closed-form analytical expression for the outage probability was derived. Furthermore, due to the intractability and non-convexity of the formulated problem, we applied an iterative algorithm based on the steepest descent method to arrive at a desirable objective. Simulations reveal that the outage probability was improved by 62.7% in the optimized-location fixed-power (OLFP) scheme, 79.3% in the optimized-power fixed-location (OPFL) scheme, and 94.2% in the optimized-location optimized-power (OLOP) scheme, as against the fixed-location and fixed-power (FLFP) scheme (i.e., without optimization). Lastly, we present an optimal relay selection strategy that chooses an appropriate relay node from randomly distributed relaying candidates.
|
2008.08750
|
Sayed Kamaledin Ghiasi-Shirazi
|
Ramin Zarei Sabzevar, Kamaledin Ghiasi-Shirazi, Ahad Harati
|
Prototype-based interpretation of the functionality of neurons in
winner-take-all neural networks
| null | null | null | null |
cs.LG cs.NE stat.ML
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
Prototype-based learning (PbL) using a winner-take-all (WTA) network based on
minimum Euclidean distance (ED-WTA) is an intuitive approach to multiclass
classification. By constructing meaningful class centers, PbL provides higher
interpretability and generalization than hyperplane-based learning (HbL)
methods based on maximum Inner Product (IP-WTA) and can efficiently detect and
reject samples that do not belong to any classes. In this paper, we first prove
the equivalence of IP-WTA and ED-WTA from a representational point of view.
Then, we show that naively using this equivalence leads to unintuitive ED-WTA
networks in which the centers have high distances to data that they represent.
We propose $\pm$ED-WTA which models each neuron with two prototypes: one
positive prototype representing samples that are modeled by this neuron and a
negative prototype representing the samples that are erroneously won by that
neuron during training. We propose a novel training algorithm for the
$\pm$ED-WTA network, which cleverly switches between updating the positive and
negative prototypes and is essential to the emergence of interpretable
prototypes. Unexpectedly, we observed that the negative prototype of each
neuron is indistinguishably similar to the positive one. The rationale behind
this observation is that the training data that are mistaken with a prototype
are indeed similar to it. The main finding of this paper is this interpretation
of the functionality of neurons as computing the difference between the
distances to a positive and a negative prototype, which is in agreement with
the BCM theory. In our experiments, we show that the proposed $\pm$ED-WTA
method constructs highly interpretable prototypes that can be successfully used
for detecting outlier and adversarial examples.
|
[
{
"created": "Thu, 20 Aug 2020 03:15:37 GMT",
"version": "v1"
}
] |
2020-08-21
|
[
[
"Sabzevar",
"Ramin Zarei",
""
],
[
"Ghiasi-Shirazi",
"Kamaledin",
""
],
[
"Harati",
"Ahad",
""
]
] |
Prototype-based learning (PbL) using a winner-take-all (WTA) network based on minimum Euclidean distance (ED-WTA) is an intuitive approach to multiclass classification. By constructing meaningful class centers, PbL provides higher interpretability and generalization than hyperplane-based learning (HbL) methods based on maximum Inner Product (IP-WTA) and can efficiently detect and reject samples that do not belong to any classes. In this paper, we first prove the equivalence of IP-WTA and ED-WTA from a representational point of view. Then, we show that naively using this equivalence leads to unintuitive ED-WTA networks in which the centers have high distances to data that they represent. We propose $\pm$ED-WTA which models each neuron with two prototypes: one positive prototype representing samples that are modeled by this neuron and a negative prototype representing the samples that are erroneously won by that neuron during training. We propose a novel training algorithm for the $\pm$ED-WTA network, which cleverly switches between updating the positive and negative prototypes and is essential to the emergence of interpretable prototypes. Unexpectedly, we observed that the negative prototype of each neuron is indistinguishably similar to the positive one. The rationale behind this observation is that the training data that are mistaken with a prototype are indeed similar to it. The main finding of this paper is this interpretation of the functionality of neurons as computing the difference between the distances to a positive and a negative prototype, which is in agreement with the BCM theory. In our experiments, we show that the proposed $\pm$ED-WTA method constructs highly interpretable prototypes that can be successfully used for detecting outlier and adversarial examples.
|
2011.05961
|
Orpaz Goldstein
|
Orpaz Goldstein, Mohammad Kachuee, Derek Shiell, Majid Sarrafzadeh
|
Real-Time Decentralized knowledge Transfer at the Edge
| null | null | null | null |
cs.LG
|
http://creativecommons.org/licenses/by-sa/4.0/
|
The proliferation of edge networks creates islands of learning agents working
on local streams of data. Transferring knowledge between these agents in
real-time without exposing private data allows for collaboration to decrease
learning time and increase model confidence. Incorporating knowledge from data
that a local model did not see creates an ability to debias a local model or
add to classification abilities on data never before seen. Transferring
knowledge in a selective decentralized approach enables models to retain their
local insights, allowing for local flavors of a machine learning model. This
approach suits the decentralized architecture of edge networks, as a local edge
node will serve a community of learning agents that will likely encounter
similar data. We propose a method based on knowledge distillation for pairwise
knowledge transfer pipelines from models trained on non-i.i.d. data and compare
it to other popular knowledge transfer methods. Additionally, we test different
scenarios of knowledge transfer network construction and show the practicality
of our approach. Our experiments show knowledge transfer using our model
outperforms standard methods in a real-time transfer scenario.
|
[
{
"created": "Wed, 11 Nov 2020 18:26:57 GMT",
"version": "v1"
},
{
"created": "Fri, 25 Dec 2020 00:16:58 GMT",
"version": "v2"
},
{
"created": "Tue, 28 Sep 2021 23:55:34 GMT",
"version": "v3"
},
{
"created": "Fri, 1 Oct 2021 16:12:29 GMT",
"version": "v4"
}
] |
2021-10-04
|
[
[
"Goldstein",
"Orpaz",
""
],
[
"Kachuee",
"Mohammad",
""
],
[
"Shiell",
"Derek",
""
],
[
"Sarrafzadeh",
"Majid",
""
]
] |
The proliferation of edge networks creates islands of learning agents working on local streams of data. Transferring knowledge between these agents in real-time without exposing private data allows for collaboration to decrease learning time and increase model confidence. Incorporating knowledge from data that a local model did not see creates an ability to debias a local model or add to classification abilities on data never before seen. Transferring knowledge in a selective decentralized approach enables models to retain their local insights, allowing for local flavors of a machine learning model. This approach suits the decentralized architecture of edge networks, as a local edge node will serve a community of learning agents that will likely encounter similar data. We propose a method based on knowledge distillation for pairwise knowledge transfer pipelines from models trained on non-i.i.d. data and compare it to other popular knowledge transfer methods. Additionally, we test different scenarios of knowledge transfer network construction and show the practicality of our approach. Our experiments show knowledge transfer using our model outperforms standard methods in a real-time transfer scenario.
|
2304.02560
|
Kumara Kahatapitiya
|
Kumara Kahatapitiya, Anurag Arnab, Arsha Nagrani, Michael S. Ryoo
|
VicTR: Video-conditioned Text Representations for Activity Recognition
|
To appear at CVPR 2024
| null | null | null |
cs.CV
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
Vision-Language models (VLMs) have excelled in the image-domain -- especially
in zero-shot settings -- thanks to the availability of vast pretraining data
(i.e., paired image-text samples). However for videos, such paired data is not
as abundant. Therefore, video-VLMs are usually designed by adapting pretrained
image-VLMs to the video-domain, instead of training from scratch. All such
recipes rely on augmenting visual embeddings with temporal information (i.e.,
image $\rightarrow$ video), often keeping text embeddings unchanged or even
being discarded. In this paper, we argue the contrary, that better video-VLMs
can be designed by focusing more on augmenting text, rather than visual
information. More specifically, we introduce Video-conditioned Text
Representations (VicTR): a form of text embeddings optimized w.r.t. visual
embeddings, creating a more-flexible contrastive latent space. Our model can
further make use of freely-available semantic information, in the form of
visually-grounded auxiliary text (e.g. object or scene information). We
evaluate our model on few-shot, zero-shot (HMDB-51, UCF-101), short-form
(Kinetics-400) and long-form (Charades) activity recognition benchmarks,
showing strong performance among video-VLMs.
|
[
{
"created": "Wed, 5 Apr 2023 16:30:36 GMT",
"version": "v1"
},
{
"created": "Fri, 29 Mar 2024 16:56:33 GMT",
"version": "v2"
}
] |
2024-04-01
|
[
[
"Kahatapitiya",
"Kumara",
""
],
[
"Arnab",
"Anurag",
""
],
[
"Nagrani",
"Arsha",
""
],
[
"Ryoo",
"Michael S.",
""
]
] |
Vision-Language models (VLMs) have excelled in the image-domain -- especially in zero-shot settings -- thanks to the availability of vast pretraining data (i.e., paired image-text samples). However for videos, such paired data is not as abundant. Therefore, video-VLMs are usually designed by adapting pretrained image-VLMs to the video-domain, instead of training from scratch. All such recipes rely on augmenting visual embeddings with temporal information (i.e., image $\rightarrow$ video), often keeping text embeddings unchanged or even being discarded. In this paper, we argue the contrary, that better video-VLMs can be designed by focusing more on augmenting text, rather than visual information. More specifically, we introduce Video-conditioned Text Representations (VicTR): a form of text embeddings optimized w.r.t. visual embeddings, creating a more-flexible contrastive latent space. Our model can further make use of freely-available semantic information, in the form of visually-grounded auxiliary text (e.g. object or scene information). We evaluate our model on few-shot, zero-shot (HMDB-51, UCF-101), short-form (Kinetics-400) and long-form (Charades) activity recognition benchmarks, showing strong performance among video-VLMs.
|
2105.14465
|
Pradipta Biswas
|
Gowdham Prabhakar and Pradipta Biswas
|
A Brief Survey on Interactive Automotive UI
| null | null | null | null |
cs.HC
|
http://creativecommons.org/licenses/by/4.0/
|
Automotive User Interface (AutoUI) is relatively a new discipline in the
context of both Transportation Engineering and Human Machine Interaction (HMI).
It covers various HMI aspects both inside and outside vehicle ranging from
operating the vehicle itself, undertaking various secondary tasks, driver
behaviour analysis, cognitive load estimation and so on. This review paper
discusses various interactive HMI inside a vehicle used for undertaking
secondary tasks. We divided recent HMIs through four sections on virtual touch
interfaces, wearable devices, speech recognition and non-visual interfaces and
eye gaze controlled systems. Finally, we summarized advantages and
disadvantages of various technologies.
|
[
{
"created": "Sun, 30 May 2021 08:37:35 GMT",
"version": "v1"
}
] |
2021-06-01
|
[
[
"Prabhakar",
"Gowdham",
""
],
[
"Biswas",
"Pradipta",
""
]
] |
Automotive User Interface (AutoUI) is relatively a new discipline in the context of both Transportation Engineering and Human Machine Interaction (HMI). It covers various HMI aspects both inside and outside vehicle ranging from operating the vehicle itself, undertaking various secondary tasks, driver behaviour analysis, cognitive load estimation and so on. This review paper discusses various interactive HMI inside a vehicle used for undertaking secondary tasks. We divided recent HMIs through four sections on virtual touch interfaces, wearable devices, speech recognition and non-visual interfaces and eye gaze controlled systems. Finally, we summarized advantages and disadvantages of various technologies.
|
1707.03319
|
Rahmat Widia Sembiring
|
Dewi Sartika Ginting, Kristin Sitompul, Jasael Simanulang, Rahmat
Widia Sembiring, Muhammad Zarlis
|
Modification of Symmetric Cryptography with Combining Affine Chiper and
Caesar Chiper which Dynamic Nature in Matrix of Chiper Transposition by
Applying Flow Pattern in the Planting Rice
|
2nd International Conference of Computer, Environment, Social
Science, Health Science, Agriculture & Technology (ICEST) 2017
|
Advances in Science, Technology and Engineering Systems Journal
(ASTESJ), Adv. Sci. Technol. Eng. Syst. J. 2(5), 1-5 (2017)
|
10.25046/aj020502
| null |
cs.CR
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
Classical cryptography is a way of disguising the news done by the people
when there was no computer. The goal is to protect information by way of
encoding. This paper describesa modification of classical algorithms to make
cryptanalis difficult to steal undisclosed messages. There are three types of
classical algorithms that are combined affine chiper, Caesar chiper and chiper
transposition. Where for chiperteks affine chiper and Caesar chiper can be
looped as much as the initial key, because the result can be varied as much as
key value, then affine chiper and Caesar chiper in this case is dynamic. Then
the results of the affine and Caesar will be combined in the transposition
chiper matrix by applying the pattern of rice cultivation path and for
chipertext retrieval by finally applying the pattern of rice planting path. And
the final digit of the digit shown in the form of binary digits so that 5
characters can be changed to 80 digit bits are scrambled. Thus the cryptanalyst
will be more difficult and takes a very long time to hack information that has
been kept secret.
|
[
{
"created": "Tue, 11 Jul 2017 15:19:33 GMT",
"version": "v1"
}
] |
2017-07-12
|
[
[
"Ginting",
"Dewi Sartika",
""
],
[
"Sitompul",
"Kristin",
""
],
[
"Simanulang",
"Jasael",
""
],
[
"Sembiring",
"Rahmat Widia",
""
],
[
"Zarlis",
"Muhammad",
""
]
] |
Classical cryptography is a way of disguising the news done by the people when there was no computer. The goal is to protect information by way of encoding. This paper describesa modification of classical algorithms to make cryptanalis difficult to steal undisclosed messages. There are three types of classical algorithms that are combined affine chiper, Caesar chiper and chiper transposition. Where for chiperteks affine chiper and Caesar chiper can be looped as much as the initial key, because the result can be varied as much as key value, then affine chiper and Caesar chiper in this case is dynamic. Then the results of the affine and Caesar will be combined in the transposition chiper matrix by applying the pattern of rice cultivation path and for chipertext retrieval by finally applying the pattern of rice planting path. And the final digit of the digit shown in the form of binary digits so that 5 characters can be changed to 80 digit bits are scrambled. Thus the cryptanalyst will be more difficult and takes a very long time to hack information that has been kept secret.
|
2306.02658
|
Etrit Haxholli
|
Etrit Haxholli, Marco Lorenzi
|
Faster Training of Diffusion Models and Improved Density Estimation via
Parallel Score Matching
| null | null | null | null |
cs.LG stat.ML
|
http://creativecommons.org/licenses/by/4.0/
|
In Diffusion Probabilistic Models (DPMs), the task of modeling the score
evolution via a single time-dependent neural network necessitates extended
training periods and may potentially impede modeling flexibility and capacity.
To counteract these challenges, we propose leveraging the independence of
learning tasks at different time points inherent to DPMs. More specifically, we
partition the learning task by utilizing independent networks, each dedicated
to learning the evolution of scores within a specific time sub-interval.
Further, inspired by residual flows, we extend this strategy to its logical
conclusion by employing separate networks to independently model the score at
each individual time point. As empirically demonstrated on synthetic and image
datasets, our approach not only significantly accelerates the training process
by introducing an additional layer of parallelization atop data
parallelization, but it also enhances density estimation performance when
compared to the conventional training methodology for DPMs.
|
[
{
"created": "Mon, 5 Jun 2023 07:47:30 GMT",
"version": "v1"
}
] |
2023-06-06
|
[
[
"Haxholli",
"Etrit",
""
],
[
"Lorenzi",
"Marco",
""
]
] |
In Diffusion Probabilistic Models (DPMs), the task of modeling the score evolution via a single time-dependent neural network necessitates extended training periods and may potentially impede modeling flexibility and capacity. To counteract these challenges, we propose leveraging the independence of learning tasks at different time points inherent to DPMs. More specifically, we partition the learning task by utilizing independent networks, each dedicated to learning the evolution of scores within a specific time sub-interval. Further, inspired by residual flows, we extend this strategy to its logical conclusion by employing separate networks to independently model the score at each individual time point. As empirically demonstrated on synthetic and image datasets, our approach not only significantly accelerates the training process by introducing an additional layer of parallelization atop data parallelization, but it also enhances density estimation performance when compared to the conventional training methodology for DPMs.
|
2307.11335
|
Wenbo Hu
|
Wenbo Hu, Yuling Wang, Lin Ma, Bangbang Yang, Lin Gao, Xiao Liu,
Yuewen Ma
|
Tri-MipRF: Tri-Mip Representation for Efficient Anti-Aliasing Neural
Radiance Fields
|
Accepted to ICCV 2023 Project page:
https://wbhu.github.io/projects/Tri-MipRF
|
ICCV 2023
| null | null |
cs.CV cs.AI cs.GR
|
http://creativecommons.org/licenses/by-nc-sa/4.0/
|
Despite the tremendous progress in neural radiance fields (NeRF), we still
face a dilemma of the trade-off between quality and efficiency, e.g., MipNeRF
presents fine-detailed and anti-aliased renderings but takes days for training,
while Instant-ngp can accomplish the reconstruction in a few minutes but
suffers from blurring or aliasing when rendering at various distances or
resolutions due to ignoring the sampling area. To this end, we propose a novel
Tri-Mip encoding that enables both instant reconstruction and anti-aliased
high-fidelity rendering for neural radiance fields. The key is to factorize the
pre-filtered 3D feature spaces in three orthogonal mipmaps. In this way, we can
efficiently perform 3D area sampling by taking advantage of 2D pre-filtered
feature maps, which significantly elevates the rendering quality without
sacrificing efficiency. To cope with the novel Tri-Mip representation, we
propose a cone-casting rendering technique to efficiently sample anti-aliased
3D features with the Tri-Mip encoding considering both pixel imaging and
observing distance. Extensive experiments on both synthetic and real-world
datasets demonstrate our method achieves state-of-the-art rendering quality and
reconstruction speed while maintaining a compact representation that reduces
25% model size compared against Instant-ngp.
|
[
{
"created": "Fri, 21 Jul 2023 03:47:28 GMT",
"version": "v1"
}
] |
2023-07-24
|
[
[
"Hu",
"Wenbo",
""
],
[
"Wang",
"Yuling",
""
],
[
"Ma",
"Lin",
""
],
[
"Yang",
"Bangbang",
""
],
[
"Gao",
"Lin",
""
],
[
"Liu",
"Xiao",
""
],
[
"Ma",
"Yuewen",
""
]
] |
Despite the tremendous progress in neural radiance fields (NeRF), we still face a dilemma of the trade-off between quality and efficiency, e.g., MipNeRF presents fine-detailed and anti-aliased renderings but takes days for training, while Instant-ngp can accomplish the reconstruction in a few minutes but suffers from blurring or aliasing when rendering at various distances or resolutions due to ignoring the sampling area. To this end, we propose a novel Tri-Mip encoding that enables both instant reconstruction and anti-aliased high-fidelity rendering for neural radiance fields. The key is to factorize the pre-filtered 3D feature spaces in three orthogonal mipmaps. In this way, we can efficiently perform 3D area sampling by taking advantage of 2D pre-filtered feature maps, which significantly elevates the rendering quality without sacrificing efficiency. To cope with the novel Tri-Mip representation, we propose a cone-casting rendering technique to efficiently sample anti-aliased 3D features with the Tri-Mip encoding considering both pixel imaging and observing distance. Extensive experiments on both synthetic and real-world datasets demonstrate our method achieves state-of-the-art rendering quality and reconstruction speed while maintaining a compact representation that reduces 25% model size compared against Instant-ngp.
|
1802.07546
|
Johannes Fauser
|
Johannes Fauser and Georgios Sakas and Anirban Mukhopadhyay
|
Planning Nonlinear Access Paths for Temporal Bone Surgery
|
To be published in International Journal on Computer Assisted
Radiology and Surgery (IJCARS), Spl. Issue IPCAI 2018
| null | null | null |
cs.RO
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
Purpose: Interventions at the otobasis operate in the narrow region of the
temporal bone where several highly sensitive organs define obstacles with
minimal clearance for surgical instruments. Nonlinear trajectories for
potential minimally-invasive interventions can provide larger distances to risk
structures and optimized orientations of surgical instruments, thus improving
clinical outcomes when compared to existing linear approaches. In this paper,
we present fast and accurate planning methods for such nonlinear access paths.
Methods: We define a specific motion planning problem in SE(3) = R3 x SO(3)
with notable constraints in computation time and goal pose that reflect the
requirements of temporal bone surgery.We then present k-RRT-Connect: two
suitable motion planners based on bidirectional Rapidly-exploring Random Trees
(RRT) to solve this problem efficiently. Results: The benefits of k-RRT-Connect
are demonstrated on real CT data of patients. Their general performance is
shown on a large set of realistic synthetic anatomies. We also show that these
new algorithms outperform state of the art methods based on circular arcs or
Bezier-Splines when applied to this specific problem. Conclusion: With this
work we demonstrate that pre- and intra-operative planning of nonlinear access
paths is possible for minimally-invasive surgeries at the otobasis.
|
[
{
"created": "Wed, 21 Feb 2018 12:53:23 GMT",
"version": "v1"
}
] |
2018-02-22
|
[
[
"Fauser",
"Johannes",
""
],
[
"Sakas",
"Georgios",
""
],
[
"Mukhopadhyay",
"Anirban",
""
]
] |
Purpose: Interventions at the otobasis operate in the narrow region of the temporal bone where several highly sensitive organs define obstacles with minimal clearance for surgical instruments. Nonlinear trajectories for potential minimally-invasive interventions can provide larger distances to risk structures and optimized orientations of surgical instruments, thus improving clinical outcomes when compared to existing linear approaches. In this paper, we present fast and accurate planning methods for such nonlinear access paths. Methods: We define a specific motion planning problem in SE(3) = R3 x SO(3) with notable constraints in computation time and goal pose that reflect the requirements of temporal bone surgery.We then present k-RRT-Connect: two suitable motion planners based on bidirectional Rapidly-exploring Random Trees (RRT) to solve this problem efficiently. Results: The benefits of k-RRT-Connect are demonstrated on real CT data of patients. Their general performance is shown on a large set of realistic synthetic anatomies. We also show that these new algorithms outperform state of the art methods based on circular arcs or Bezier-Splines when applied to this specific problem. Conclusion: With this work we demonstrate that pre- and intra-operative planning of nonlinear access paths is possible for minimally-invasive surgeries at the otobasis.
|
2112.02779
|
Kwonyoung Ryu
|
Wei Dong, Kwonyoung Ryu, Michael Kaess, Jaesik Park
|
Revisiting LiDAR Registration and Reconstruction: A Range Image
Perspective
|
14 pages, 9 figures. This paper is under the review
| null | null | null |
cs.CV cs.RO
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
Spinning LiDAR data are prevalent for 3D vision tasks. Since LiDAR data is
presented in the form of point clouds, expensive 3D operations are usually
required. This paper revisits spinning LiDAR scan formation and presents a
cylindrical range image representation with a ray-wise projection/unprojection
model. It is built upon raw scans and supports lossless conversion from 2D to
3D, allowing fast 2D operations, including 2D index-based neighbor search and
downsampling. We then propose, to the best of our knowledge, the first
multi-scale registration and dense signed distance function (SDF)
reconstruction system for LiDAR range images. We further collect a dataset of
indoor and outdoor LiDAR scenes in the posed range image format. A
comprehensive evaluation of registration and reconstruction is conducted on the
proposed dataset and the KITTI dataset. Experiments demonstrate that our
approach outperforms surface reconstruction baselines and achieves similar
performance to state-of-the-art LiDAR registration methods, including a modern
learning-based registration approach. Thanks to the simplicity, our
registration runs at 100Hz and SDF reconstruction in real time. The dataset and
a modularized C++/Python toolbox will be released.
|
[
{
"created": "Mon, 6 Dec 2021 04:28:32 GMT",
"version": "v1"
},
{
"created": "Mon, 28 Mar 2022 22:38:28 GMT",
"version": "v2"
}
] |
2022-03-30
|
[
[
"Dong",
"Wei",
""
],
[
"Ryu",
"Kwonyoung",
""
],
[
"Kaess",
"Michael",
""
],
[
"Park",
"Jaesik",
""
]
] |
Spinning LiDAR data are prevalent for 3D vision tasks. Since LiDAR data is presented in the form of point clouds, expensive 3D operations are usually required. This paper revisits spinning LiDAR scan formation and presents a cylindrical range image representation with a ray-wise projection/unprojection model. It is built upon raw scans and supports lossless conversion from 2D to 3D, allowing fast 2D operations, including 2D index-based neighbor search and downsampling. We then propose, to the best of our knowledge, the first multi-scale registration and dense signed distance function (SDF) reconstruction system for LiDAR range images. We further collect a dataset of indoor and outdoor LiDAR scenes in the posed range image format. A comprehensive evaluation of registration and reconstruction is conducted on the proposed dataset and the KITTI dataset. Experiments demonstrate that our approach outperforms surface reconstruction baselines and achieves similar performance to state-of-the-art LiDAR registration methods, including a modern learning-based registration approach. Thanks to the simplicity, our registration runs at 100Hz and SDF reconstruction in real time. The dataset and a modularized C++/Python toolbox will be released.
|
2104.11568
|
Lorin Sweeney
|
Lorin Sweeney, Graham Healy, Alan F. Smeaton
|
The Influence of Audio on Video Memorability with an Audio Gestalt
Regulated Video Memorability System
|
6 pages, 3 figures, 4 tables, paper accepted in CBMI 2021 for
publication and oral presentation
| null |
10.1109/CBMI50038.2021.9461903
| null |
cs.MM cs.CV cs.LG
|
http://creativecommons.org/licenses/by/4.0/
|
Memories are the tethering threads that tie us to the world, and memorability
is the measure of their tensile strength. The threads of memory are spun from
fibres of many modalities, obscuring the contribution of a single fibre to a
thread's overall tensile strength. Unfurling these fibres is the key to
understanding the nature of their interaction, and how we can ultimately create
more meaningful media content. In this paper, we examine the influence of audio
on video recognition memorability, finding evidence to suggest that it can
facilitate overall video recognition memorability rich in high-level (gestalt)
audio features. We introduce a novel multimodal deep learning-based late-fusion
system that uses audio gestalt to estimate the influence of a given video's
audio on its overall short-term recognition memorability, and selectively
leverages audio features to make a prediction accordingly. We benchmark our
audio gestalt based system on the Memento10k short-term video memorability
dataset, achieving top-2 state-of-the-art results.
|
[
{
"created": "Fri, 23 Apr 2021 12:53:33 GMT",
"version": "v1"
}
] |
2021-07-02
|
[
[
"Sweeney",
"Lorin",
""
],
[
"Healy",
"Graham",
""
],
[
"Smeaton",
"Alan F.",
""
]
] |
Memories are the tethering threads that tie us to the world, and memorability is the measure of their tensile strength. The threads of memory are spun from fibres of many modalities, obscuring the contribution of a single fibre to a thread's overall tensile strength. Unfurling these fibres is the key to understanding the nature of their interaction, and how we can ultimately create more meaningful media content. In this paper, we examine the influence of audio on video recognition memorability, finding evidence to suggest that it can facilitate overall video recognition memorability rich in high-level (gestalt) audio features. We introduce a novel multimodal deep learning-based late-fusion system that uses audio gestalt to estimate the influence of a given video's audio on its overall short-term recognition memorability, and selectively leverages audio features to make a prediction accordingly. We benchmark our audio gestalt based system on the Memento10k short-term video memorability dataset, achieving top-2 state-of-the-art results.
|
2105.01011
|
Peng Liang
|
Liming Fu, Peng Liang, Xueying Li, Chen Yang
|
A Machine Learning Based Ensemble Method for Automatic Multiclass
Classification of Decisions
|
The 25th International Conference on Evaluation and Assessment in
Software Engineering (EASE)
| null |
10.1145/3463274.3463325
| null |
cs.SE
|
http://creativecommons.org/licenses/by/4.0/
|
Stakeholders make various types of decisions with respect to requirements,
design, management, and so on during the software development life cycle.
Nevertheless, these decisions are typically not well documented and classified
due to limited human resources, time, and budget. To this end, automatic
approaches provide a promising way. In this paper, we aimed at automatically
classifying decisions into five types to help stakeholders better document and
understand decisions. First, we collected a dataset from the Hibernate
developer mailing list. We then experimented and evaluated 270 configurations
regarding feature selection, feature extraction techniques, and machine
learning classifiers to seek the best configuration for classifying decisions.
Especially, we applied an ensemble learning method and constructed ensemble
classifiers to compare the performance between ensemble classifiers and base
classifiers. Our experiment results show that (1) feature selection can
decently improve the classification results; (2) ensemble classifiers can
outperform base classifiers provided that ensemble classifiers are well
constructed; (3) BoW + 50% features selected by feature selection with an
ensemble classifier that combines Na\"ive Bayes (NB), Logistic Regression (LR),
and Support Vector Machine (SVM) achieves the best classification result (with
a weighted precision of 0.750, a weighted recall of 0.739, and a weighted
F1-score of 0.727) among all the configurations. Our work can benefit various
types of stakeholders in software development through providing an automatic
approach for effectively classifying decisions into specific types that are
relevant to their interests.
|
[
{
"created": "Mon, 3 May 2021 16:55:00 GMT",
"version": "v1"
},
{
"created": "Tue, 4 May 2021 04:21:23 GMT",
"version": "v2"
}
] |
2021-05-05
|
[
[
"Fu",
"Liming",
""
],
[
"Liang",
"Peng",
""
],
[
"Li",
"Xueying",
""
],
[
"Yang",
"Chen",
""
]
] |
Stakeholders make various types of decisions with respect to requirements, design, management, and so on during the software development life cycle. Nevertheless, these decisions are typically not well documented and classified due to limited human resources, time, and budget. To this end, automatic approaches provide a promising way. In this paper, we aimed at automatically classifying decisions into five types to help stakeholders better document and understand decisions. First, we collected a dataset from the Hibernate developer mailing list. We then experimented and evaluated 270 configurations regarding feature selection, feature extraction techniques, and machine learning classifiers to seek the best configuration for classifying decisions. Especially, we applied an ensemble learning method and constructed ensemble classifiers to compare the performance between ensemble classifiers and base classifiers. Our experiment results show that (1) feature selection can decently improve the classification results; (2) ensemble classifiers can outperform base classifiers provided that ensemble classifiers are well constructed; (3) BoW + 50% features selected by feature selection with an ensemble classifier that combines Na\"ive Bayes (NB), Logistic Regression (LR), and Support Vector Machine (SVM) achieves the best classification result (with a weighted precision of 0.750, a weighted recall of 0.739, and a weighted F1-score of 0.727) among all the configurations. Our work can benefit various types of stakeholders in software development through providing an automatic approach for effectively classifying decisions into specific types that are relevant to their interests.
|
2407.04272
|
Dingwen Tao
|
Hao Feng, Boyuan Zhang, Fanjiang Ye, Min Si, Ching-Hsiang Chu, Jiannan
Tian, Chunxing Yin, Summer Deng, Yuchen Hao, Pavan Balaji, Tong Geng, Dingwen
Tao
|
Accelerating Communication in Deep Learning Recommendation Model
Training with Dual-Level Adaptive Lossy Compression
|
accepted by SC '24
| null | null | null |
cs.LG cs.DC
|
http://creativecommons.org/licenses/by/4.0/
|
DLRM is a state-of-the-art recommendation system model that has gained
widespread adoption across various industry applications. The large size of
DLRM models, however, necessitates the use of multiple devices/GPUs for
efficient training. A significant bottleneck in this process is the
time-consuming all-to-all communication required to collect embedding data from
all devices. To mitigate this, we introduce a method that employs error-bounded
lossy compression to reduce the communication data size and accelerate DLRM
training. We develop a novel error-bounded lossy compression algorithm,
informed by an in-depth analysis of embedding data features, to achieve high
compression ratios. Moreover, we introduce a dual-level adaptive strategy for
error-bound adjustment, spanning both table-wise and iteration-wise aspects, to
balance the compression benefits with the potential impacts on accuracy. We
further optimize our compressor for PyTorch tensors on GPUs, minimizing
compression overhead. Evaluation shows that our method achieves a 1.38$\times$
training speedup with a minimal accuracy impact.
|
[
{
"created": "Fri, 5 Jul 2024 05:55:18 GMT",
"version": "v1"
},
{
"created": "Mon, 8 Jul 2024 05:53:10 GMT",
"version": "v2"
},
{
"created": "Thu, 11 Jul 2024 15:31:53 GMT",
"version": "v3"
}
] |
2024-07-12
|
[
[
"Feng",
"Hao",
""
],
[
"Zhang",
"Boyuan",
""
],
[
"Ye",
"Fanjiang",
""
],
[
"Si",
"Min",
""
],
[
"Chu",
"Ching-Hsiang",
""
],
[
"Tian",
"Jiannan",
""
],
[
"Yin",
"Chunxing",
""
],
[
"Deng",
"Summer",
""
],
[
"Hao",
"Yuchen",
""
],
[
"Balaji",
"Pavan",
""
],
[
"Geng",
"Tong",
""
],
[
"Tao",
"Dingwen",
""
]
] |
DLRM is a state-of-the-art recommendation system model that has gained widespread adoption across various industry applications. The large size of DLRM models, however, necessitates the use of multiple devices/GPUs for efficient training. A significant bottleneck in this process is the time-consuming all-to-all communication required to collect embedding data from all devices. To mitigate this, we introduce a method that employs error-bounded lossy compression to reduce the communication data size and accelerate DLRM training. We develop a novel error-bounded lossy compression algorithm, informed by an in-depth analysis of embedding data features, to achieve high compression ratios. Moreover, we introduce a dual-level adaptive strategy for error-bound adjustment, spanning both table-wise and iteration-wise aspects, to balance the compression benefits with the potential impacts on accuracy. We further optimize our compressor for PyTorch tensors on GPUs, minimizing compression overhead. Evaluation shows that our method achieves a 1.38$\times$ training speedup with a minimal accuracy impact.
|
2403.02234
|
Fangzhou Hong
|
Fangzhou Hong, Jiaxiang Tang, Ziang Cao, Min Shi, Tong Wu, Zhaoxi
Chen, Shuai Yang, Tengfei Wang, Liang Pan, Dahua Lin, Ziwei Liu
|
3DTopia: Large Text-to-3D Generation Model with Hybrid Diffusion Priors
|
Code available at https://github.com/3DTopia/3DTopia
| null | null | null |
cs.CV
|
http://creativecommons.org/licenses/by-nc-sa/4.0/
|
We present a two-stage text-to-3D generation system, namely 3DTopia, which
generates high-quality general 3D assets within 5 minutes using hybrid
diffusion priors. The first stage samples from a 3D diffusion prior directly
learned from 3D data. Specifically, it is powered by a text-conditioned
tri-plane latent diffusion model, which quickly generates coarse 3D samples for
fast prototyping. The second stage utilizes 2D diffusion priors to further
refine the texture of coarse 3D models from the first stage. The refinement
consists of both latent and pixel space optimization for high-quality texture
generation. To facilitate the training of the proposed system, we clean and
caption the largest open-source 3D dataset, Objaverse, by combining the power
of vision language models and large language models. Experiment results are
reported qualitatively and quantitatively to show the performance of the
proposed system. Our codes and models are available at
https://github.com/3DTopia/3DTopia
|
[
{
"created": "Mon, 4 Mar 2024 17:26:28 GMT",
"version": "v1"
},
{
"created": "Tue, 7 May 2024 03:25:50 GMT",
"version": "v2"
}
] |
2024-05-08
|
[
[
"Hong",
"Fangzhou",
""
],
[
"Tang",
"Jiaxiang",
""
],
[
"Cao",
"Ziang",
""
],
[
"Shi",
"Min",
""
],
[
"Wu",
"Tong",
""
],
[
"Chen",
"Zhaoxi",
""
],
[
"Yang",
"Shuai",
""
],
[
"Wang",
"Tengfei",
""
],
[
"Pan",
"Liang",
""
],
[
"Lin",
"Dahua",
""
],
[
"Liu",
"Ziwei",
""
]
] |
We present a two-stage text-to-3D generation system, namely 3DTopia, which generates high-quality general 3D assets within 5 minutes using hybrid diffusion priors. The first stage samples from a 3D diffusion prior directly learned from 3D data. Specifically, it is powered by a text-conditioned tri-plane latent diffusion model, which quickly generates coarse 3D samples for fast prototyping. The second stage utilizes 2D diffusion priors to further refine the texture of coarse 3D models from the first stage. The refinement consists of both latent and pixel space optimization for high-quality texture generation. To facilitate the training of the proposed system, we clean and caption the largest open-source 3D dataset, Objaverse, by combining the power of vision language models and large language models. Experiment results are reported qualitatively and quantitatively to show the performance of the proposed system. Our codes and models are available at https://github.com/3DTopia/3DTopia
|
1510.05182
|
Fereydoun Farrahi Moghaddam
|
Fereydoun Farrahi Moghaddam and Mohamed Cheriet
|
Sustainability-Aware Cloud Computing Using Virtual Carbon Tax
| null | null | null | null |
cs.DC
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
In this paper, a solution for sustainable cloud system is proposed and then
implemented on a real testbed. The solution composes of optimization of a
profit model and introduction of virtual carbon tax to limit environmental
footprint of the cloud. The proposed multi-criteria optimizer of the cloud
system suggests new optimum CPU frequencies for CPU-cores when the local grid
energy mix or the cloud workload changes. The cloud system is implemented on a
blade system, and proper middlewares are developed to interact with the blades.
The experimental results show that it is possible to significantly decrease the
targeted environmental footprint of the system and keep it profitable.
|
[
{
"created": "Sat, 17 Oct 2015 23:33:20 GMT",
"version": "v1"
},
{
"created": "Wed, 1 Nov 2017 12:55:42 GMT",
"version": "v2"
}
] |
2017-11-02
|
[
[
"Moghaddam",
"Fereydoun Farrahi",
""
],
[
"Cheriet",
"Mohamed",
""
]
] |
In this paper, a solution for sustainable cloud system is proposed and then implemented on a real testbed. The solution composes of optimization of a profit model and introduction of virtual carbon tax to limit environmental footprint of the cloud. The proposed multi-criteria optimizer of the cloud system suggests new optimum CPU frequencies for CPU-cores when the local grid energy mix or the cloud workload changes. The cloud system is implemented on a blade system, and proper middlewares are developed to interact with the blades. The experimental results show that it is possible to significantly decrease the targeted environmental footprint of the system and keep it profitable.
|
1705.08568
|
Grant Storey
|
Grant Storey, Dillon Reisman, Jonathan Mayer, Arvind Narayanan
|
The Future of Ad Blocking: An Analytical Framework and New Techniques
| null | null | null | null |
cs.CR
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
We present a systematic study of ad blocking - and the associated "arms race"
- as a security problem. We model ad blocking as a state space with four states
and six state transitions, which correspond to techniques that can be deployed
by either publishers or ad blockers. We argue that this is a complete model of
the system. We propose several new ad blocking techniques, including ones that
borrow ideas from rootkits to prevent detection by anti-ad blocking scripts.
Another technique uses the insight that ads must be recognizable by humans to
comply with laws and industry self-regulation. We have built prototype
implementations of three of these techniques, successfully blocking ads and
evading detection. We systematically evaluate our proposed techniques, along
with existing ones, in terms of security, practicality, and legality. We
characterize the order of growth of the development effort required to
create/maintain ad blockers as a function of the growth of the web. Based on
our state-space model, our new techniques, and this systematization, we offer
insights into the likely "end game" of the arms race. We challenge the
widespread assumption that the arms race will escalate indefinitely, and
instead identify a combination of evolving technical and legal factors that
will determine the outcome.
|
[
{
"created": "Wed, 24 May 2017 00:28:51 GMT",
"version": "v1"
}
] |
2017-05-25
|
[
[
"Storey",
"Grant",
""
],
[
"Reisman",
"Dillon",
""
],
[
"Mayer",
"Jonathan",
""
],
[
"Narayanan",
"Arvind",
""
]
] |
We present a systematic study of ad blocking - and the associated "arms race" - as a security problem. We model ad blocking as a state space with four states and six state transitions, which correspond to techniques that can be deployed by either publishers or ad blockers. We argue that this is a complete model of the system. We propose several new ad blocking techniques, including ones that borrow ideas from rootkits to prevent detection by anti-ad blocking scripts. Another technique uses the insight that ads must be recognizable by humans to comply with laws and industry self-regulation. We have built prototype implementations of three of these techniques, successfully blocking ads and evading detection. We systematically evaluate our proposed techniques, along with existing ones, in terms of security, practicality, and legality. We characterize the order of growth of the development effort required to create/maintain ad blockers as a function of the growth of the web. Based on our state-space model, our new techniques, and this systematization, we offer insights into the likely "end game" of the arms race. We challenge the widespread assumption that the arms race will escalate indefinitely, and instead identify a combination of evolving technical and legal factors that will determine the outcome.
|
1603.05623
|
Dimitri Van De Ville
|
Dimitri Van De Ville
|
Steering Macro-Scale Network Community Structure by Micro-Scale Features
|
15 pages, 7 figures
| null | null | null |
cs.SI cs.CE cs.DM
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
Network science plays an increasingly important role to model complex data in
many scientific disciplines. One notable feature of network organization is
community structure, which refers to clusters of tightly interconnected nodes.
A prominent problem is how to investigate the relationship between macro-scale
modules that are retrieved by optimizing global network measures, and
micro-scale structure that are defined by specific queries of the analysis
(e.g., nodal features). By generalizing fundamental concepts of joint
space-frequency localization to network theory, here we propose a flexible
framework to study interactions between micro- and macro-structure. Similar to
pointing and focusing a magnifying glass, the analysis can be directed to
specific micro-scale structure, while the degree of interaction with the
macro-scale community structure can be seamlessly controlled. In addition, the
method is computationally efficient as a result of the underlying
low-dimensional optimization problem.
|
[
{
"created": "Thu, 17 Mar 2016 19:11:46 GMT",
"version": "v1"
}
] |
2016-03-18
|
[
[
"Van De Ville",
"Dimitri",
""
]
] |
Network science plays an increasingly important role to model complex data in many scientific disciplines. One notable feature of network organization is community structure, which refers to clusters of tightly interconnected nodes. A prominent problem is how to investigate the relationship between macro-scale modules that are retrieved by optimizing global network measures, and micro-scale structure that are defined by specific queries of the analysis (e.g., nodal features). By generalizing fundamental concepts of joint space-frequency localization to network theory, here we propose a flexible framework to study interactions between micro- and macro-structure. Similar to pointing and focusing a magnifying glass, the analysis can be directed to specific micro-scale structure, while the degree of interaction with the macro-scale community structure can be seamlessly controlled. In addition, the method is computationally efficient as a result of the underlying low-dimensional optimization problem.
|
1710.08338
|
Mehdi Samiee
|
M. Samiee, M. Zayernouri. Mark M. Meerschaert
|
A Unified Spectral Method for FPDEs with Two-sided Derivatives; A Fast
Solver
| null |
https://doi.org/10.1016/j.jcp.2018.02.014
|
10.1016/j.jcp.2018.02.014
| null |
cs.CE math.NA
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
We develop a unified Petrov-Galerkin spectral method for a class of
fractional partial differential equations with two-sided derivatives and
constant coefficients of the form $ _{0}{\mathcal{D}}_{t}^{2\tau}u^{} +
\sum_{i=1}^{d}$ $[c_{l_i}$ $_{a_i}{\mathcal{D}}_{x_i}^{2\mu_i} u^{} +c_{r_i}$
$_{x_i}{\mathcal{D}}_{b_i}^{2\mu_i}$ $u^{} ] +$ $\gamma$ $u^{} = \sum_{j=1}^{d}
[ \kappa_{l_j}$ $_{a_j}{\mathcal{D}}_{x_j}^{2\nu_j} u^{}$ $+\kappa_{r_j}$
$_{x_j}{\mathcal{D}}_{b_j}^{2\nu_j}$ $u^{} ]$ $+ f$, where $2\tau \in (0,2)$,
$2\mu_i \in (0,1)$ and $2\nu_j \in (1,2)$, in a ($1+d$)-dimensional
\textit{space-time} hypercube, $d = 1, 2, 3, \cdots$, subject to homogeneous
Dirichlet initial/boundary conditions. We employ the eigenfunctions of the
fractional Sturm-Liouville eigen-problems of the first kind in
\cite{zayernouri2013fractional}, called \textit{Jacobi poly-fractonomial}s, as
temporal bases, and the eigen-functions of the boundary-value problem of the
second kind as temporal test functions. Next, we construct our spatial
basis/test functions using Legendre polynomials, yielding mass matrices being
independent of the spatial fractional orders ($\mu_i, \, \nu_j, \, i,
\,j=1,2,\cdots,d$). Furthermore, we formulate a novel unified fast linear
solver for the resulting high-dimensional linear system based on the solution
of generalized eigen-problem of spatial mass matrices with respect to the
corresponding stiffness matrices, hence, making the complexity of the problem
optimal, i.e., $\mathcal{O}(N^{d+2})$. We carry out several numerical test
cases to examine the CPU time and convergence rate of the method. The
corresponding stability and error analysis of the Petrov-Galerkin method are
carried out in \cite{samiee2016Unified2}.
|
[
{
"created": "Sun, 15 Oct 2017 20:49:56 GMT",
"version": "v1"
}
] |
2019-10-02
|
[
[
"Samiee",
"M.",
""
],
[
"Meerschaert",
"M. Zayernouri. Mark M.",
""
]
] |
We develop a unified Petrov-Galerkin spectral method for a class of fractional partial differential equations with two-sided derivatives and constant coefficients of the form $ _{0}{\mathcal{D}}_{t}^{2\tau}u^{} + \sum_{i=1}^{d}$ $[c_{l_i}$ $_{a_i}{\mathcal{D}}_{x_i}^{2\mu_i} u^{} +c_{r_i}$ $_{x_i}{\mathcal{D}}_{b_i}^{2\mu_i}$ $u^{} ] +$ $\gamma$ $u^{} = \sum_{j=1}^{d} [ \kappa_{l_j}$ $_{a_j}{\mathcal{D}}_{x_j}^{2\nu_j} u^{}$ $+\kappa_{r_j}$ $_{x_j}{\mathcal{D}}_{b_j}^{2\nu_j}$ $u^{} ]$ $+ f$, where $2\tau \in (0,2)$, $2\mu_i \in (0,1)$ and $2\nu_j \in (1,2)$, in a ($1+d$)-dimensional \textit{space-time} hypercube, $d = 1, 2, 3, \cdots$, subject to homogeneous Dirichlet initial/boundary conditions. We employ the eigenfunctions of the fractional Sturm-Liouville eigen-problems of the first kind in \cite{zayernouri2013fractional}, called \textit{Jacobi poly-fractonomial}s, as temporal bases, and the eigen-functions of the boundary-value problem of the second kind as temporal test functions. Next, we construct our spatial basis/test functions using Legendre polynomials, yielding mass matrices being independent of the spatial fractional orders ($\mu_i, \, \nu_j, \, i, \,j=1,2,\cdots,d$). Furthermore, we formulate a novel unified fast linear solver for the resulting high-dimensional linear system based on the solution of generalized eigen-problem of spatial mass matrices with respect to the corresponding stiffness matrices, hence, making the complexity of the problem optimal, i.e., $\mathcal{O}(N^{d+2})$. We carry out several numerical test cases to examine the CPU time and convergence rate of the method. The corresponding stability and error analysis of the Petrov-Galerkin method are carried out in \cite{samiee2016Unified2}.
|
2201.07060
|
Sourav Mondal
|
Sourav Mondal and Marco Ruffini
|
A Min-Max Fair Resource Allocation Framework for Optical x-haul and
DU/CU in Multi-tenant O-RANs
|
This article is accepted for publication in IEEE International
Conference on Communications (ICC) 2022. Copyright @ IEEE
| null | null | null |
cs.NI
|
http://creativecommons.org/licenses/by-nc-sa/4.0/
|
The recently proposed open-radio access network (O-RAN) architecture embraces
cloudification and network function virtualization techniques to perform the
base-band function processing by dis-aggregated radio units (RUs), distributed
units (DUs), and centralized units (CUs). This enables the cloud-RAN vision in
full, where mobile network operators (MNOs) could install their own RUs, but
then lease on-demand computational resources for the processing of DU and CU
functions from commonly available open-cloud (O-Cloud) servers via open x-haul
interfaces due to variation of load over the day. This creates a multi-tenant
scenario where multiple MNOs share networking as well as computational
resources. In this paper, we propose a framework that dynamically allocates
x-haul and DU/CU resources in a multi-tenant O-RAN ecosystem with min-max
fairness guarantees. This framework ensures that a maximum number of RUs get
sufficient resources while minimizing the OPEX for their MNOs. Moreover, in
order to provide an access network architecture capable of sustaining
low-latency and high capacity between RUs and edge-computing devices, we
consider time-wavelength division multiplexed (TWDM) passive optical network
(PON)-based x-haul interfaces where the PON virtualization technique is used to
provide a direct optical connection between end-points. This creates a virtual
mesh interconnection among all the nodes such that the RUs can be connected to
the Edge-Clouds at macro-cell RU locations as well as to the O-Cloud servers at
the central office locations. Furthermore, we analyze the system performance
with our proposed framework and show that MNOs can operate with a better
cost-efficiency than baseline greedy resource allocation with uniform
cost-sharing.
|
[
{
"created": "Tue, 18 Jan 2022 15:38:16 GMT",
"version": "v1"
},
{
"created": "Thu, 20 Jan 2022 18:29:09 GMT",
"version": "v2"
},
{
"created": "Tue, 25 Jan 2022 03:56:32 GMT",
"version": "v3"
},
{
"created": "Tue, 22 Feb 2022 15:22:59 GMT",
"version": "v4"
}
] |
2022-02-23
|
[
[
"Mondal",
"Sourav",
""
],
[
"Ruffini",
"Marco",
""
]
] |
The recently proposed open-radio access network (O-RAN) architecture embraces cloudification and network function virtualization techniques to perform the base-band function processing by dis-aggregated radio units (RUs), distributed units (DUs), and centralized units (CUs). This enables the cloud-RAN vision in full, where mobile network operators (MNOs) could install their own RUs, but then lease on-demand computational resources for the processing of DU and CU functions from commonly available open-cloud (O-Cloud) servers via open x-haul interfaces due to variation of load over the day. This creates a multi-tenant scenario where multiple MNOs share networking as well as computational resources. In this paper, we propose a framework that dynamically allocates x-haul and DU/CU resources in a multi-tenant O-RAN ecosystem with min-max fairness guarantees. This framework ensures that a maximum number of RUs get sufficient resources while minimizing the OPEX for their MNOs. Moreover, in order to provide an access network architecture capable of sustaining low-latency and high capacity between RUs and edge-computing devices, we consider time-wavelength division multiplexed (TWDM) passive optical network (PON)-based x-haul interfaces where the PON virtualization technique is used to provide a direct optical connection between end-points. This creates a virtual mesh interconnection among all the nodes such that the RUs can be connected to the Edge-Clouds at macro-cell RU locations as well as to the O-Cloud servers at the central office locations. Furthermore, we analyze the system performance with our proposed framework and show that MNOs can operate with a better cost-efficiency than baseline greedy resource allocation with uniform cost-sharing.
|
1905.11924
|
Ari Kobren
|
Ari Kobren, Barna Saha, Andrew McCallum
|
Paper Matching with Local Fairness Constraints
|
Appears at KDD 2019 Research Track, 20 pages
| null | null | null |
cs.DS cs.DL
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
Automatically matching reviewers to papers is a crucial step of the peer
review process for venues receiving thousands of submissions. Unfortunately,
common paper matching algorithms often construct matchings suffering from two
critical problems: (1) the group of reviewers assigned to a paper do not
collectively possess sufficient expertise, and (2) reviewer workloads are
highly skewed. In this paper, we propose a novel local fairness formulation of
paper matching that directly addresses both of these issues. Since optimizing
our formulation is not always tractable, we introduce two new algorithms,
FairIR and FairFlow, for computing fair matchings that approximately optimize
the new formulation. FairIR solves a relaxation of the local fairness
formulation and then employs a rounding technique to construct a valid matching
that provably maximizes the objective and only compromises on fairness with
respect to reviewer loads and papers by a small constant. In contrast, FairFlow
is not provably guaranteed to produce fair matchings, however it can be 2x as
efficient as FairIR and an order of magnitude faster than matching algorithms
that directly optimize for fairness. Empirically, we demonstrate that both
FairIR and FairFlow improve fairness over standard matching algorithms on real
conference data. Moreover, in comparison to state-of-the-art matching
algorithms that optimize for fairness only, FairIR achieves higher objective
scores, FairFlow achieves competitive fairness, and both are capable of more
evenly allocating reviewers.
|
[
{
"created": "Tue, 28 May 2019 16:36:51 GMT",
"version": "v1"
}
] |
2019-05-29
|
[
[
"Kobren",
"Ari",
""
],
[
"Saha",
"Barna",
""
],
[
"McCallum",
"Andrew",
""
]
] |
Automatically matching reviewers to papers is a crucial step of the peer review process for venues receiving thousands of submissions. Unfortunately, common paper matching algorithms often construct matchings suffering from two critical problems: (1) the group of reviewers assigned to a paper do not collectively possess sufficient expertise, and (2) reviewer workloads are highly skewed. In this paper, we propose a novel local fairness formulation of paper matching that directly addresses both of these issues. Since optimizing our formulation is not always tractable, we introduce two new algorithms, FairIR and FairFlow, for computing fair matchings that approximately optimize the new formulation. FairIR solves a relaxation of the local fairness formulation and then employs a rounding technique to construct a valid matching that provably maximizes the objective and only compromises on fairness with respect to reviewer loads and papers by a small constant. In contrast, FairFlow is not provably guaranteed to produce fair matchings, however it can be 2x as efficient as FairIR and an order of magnitude faster than matching algorithms that directly optimize for fairness. Empirically, we demonstrate that both FairIR and FairFlow improve fairness over standard matching algorithms on real conference data. Moreover, in comparison to state-of-the-art matching algorithms that optimize for fairness only, FairIR achieves higher objective scores, FairFlow achieves competitive fairness, and both are capable of more evenly allocating reviewers.
|
2406.03897
|
Tzuf Paz-Argaman
|
Tzuf Paz-Argaman, Itai Mondshine, Asaf Achi Mordechai, and Reut
Tsarfaty
|
HeSum: a Novel Dataset for Abstractive Text Summarization in Hebrew
| null |
ACL 2024 Findings
| null | null |
cs.CL cs.AI
|
http://creativecommons.org/licenses/by/4.0/
|
While large language models (LLMs) excel in various natural language tasks in
English, their performance in lower-resourced languages like Hebrew, especially
for generative tasks such as abstractive summarization, remains unclear. The
high morphological richness in Hebrew adds further challenges due to the
ambiguity in sentence comprehension and the complexities in meaning
construction. In this paper, we address this resource and evaluation gap by
introducing HeSum, a novel benchmark specifically designed for abstractive text
summarization in Modern Hebrew. HeSum consists of 10,000 article-summary pairs
sourced from Hebrew news websites written by professionals. Linguistic analysis
confirms HeSum's high abstractness and unique morphological challenges. We show
that HeSum presents distinct difficulties for contemporary state-of-the-art
LLMs, establishing it as a valuable testbed for generative language technology
in Hebrew, and MRLs generative challenges in general.
|
[
{
"created": "Thu, 6 Jun 2024 09:36:14 GMT",
"version": "v1"
},
{
"created": "Mon, 10 Jun 2024 05:45:25 GMT",
"version": "v2"
}
] |
2024-06-11
|
[
[
"Paz-Argaman",
"Tzuf",
""
],
[
"Mondshine",
"Itai",
""
],
[
"Mordechai",
"Asaf Achi",
""
],
[
"Tsarfaty",
"Reut",
""
]
] |
While large language models (LLMs) excel in various natural language tasks in English, their performance in lower-resourced languages like Hebrew, especially for generative tasks such as abstractive summarization, remains unclear. The high morphological richness in Hebrew adds further challenges due to the ambiguity in sentence comprehension and the complexities in meaning construction. In this paper, we address this resource and evaluation gap by introducing HeSum, a novel benchmark specifically designed for abstractive text summarization in Modern Hebrew. HeSum consists of 10,000 article-summary pairs sourced from Hebrew news websites written by professionals. Linguistic analysis confirms HeSum's high abstractness and unique morphological challenges. We show that HeSum presents distinct difficulties for contemporary state-of-the-art LLMs, establishing it as a valuable testbed for generative language technology in Hebrew, and MRLs generative challenges in general.
|
2009.00530
|
Duy Phan Mr
|
Phan The Duy, Do Thi Thu Hien, Van-Hau Pham
|
A survey on Blockchain-based applications for reforming data protection,
privacy and security
|
8 pages, 2 figures
| null | null | null |
cs.CR
|
http://creativecommons.org/licenses/by/4.0/
|
The modern society, economy and industry have been changed remarkably by many
cutting-edge technologies over the last years, and many more are in development
and early implementation that will in turn led even wider spread of adoptions
and greater alteration. Blockchain technology along with other rising ones is
expected to transform virtually every aspect of global business and
individuals' lifestyle in some areas. It has been spreading with multi-sector
applications from financial services to healthcare, supply chain, and
cybersecurity emerging every passing day. Simultaneously, in the digital world,
data protection and privacy are the most enormous issues which customers,
companies and policymakers also take seriously into consideration due to the
recent increase of security breaches and surveillance in reported incidents. In
this case, blockchain has the capability and potential to revolutionize trust,
security and privacy of individual data in the online world. Hence, the purpose
of this paper is to study the actual cases of Blockchain applied in the
reformation of privacy and security field by discussing its impacts as well as
the opportunities and challenges.
|
[
{
"created": "Tue, 1 Sep 2020 16:04:57 GMT",
"version": "v1"
}
] |
2020-09-02
|
[
[
"Duy",
"Phan The",
""
],
[
"Hien",
"Do Thi Thu",
""
],
[
"Pham",
"Van-Hau",
""
]
] |
The modern society, economy and industry have been changed remarkably by many cutting-edge technologies over the last years, and many more are in development and early implementation that will in turn led even wider spread of adoptions and greater alteration. Blockchain technology along with other rising ones is expected to transform virtually every aspect of global business and individuals' lifestyle in some areas. It has been spreading with multi-sector applications from financial services to healthcare, supply chain, and cybersecurity emerging every passing day. Simultaneously, in the digital world, data protection and privacy are the most enormous issues which customers, companies and policymakers also take seriously into consideration due to the recent increase of security breaches and surveillance in reported incidents. In this case, blockchain has the capability and potential to revolutionize trust, security and privacy of individual data in the online world. Hence, the purpose of this paper is to study the actual cases of Blockchain applied in the reformation of privacy and security field by discussing its impacts as well as the opportunities and challenges.
|
2006.12645
|
Vinod Grover
|
Somashekaracharya G. Bhaskaracharya, Julien Demouth, Vinod Grover
|
Automatic Kernel Generation for Volta Tensor Cores
| null | null | null | null |
cs.PL
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
A commonly occurring computation idiom in neural networks is to perform some
pointwise operations on the result of a matrix multiplication. Such a sequence
of operations is typically represented as a computation graph in deep learning
compilers. When compiling to a GPU target, these computations can be
individually mapped to manually tuned implementations provided by libraries
such as cuBLAS and cuDNN. These libraries also provide off-the-shelf support
for targeting tensor cores in NVIDIA GPUs, which can lead to huge performance
boosts through their specialized support for mixed-precision matrix math.
Alternatively, tensor cores can be programmed directly using CUDA APIs or
inline assembly instructions, which opens up the possibility of generating
efficient CUDA kernels automatically for such computations.
Automatic kernel generation is particularly crucial when it is beneficial to
generate efficient code for an entire computation graph by fusing several
operations into a single device function instead of invoking a separate kernel
for each of them. Polyhedral compilation techniques provide a systematic
approach for the analysis and transformation of a sequence of affine
loop-nests. In this paper, we describe a polyhedral approach to generate
efficient CUDA kernels for matrix multiplication using inline assembly
instructions for programming tensor cores on NVIDIA Volta GPUs. Furthermore, we
build on this approach to generate fused kernels for computation sequences
involving matrix multiplication and pointwise operations such as bias addition,
ReLU activation etc. Experimental evaluation of these techniques show that
automatically generated kernels can provide significantly better performance
than manually tuned library implementations, with speedups ranging up to 2.55X.
|
[
{
"created": "Mon, 22 Jun 2020 22:16:00 GMT",
"version": "v1"
},
{
"created": "Mon, 29 Jun 2020 22:20:24 GMT",
"version": "v2"
},
{
"created": "Sat, 1 Aug 2020 21:41:41 GMT",
"version": "v3"
}
] |
2020-08-04
|
[
[
"Bhaskaracharya",
"Somashekaracharya G.",
""
],
[
"Demouth",
"Julien",
""
],
[
"Grover",
"Vinod",
""
]
] |
A commonly occurring computation idiom in neural networks is to perform some pointwise operations on the result of a matrix multiplication. Such a sequence of operations is typically represented as a computation graph in deep learning compilers. When compiling to a GPU target, these computations can be individually mapped to manually tuned implementations provided by libraries such as cuBLAS and cuDNN. These libraries also provide off-the-shelf support for targeting tensor cores in NVIDIA GPUs, which can lead to huge performance boosts through their specialized support for mixed-precision matrix math. Alternatively, tensor cores can be programmed directly using CUDA APIs or inline assembly instructions, which opens up the possibility of generating efficient CUDA kernels automatically for such computations. Automatic kernel generation is particularly crucial when it is beneficial to generate efficient code for an entire computation graph by fusing several operations into a single device function instead of invoking a separate kernel for each of them. Polyhedral compilation techniques provide a systematic approach for the analysis and transformation of a sequence of affine loop-nests. In this paper, we describe a polyhedral approach to generate efficient CUDA kernels for matrix multiplication using inline assembly instructions for programming tensor cores on NVIDIA Volta GPUs. Furthermore, we build on this approach to generate fused kernels for computation sequences involving matrix multiplication and pointwise operations such as bias addition, ReLU activation etc. Experimental evaluation of these techniques show that automatically generated kernels can provide significantly better performance than manually tuned library implementations, with speedups ranging up to 2.55X.
|
2111.06995
|
Zhimin Gao
|
Shuangyan Miao, Yonghong Hou, Zhimin Gao, Mingliang Xu, and Wanqing Li
|
A Central Difference Graph Convolutional Operator for Skeleton-Based
Action Recognition
|
Accepted by IEEE Transactions on Circuits and Systems for Video
Technology (TCSVT)
| null |
10.1109/TCSVT.2021.3124562
| null |
cs.CV
|
http://creativecommons.org/licenses/by-nc-nd/4.0/
|
This paper proposes a new graph convolutional operator called central
difference graph convolution (CDGC) for skeleton based action recognition. It
is not only able to aggregate node information like a vanilla graph
convolutional operation but also gradient information. Without introducing any
additional parameters, CDGC can replace vanilla graph convolution in any
existing Graph Convolutional Networks (GCNs). In addition, an accelerated
version of the CDGC is developed which greatly improves the speed of training.
Experiments on two popular large-scale datasets NTU RGB+D 60 & 120 have
demonstrated the efficacy of the proposed CDGC. Code is available at
https://github.com/iesymiao/CD-GCN.
|
[
{
"created": "Sat, 13 Nov 2021 00:02:57 GMT",
"version": "v1"
}
] |
2021-11-16
|
[
[
"Miao",
"Shuangyan",
""
],
[
"Hou",
"Yonghong",
""
],
[
"Gao",
"Zhimin",
""
],
[
"Xu",
"Mingliang",
""
],
[
"Li",
"Wanqing",
""
]
] |
This paper proposes a new graph convolutional operator called central difference graph convolution (CDGC) for skeleton based action recognition. It is not only able to aggregate node information like a vanilla graph convolutional operation but also gradient information. Without introducing any additional parameters, CDGC can replace vanilla graph convolution in any existing Graph Convolutional Networks (GCNs). In addition, an accelerated version of the CDGC is developed which greatly improves the speed of training. Experiments on two popular large-scale datasets NTU RGB+D 60 & 120 have demonstrated the efficacy of the proposed CDGC. Code is available at https://github.com/iesymiao/CD-GCN.
|
2303.16154
|
Danko Nikolic
|
Danko Nikoli\'c, Davor Andri\'c, Vjekoslav Nikoli\'c
|
Guided Transfer Learning
| null | null | null | null |
cs.LG
|
http://creativecommons.org/licenses/by-nc-sa/4.0/
|
Machine learning requires exuberant amounts of data and computation. Also,
models require equally excessive growth in the number of parameters. It is,
therefore, sensible to look for technologies that reduce these demands on
resources. Here, we propose an approach called guided transfer learning. Each
weight and bias in the network has its own guiding parameter that indicates how
much this parameter is allowed to change while learning a new task. Guiding
parameters are learned during an initial scouting process. Guided transfer
learning can result in a reduction in resources needed to train a network. In
some applications, guided transfer learning enables the network to learn from a
small amount of data. In other cases, a network with a smaller number of
parameters can learn a task which otherwise only a larger network could learn.
Guided transfer learning potentially has many applications when the amount of
data, model size, or the availability of computational resources reach their
limits.
|
[
{
"created": "Sun, 26 Mar 2023 18:21:24 GMT",
"version": "v1"
}
] |
2023-03-29
|
[
[
"Nikolić",
"Danko",
""
],
[
"Andrić",
"Davor",
""
],
[
"Nikolić",
"Vjekoslav",
""
]
] |
Machine learning requires exuberant amounts of data and computation. Also, models require equally excessive growth in the number of parameters. It is, therefore, sensible to look for technologies that reduce these demands on resources. Here, we propose an approach called guided transfer learning. Each weight and bias in the network has its own guiding parameter that indicates how much this parameter is allowed to change while learning a new task. Guiding parameters are learned during an initial scouting process. Guided transfer learning can result in a reduction in resources needed to train a network. In some applications, guided transfer learning enables the network to learn from a small amount of data. In other cases, a network with a smaller number of parameters can learn a task which otherwise only a larger network could learn. Guided transfer learning potentially has many applications when the amount of data, model size, or the availability of computational resources reach their limits.
|
1810.05934
|
Liam Li
|
Liam Li, Kevin Jamieson, Afshin Rostamizadeh, Ekaterina Gonina, Moritz
Hardt, Benjamin Recht, Ameet Talwalkar
|
A System for Massively Parallel Hyperparameter Tuning
|
v2: Corrected typo in Algorithm 1 v3: Added comparison to BOHB and
parallel version of synchronous SHA. Add PBT to experiment in Section 4.3.1
v4: Added acknowledgements and slight edit to related work
|
Conference on Machine Learning and Systems 2020
| null | null |
cs.LG stat.ML
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
Modern learning models are characterized by large hyperparameter spaces and
long training times. These properties, coupled with the rise of parallel
computing and the growing demand to productionize machine learning workloads,
motivate the need to develop mature hyperparameter optimization functionality
in distributed computing settings. We address this challenge by first
introducing a simple and robust hyperparameter optimization algorithm called
ASHA, which exploits parallelism and aggressive early-stopping to tackle
large-scale hyperparameter optimization problems. Our extensive empirical
results show that ASHA outperforms existing state-of-the-art hyperparameter
optimization methods; scales linearly with the number of workers in distributed
settings; and is suitable for massive parallelism, as demonstrated on a task
with 500 workers. We then describe several design decisions we encountered,
along with our associated solutions, when integrating ASHA in Determined AI's
end-to-end production-quality machine learning system that offers
hyperparameter tuning as a service.
|
[
{
"created": "Sat, 13 Oct 2018 22:02:52 GMT",
"version": "v1"
},
{
"created": "Wed, 17 Oct 2018 00:23:57 GMT",
"version": "v2"
},
{
"created": "Thu, 29 Nov 2018 04:41:42 GMT",
"version": "v3"
},
{
"created": "Wed, 23 Jan 2019 02:15:22 GMT",
"version": "v4"
},
{
"created": "Mon, 16 Mar 2020 01:28:21 GMT",
"version": "v5"
}
] |
2020-03-17
|
[
[
"Li",
"Liam",
""
],
[
"Jamieson",
"Kevin",
""
],
[
"Rostamizadeh",
"Afshin",
""
],
[
"Gonina",
"Ekaterina",
""
],
[
"Hardt",
"Moritz",
""
],
[
"Recht",
"Benjamin",
""
],
[
"Talwalkar",
"Ameet",
""
]
] |
Modern learning models are characterized by large hyperparameter spaces and long training times. These properties, coupled with the rise of parallel computing and the growing demand to productionize machine learning workloads, motivate the need to develop mature hyperparameter optimization functionality in distributed computing settings. We address this challenge by first introducing a simple and robust hyperparameter optimization algorithm called ASHA, which exploits parallelism and aggressive early-stopping to tackle large-scale hyperparameter optimization problems. Our extensive empirical results show that ASHA outperforms existing state-of-the-art hyperparameter optimization methods; scales linearly with the number of workers in distributed settings; and is suitable for massive parallelism, as demonstrated on a task with 500 workers. We then describe several design decisions we encountered, along with our associated solutions, when integrating ASHA in Determined AI's end-to-end production-quality machine learning system that offers hyperparameter tuning as a service.
|
1705.02883
|
Umar Iqbal
|
Umar Iqbal, Andreas Doering, Hashim Yasin, Bj\"orn Kr\"uger, Andreas
Weber, Juergen Gall
|
A Dual-Source Approach for 3D Human Pose Estimation from a Single Image
|
under consideration at Computer Vision and Image Understanding.
Extended version of CVPR-2016 paper, arXiv:1509.06720
| null | null | null |
cs.CV
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
In this work we address the challenging problem of 3D human pose estimation
from single images. Recent approaches learn deep neural networks to regress 3D
pose directly from images. One major challenge for such methods, however, is
the collection of training data. Specifically, collecting large amounts of
training data containing unconstrained images annotated with accurate 3D poses
is infeasible. We therefore propose to use two independent training sources.
The first source consists of accurate 3D motion capture data, and the second
source consists of unconstrained images with annotated 2D poses. To integrate
both sources, we propose a dual-source approach that combines 2D pose
estimation with efficient 3D pose retrieval. To this end, we first convert the
motion capture data into a normalized 2D pose space, and separately learn a 2D
pose estimation model from the image data. During inference, we estimate the 2D
pose and efficiently retrieve the nearest 3D poses. We then jointly estimate a
mapping from the 3D pose space to the image and reconstruct the 3D pose. We
provide a comprehensive evaluation of the proposed method and experimentally
demonstrate the effectiveness of our approach, even when the skeleton
structures of the two sources differ substantially.
|
[
{
"created": "Mon, 8 May 2017 14:03:48 GMT",
"version": "v1"
},
{
"created": "Wed, 6 Sep 2017 13:24:52 GMT",
"version": "v2"
}
] |
2017-09-07
|
[
[
"Iqbal",
"Umar",
""
],
[
"Doering",
"Andreas",
""
],
[
"Yasin",
"Hashim",
""
],
[
"Krüger",
"Björn",
""
],
[
"Weber",
"Andreas",
""
],
[
"Gall",
"Juergen",
""
]
] |
In this work we address the challenging problem of 3D human pose estimation from single images. Recent approaches learn deep neural networks to regress 3D pose directly from images. One major challenge for such methods, however, is the collection of training data. Specifically, collecting large amounts of training data containing unconstrained images annotated with accurate 3D poses is infeasible. We therefore propose to use two independent training sources. The first source consists of accurate 3D motion capture data, and the second source consists of unconstrained images with annotated 2D poses. To integrate both sources, we propose a dual-source approach that combines 2D pose estimation with efficient 3D pose retrieval. To this end, we first convert the motion capture data into a normalized 2D pose space, and separately learn a 2D pose estimation model from the image data. During inference, we estimate the 2D pose and efficiently retrieve the nearest 3D poses. We then jointly estimate a mapping from the 3D pose space to the image and reconstruct the 3D pose. We provide a comprehensive evaluation of the proposed method and experimentally demonstrate the effectiveness of our approach, even when the skeleton structures of the two sources differ substantially.
|
2404.17597
|
Alexander Rogiers
|
Alexander Rogiers, Maarten Buyl, Bo Kang, and Tijl De Bie
|
KamerRaad: Enhancing Information Retrieval in Belgian National Politics
through Hierarchical Summarization and Conversational Interfaces
|
4 pages, 2 figures, submitted to 2024 ECML-PKDD demo track
| null | null | null |
cs.IR
|
http://creativecommons.org/licenses/by/4.0/
|
KamerRaad is an AI tool that leverages large language models to help citizens
interactively engage with Belgian political information. The tool extracts and
concisely summarizes key excerpts from parliamentary proceedings, followed by
the potential for interaction based on generative AI that allows users to
steadily build up their understanding. KamerRaad's front-end, built with
Streamlit, facilitates easy interaction, while the back-end employs open-source
models for text embedding and generation to ensure accurate and relevant
responses. By collecting feedback, we intend to enhance the relevancy of our
source retrieval and the quality of our summarization, thereby enriching the
user experience with a focus on source-driven dialogue.
|
[
{
"created": "Mon, 22 Apr 2024 15:01:39 GMT",
"version": "v1"
}
] |
2024-04-30
|
[
[
"Rogiers",
"Alexander",
""
],
[
"Buyl",
"Maarten",
""
],
[
"Kang",
"Bo",
""
],
[
"De Bie",
"Tijl",
""
]
] |
KamerRaad is an AI tool that leverages large language models to help citizens interactively engage with Belgian political information. The tool extracts and concisely summarizes key excerpts from parliamentary proceedings, followed by the potential for interaction based on generative AI that allows users to steadily build up their understanding. KamerRaad's front-end, built with Streamlit, facilitates easy interaction, while the back-end employs open-source models for text embedding and generation to ensure accurate and relevant responses. By collecting feedback, we intend to enhance the relevancy of our source retrieval and the quality of our summarization, thereby enriching the user experience with a focus on source-driven dialogue.
|
2202.11784
|
Jiajia Zhang
|
Jiajia Zhang and Jiyuan Tian and Dibin Zhu and Yang Liu and Shyam
Prasad
|
Design and experimental investigation of a vibro-impact self-propelled
capsule robot with orientation control
|
ICRA 2022 Conference paper
| null | null | null |
cs.RO cs.SY eess.SY
|
http://creativecommons.org/licenses/by-nc-sa/4.0/
|
This paper presents a novel design and experimental investigation for a
self-propelled capsule robot that can be used for painless colonoscopy during a
retrograde progression from the patient's rectum. The steerable robot is driven
forward and backward via its internal vibration and impact with orientation
control by using an electromagnetic actuator. The actuator contains four sets
of coils and a shaft made by permanent magnet. The shaft can be excited
linearly in a controllable and tilted angle, so guide the progression
orientation of the robot. Two control strategies are studied in this work and
compared via simulation and experiment. Extensive results are presented to
demonstrate the progression efficiency of the robot and its potential for
robotic colonoscopy.
|
[
{
"created": "Wed, 23 Feb 2022 21:00:32 GMT",
"version": "v1"
},
{
"created": "Tue, 1 Mar 2022 18:52:30 GMT",
"version": "v2"
}
] |
2022-03-02
|
[
[
"Zhang",
"Jiajia",
""
],
[
"Tian",
"Jiyuan",
""
],
[
"Zhu",
"Dibin",
""
],
[
"Liu",
"Yang",
""
],
[
"Prasad",
"Shyam",
""
]
] |
This paper presents a novel design and experimental investigation for a self-propelled capsule robot that can be used for painless colonoscopy during a retrograde progression from the patient's rectum. The steerable robot is driven forward and backward via its internal vibration and impact with orientation control by using an electromagnetic actuator. The actuator contains four sets of coils and a shaft made by permanent magnet. The shaft can be excited linearly in a controllable and tilted angle, so guide the progression orientation of the robot. Two control strategies are studied in this work and compared via simulation and experiment. Extensive results are presented to demonstrate the progression efficiency of the robot and its potential for robotic colonoscopy.
|
2202.04513
|
Tom Sterkenburg
|
Tom F. Sterkenburg, Peter D. Gr\"unwald
|
The no-free-lunch theorems of supervised learning
| null |
Synthese 199:9979-10015 (2021)
|
10.1007/s11229-021-03233-1
| null |
cs.LG
|
http://creativecommons.org/licenses/by/4.0/
|
The no-free-lunch theorems promote a skeptical conclusion that all possible
machine learning algorithms equally lack justification. But how could this
leave room for a learning theory, that shows that some algorithms are better
than others? Drawing parallels to the philosophy of induction, we point out
that the no-free-lunch results presuppose a conception of learning algorithms
as purely data-driven. On this conception, every algorithm must have an
inherent inductive bias, that wants justification. We argue that many standard
learning algorithms should rather be understood as model-dependent: in each
application they also require for input a model, representing a bias. Generic
algorithms themselves, they can be given a model-relative justification.
|
[
{
"created": "Wed, 9 Feb 2022 15:24:30 GMT",
"version": "v1"
}
] |
2022-02-10
|
[
[
"Sterkenburg",
"Tom F.",
""
],
[
"Grünwald",
"Peter D.",
""
]
] |
The no-free-lunch theorems promote a skeptical conclusion that all possible machine learning algorithms equally lack justification. But how could this leave room for a learning theory, that shows that some algorithms are better than others? Drawing parallels to the philosophy of induction, we point out that the no-free-lunch results presuppose a conception of learning algorithms as purely data-driven. On this conception, every algorithm must have an inherent inductive bias, that wants justification. We argue that many standard learning algorithms should rather be understood as model-dependent: in each application they also require for input a model, representing a bias. Generic algorithms themselves, they can be given a model-relative justification.
|
2307.09762
|
Abhishek Ajayakumar
|
Abhishek Ajayakumar, Soumyendu Raha
|
Reinforcing POD-based model reduction techniques in reaction-diffusion
complex networks using stochastic filtering and pattern recognition
|
19 pages, 6 figures
| null | null | null |
cs.CE cs.AI cs.LG math.OC
|
http://creativecommons.org/licenses/by/4.0/
|
Complex networks are used to model many real-world systems. However, the
dimensionality of these systems can make them challenging to analyze.
Dimensionality reduction techniques like POD can be used in such cases.
However, these models are susceptible to perturbations in the input data. We
propose an algorithmic framework that combines techniques from pattern
recognition (PR) and stochastic filtering theory to enhance the output of such
models. The results of our study show that our method can improve the accuracy
of the surrogate model under perturbed inputs. Deep Neural Networks (DNNs) are
susceptible to adversarial attacks. However, recent research has revealed that
Neural Ordinary Differential Equations (neural ODEs) exhibit robustness in
specific applications. We benchmark our algorithmic framework with the neural
ODE-based approach as a reference.
|
[
{
"created": "Wed, 19 Jul 2023 05:45:05 GMT",
"version": "v1"
},
{
"created": "Sat, 16 Sep 2023 14:09:43 GMT",
"version": "v2"
}
] |
2023-09-19
|
[
[
"Ajayakumar",
"Abhishek",
""
],
[
"Raha",
"Soumyendu",
""
]
] |
Complex networks are used to model many real-world systems. However, the dimensionality of these systems can make them challenging to analyze. Dimensionality reduction techniques like POD can be used in such cases. However, these models are susceptible to perturbations in the input data. We propose an algorithmic framework that combines techniques from pattern recognition (PR) and stochastic filtering theory to enhance the output of such models. The results of our study show that our method can improve the accuracy of the surrogate model under perturbed inputs. Deep Neural Networks (DNNs) are susceptible to adversarial attacks. However, recent research has revealed that Neural Ordinary Differential Equations (neural ODEs) exhibit robustness in specific applications. We benchmark our algorithmic framework with the neural ODE-based approach as a reference.
|
2208.07601
|
Wenhao Ye
|
Wenhao Ye, Huihui Wu, Shitong Wu, Yizhu Wang, Wenyi Zhang, Hao Wu and
Bo Bai
|
An Optimal Transport Approach to the Computation of the LM Rate
| null | null | null | null |
cs.IT math.IT stat.CO
|
http://creativecommons.org/licenses/by/4.0/
|
Mismatch capacity characterizes the highest information rate for a channel
under a prescribed decoding metric, and is thus a highly relevant fundamental
performance metric when dealing with many practically important communication
scenarios. Compared with the frequently used generalized mutual information
(GMI), the LM rate has been known as a tighter lower bound of the mismatch
capacity. The computation of the LM rate, however, has been a difficult task,
due to the fact that the LM rate involves a maximization over a function of the
channel input, which becomes challenging as the input alphabet size grows, and
direct numerical methods (e.g., interior point methods) suffer from intensive
memory and computational resource requirements. Noting that the computation of
the LM rate can also be formulated as an entropy-based optimization problem
with constraints, in this work, we transform the task into an optimal transport
(OT) problem with an extra constraint. This allows us to efficiently and
accurately accomplish our task by using the well-known Sinkhorn algorithm.
Indeed, only a few iterations are required for convergence, due to the fact
that the formulated problem does not contain additional regularization terms.
Moreover, we convert the extra constraint into a root-finding procedure for a
one-dimensional monotonic function. Numerical experiments demonstrate the
feasibility and efficiency of our OT approach to the computation of the LM
rate.
|
[
{
"created": "Tue, 16 Aug 2022 08:33:20 GMT",
"version": "v1"
}
] |
2022-08-17
|
[
[
"Ye",
"Wenhao",
""
],
[
"Wu",
"Huihui",
""
],
[
"Wu",
"Shitong",
""
],
[
"Wang",
"Yizhu",
""
],
[
"Zhang",
"Wenyi",
""
],
[
"Wu",
"Hao",
""
],
[
"Bai",
"Bo",
""
]
] |
Mismatch capacity characterizes the highest information rate for a channel under a prescribed decoding metric, and is thus a highly relevant fundamental performance metric when dealing with many practically important communication scenarios. Compared with the frequently used generalized mutual information (GMI), the LM rate has been known as a tighter lower bound of the mismatch capacity. The computation of the LM rate, however, has been a difficult task, due to the fact that the LM rate involves a maximization over a function of the channel input, which becomes challenging as the input alphabet size grows, and direct numerical methods (e.g., interior point methods) suffer from intensive memory and computational resource requirements. Noting that the computation of the LM rate can also be formulated as an entropy-based optimization problem with constraints, in this work, we transform the task into an optimal transport (OT) problem with an extra constraint. This allows us to efficiently and accurately accomplish our task by using the well-known Sinkhorn algorithm. Indeed, only a few iterations are required for convergence, due to the fact that the formulated problem does not contain additional regularization terms. Moreover, we convert the extra constraint into a root-finding procedure for a one-dimensional monotonic function. Numerical experiments demonstrate the feasibility and efficiency of our OT approach to the computation of the LM rate.
|
1401.1458
|
Young-Ho Eom
|
Young-Ho Eom, Hang-Hyun Jo
|
Generalized friendship paradox in complex networks: The case of
scientific collaboration
|
Published in Scientific Reports. 9 pages, 3 figures
|
Scientific Reports 4, 4603 (2014)
|
10.1038/srep04603
| null |
cs.SI physics.data-an physics.soc-ph
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
The friendship paradox states that your friends have on average more friends
than you have. Does the paradox "hold" for other individual characteristics
like income or happiness? To address this question, we generalize the
friendship paradox for arbitrary node characteristics in complex networks. By
analyzing two coauthorship networks of Physical Review journals and Google
Scholar profiles, we find that the generalized friendship paradox (GFP) holds
at the individual and network levels for various characteristics, including the
number of coauthors, the number of citations, and the number of publications.
The origin of the GFP is shown to be rooted in positive correlations between
degree and characteristics. As a fruitful application of the GFP, we suggest
effective and efficient sampling methods for identifying high characteristic
nodes in large-scale networks. Our study on the GFP can shed lights on
understanding the interplay between network structure and node characteristics
in complex networks.
|
[
{
"created": "Tue, 7 Jan 2014 17:51:14 GMT",
"version": "v1"
},
{
"created": "Mon, 31 Mar 2014 16:59:54 GMT",
"version": "v2"
},
{
"created": "Thu, 10 Apr 2014 09:21:46 GMT",
"version": "v3"
}
] |
2014-04-11
|
[
[
"Eom",
"Young-Ho",
""
],
[
"Jo",
"Hang-Hyun",
""
]
] |
The friendship paradox states that your friends have on average more friends than you have. Does the paradox "hold" for other individual characteristics like income or happiness? To address this question, we generalize the friendship paradox for arbitrary node characteristics in complex networks. By analyzing two coauthorship networks of Physical Review journals and Google Scholar profiles, we find that the generalized friendship paradox (GFP) holds at the individual and network levels for various characteristics, including the number of coauthors, the number of citations, and the number of publications. The origin of the GFP is shown to be rooted in positive correlations between degree and characteristics. As a fruitful application of the GFP, we suggest effective and efficient sampling methods for identifying high characteristic nodes in large-scale networks. Our study on the GFP can shed lights on understanding the interplay between network structure and node characteristics in complex networks.
|
2210.11065
|
Digbalay Bose
|
Digbalay Bose, Rajat Hebbar, Krishna Somandepalli, Haoyang Zhang, Yin
Cui, Kree Cole-McLaughlin, Huisheng Wang, Shrikanth Narayanan
|
MovieCLIP: Visual Scene Recognition in Movies
|
Accepted to 2023 IEEE/CVF Winter Conference on Applications of
Computer Vision (WACV 2023). Project website with supplemental material:
https://sail.usc.edu/~mica/MovieCLIP/. Revised version with updated author
affiliations
| null | null | null |
cs.CV cs.CL cs.MM
|
http://creativecommons.org/licenses/by/4.0/
|
Longform media such as movies have complex narrative structures, with events
spanning a rich variety of ambient visual scenes. Domain specific challenges
associated with visual scenes in movies include transitions, person coverage,
and a wide array of real-life and fictional scenarios. Existing visual scene
datasets in movies have limited taxonomies and don't consider the visual scene
transition within movie clips. In this work, we address the problem of visual
scene recognition in movies by first automatically curating a new and extensive
movie-centric taxonomy of 179 scene labels derived from movie scripts and
auxiliary web-based video datasets. Instead of manual annotations which can be
expensive, we use CLIP to weakly label 1.12 million shots from 32K movie clips
based on our proposed taxonomy. We provide baseline visual models trained on
the weakly labeled dataset called MovieCLIP and evaluate them on an independent
dataset verified by human raters. We show that leveraging features from models
pretrained on MovieCLIP benefits downstream tasks such as multi-label scene and
genre classification of web videos and movie trailers.
|
[
{
"created": "Thu, 20 Oct 2022 07:38:56 GMT",
"version": "v1"
},
{
"created": "Sun, 23 Oct 2022 01:25:13 GMT",
"version": "v2"
}
] |
2022-10-25
|
[
[
"Bose",
"Digbalay",
""
],
[
"Hebbar",
"Rajat",
""
],
[
"Somandepalli",
"Krishna",
""
],
[
"Zhang",
"Haoyang",
""
],
[
"Cui",
"Yin",
""
],
[
"Cole-McLaughlin",
"Kree",
""
],
[
"Wang",
"Huisheng",
""
],
[
"Narayanan",
"Shrikanth",
""
]
] |
Longform media such as movies have complex narrative structures, with events spanning a rich variety of ambient visual scenes. Domain specific challenges associated with visual scenes in movies include transitions, person coverage, and a wide array of real-life and fictional scenarios. Existing visual scene datasets in movies have limited taxonomies and don't consider the visual scene transition within movie clips. In this work, we address the problem of visual scene recognition in movies by first automatically curating a new and extensive movie-centric taxonomy of 179 scene labels derived from movie scripts and auxiliary web-based video datasets. Instead of manual annotations which can be expensive, we use CLIP to weakly label 1.12 million shots from 32K movie clips based on our proposed taxonomy. We provide baseline visual models trained on the weakly labeled dataset called MovieCLIP and evaluate them on an independent dataset verified by human raters. We show that leveraging features from models pretrained on MovieCLIP benefits downstream tasks such as multi-label scene and genre classification of web videos and movie trailers.
|
1807.04040
|
Jeevan Manavalan
|
Jeevan Manavalan, Matthew Howard
|
Learning Singularity Avoidance
| null | null | null | null |
cs.RO cs.LG
|
http://creativecommons.org/licenses/by/4.0/
|
With the increase in complexity of robotic systems and the rise in non-expert
users, it can be assumed that task constraints are not explicitly known. In
tasks where avoiding singularity is critical to its success, this paper
provides an approach, especially for non-expert users, for the system to learn
the constraints contained in a set of demonstrations, such that they can be
used to optimise an autonomous controller to avoid singularity, without having
to explicitly know the task constraints. The proposed approach avoids
singularity, and thereby unpredictable behaviour when carrying out a task, by
maximising the learnt manipulability throughout the motion of the constrained
system, and is not limited to kinematic systems. Its benefits are demonstrated
through comparisons with other control policies which show that the constrained
manipulability of a system learnt through demonstration can be used to avoid
singularities in cases where these other policies would fail. In the absence of
the systems manipulability subject to a tasks constraints, the proposed
approach can be used instead to infer these with results showing errors less
than 10^-5 in 3DOF simulated systems as well as 10^-2 using a 7DOF real world
robotic system.
|
[
{
"created": "Wed, 11 Jul 2018 09:46:05 GMT",
"version": "v1"
},
{
"created": "Mon, 25 Mar 2019 22:03:01 GMT",
"version": "v2"
}
] |
2019-03-27
|
[
[
"Manavalan",
"Jeevan",
""
],
[
"Howard",
"Matthew",
""
]
] |
With the increase in complexity of robotic systems and the rise in non-expert users, it can be assumed that task constraints are not explicitly known. In tasks where avoiding singularity is critical to its success, this paper provides an approach, especially for non-expert users, for the system to learn the constraints contained in a set of demonstrations, such that they can be used to optimise an autonomous controller to avoid singularity, without having to explicitly know the task constraints. The proposed approach avoids singularity, and thereby unpredictable behaviour when carrying out a task, by maximising the learnt manipulability throughout the motion of the constrained system, and is not limited to kinematic systems. Its benefits are demonstrated through comparisons with other control policies which show that the constrained manipulability of a system learnt through demonstration can be used to avoid singularities in cases where these other policies would fail. In the absence of the systems manipulability subject to a tasks constraints, the proposed approach can be used instead to infer these with results showing errors less than 10^-5 in 3DOF simulated systems as well as 10^-2 using a 7DOF real world robotic system.
|
2005.09512
|
Frederico Gadelha Guimaraes
|
Leonardo Augusto Ferreira and Frederico Gadelha Guimar\~aes and
Rodrigo Silva
|
Applying Genetic Programming to Improve Interpretability in Machine
Learning Models
|
8 pages, 8 figures, submitted and accepted to 2020 IEEE Congress on
Evolutionary Computation (IEEE CEC 2020). Copyright 2020 IEEE. Personal use
of this material is permitted. Permission from IEEE must be obtained for all
other uses
| null | null | null |
cs.LG cs.AI cs.NE cs.SC
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
Explainable Artificial Intelligence (or xAI) has become an important research
topic in the fields of Machine Learning and Deep Learning. In this paper, we
propose a Genetic Programming (GP) based approach, named Genetic Programming
Explainer (GPX), to the problem of explaining decisions computed by AI systems.
The method generates a noise set located in the neighborhood of the point of
interest, whose prediction should be explained, and fits a local explanation
model for the analyzed sample. The tree structure generated by GPX provides a
comprehensible analytical, possibly non-linear, symbolic expression which
reflects the local behavior of the complex model. We considered three machine
learning techniques that can be recognized as complex black-box models: Random
Forest, Deep Neural Network and Support Vector Machine in twenty data sets for
regression and classifications problems. Our results indicate that the GPX is
able to produce more accurate understanding of complex models than the state of
the art. The results validate the proposed approach as a novel way to deploy GP
to improve interpretability.
|
[
{
"created": "Mon, 18 May 2020 16:09:49 GMT",
"version": "v1"
}
] |
2020-05-20
|
[
[
"Ferreira",
"Leonardo Augusto",
""
],
[
"Guimarães",
"Frederico Gadelha",
""
],
[
"Silva",
"Rodrigo",
""
]
] |
Explainable Artificial Intelligence (or xAI) has become an important research topic in the fields of Machine Learning and Deep Learning. In this paper, we propose a Genetic Programming (GP) based approach, named Genetic Programming Explainer (GPX), to the problem of explaining decisions computed by AI systems. The method generates a noise set located in the neighborhood of the point of interest, whose prediction should be explained, and fits a local explanation model for the analyzed sample. The tree structure generated by GPX provides a comprehensible analytical, possibly non-linear, symbolic expression which reflects the local behavior of the complex model. We considered three machine learning techniques that can be recognized as complex black-box models: Random Forest, Deep Neural Network and Support Vector Machine in twenty data sets for regression and classifications problems. Our results indicate that the GPX is able to produce more accurate understanding of complex models than the state of the art. The results validate the proposed approach as a novel way to deploy GP to improve interpretability.
|
2011.14365
|
Weifeng Zhu
|
Jiazhu Dai, Weifeng Zhu, Xiangfeng Luo
|
A Targeted Universal Attack on Graph Convolutional Network
| null | null | null | null |
cs.LG cs.CR
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
Graph-structured data exist in numerous applications in real life. As a
state-of-the-art graph neural network, the graph convolutional network (GCN)
plays an important role in processing graph-structured data. However, a recent
study reported that GCNs are also vulnerable to adversarial attacks, which
means that GCN models may suffer malicious attacks with unnoticeable
modifications of the data. Among all the adversarial attacks on GCNs, there is
a special kind of attack method called the universal adversarial attack, which
generates a perturbation that can be applied to any sample and causes GCN
models to output incorrect results. Although universal adversarial attacks in
computer vision have been extensively researched, there are few research works
on universal adversarial attacks on graph structured data. In this paper, we
propose a targeted universal adversarial attack against GCNs. Our method
employs a few nodes as the attack nodes. The attack capability of the attack
nodes is enhanced through a small number of fake nodes connected to them.
During an attack, any victim node will be misclassified by the GCN as the
attack node class as long as it is linked to them. The experiments on three
popular datasets show that the average attack success rate of the proposed
attack on any victim node in the graph reaches 83% when using only 3 attack
nodes and 6 fake nodes. We hope that our work will make the community aware of
the threat of this type of attack and raise the attention given to its future
defense.
|
[
{
"created": "Sun, 29 Nov 2020 13:19:53 GMT",
"version": "v1"
}
] |
2020-12-01
|
[
[
"Dai",
"Jiazhu",
""
],
[
"Zhu",
"Weifeng",
""
],
[
"Luo",
"Xiangfeng",
""
]
] |
Graph-structured data exist in numerous applications in real life. As a state-of-the-art graph neural network, the graph convolutional network (GCN) plays an important role in processing graph-structured data. However, a recent study reported that GCNs are also vulnerable to adversarial attacks, which means that GCN models may suffer malicious attacks with unnoticeable modifications of the data. Among all the adversarial attacks on GCNs, there is a special kind of attack method called the universal adversarial attack, which generates a perturbation that can be applied to any sample and causes GCN models to output incorrect results. Although universal adversarial attacks in computer vision have been extensively researched, there are few research works on universal adversarial attacks on graph structured data. In this paper, we propose a targeted universal adversarial attack against GCNs. Our method employs a few nodes as the attack nodes. The attack capability of the attack nodes is enhanced through a small number of fake nodes connected to them. During an attack, any victim node will be misclassified by the GCN as the attack node class as long as it is linked to them. The experiments on three popular datasets show that the average attack success rate of the proposed attack on any victim node in the graph reaches 83% when using only 3 attack nodes and 6 fake nodes. We hope that our work will make the community aware of the threat of this type of attack and raise the attention given to its future defense.
|
2202.05977
|
Yc Huo
|
Hangming Fan, Rui Wang, Yuchi Huo, Hujun Bao
|
Real-time Monte Carlo Denoising with Weight Sharing Kernel Prediction
Network
| null |
Computer Graphics Forum. 2021, 40(4): 15-27
|
10.1111/cgf.14338
| null |
cs.GR
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
Real-time Monte Carlo denoising aims at removing severe noise under low
samples per pixel (spp) in a strict time budget. Recently, kernel-prediction
methods use a neural network to predict each pixel's filtering kernel and have
shown a great potential to remove Monte Carlo noise. However, the heavy
computation overhead blocks these methods from real-time applications. This
paper expands the kernel-prediction method and proposes a novel approach to
denoise very low spp (e.g., 1-spp) Monte Carlo path traced images at real-time
frame rates. Instead of using the neural network to directly predict the kernel
map, i.e., the complete weights of each per-pixel filtering kernel, we predict
an encoding of the kernel map, followed by a high-efficiency decoder with
unfolding operations for a high-quality reconstruction of the filtering
kernels. The kernel map encoding yields a compact single-channel representation
of the kernel map, which can significantly reduce the kernel-prediction
network's throughput. In addition, we adopt a scalable kernel fusion module to
improve denoising quality. The proposed approach preserves kernel prediction
methods' denoising quality while roughly halving its denoising time for 1-spp
noisy inputs. In addition, compared with the recent neural bilateral grid-based
real-time denoiser, our approach benefits from the high parallelism of
kernel-based reconstruction and produces better denoising results at equal
time.
|
[
{
"created": "Sat, 12 Feb 2022 04:21:37 GMT",
"version": "v1"
},
{
"created": "Fri, 25 Feb 2022 09:16:14 GMT",
"version": "v2"
}
] |
2022-02-28
|
[
[
"Fan",
"Hangming",
""
],
[
"Wang",
"Rui",
""
],
[
"Huo",
"Yuchi",
""
],
[
"Bao",
"Hujun",
""
]
] |
Real-time Monte Carlo denoising aims at removing severe noise under low samples per pixel (spp) in a strict time budget. Recently, kernel-prediction methods use a neural network to predict each pixel's filtering kernel and have shown a great potential to remove Monte Carlo noise. However, the heavy computation overhead blocks these methods from real-time applications. This paper expands the kernel-prediction method and proposes a novel approach to denoise very low spp (e.g., 1-spp) Monte Carlo path traced images at real-time frame rates. Instead of using the neural network to directly predict the kernel map, i.e., the complete weights of each per-pixel filtering kernel, we predict an encoding of the kernel map, followed by a high-efficiency decoder with unfolding operations for a high-quality reconstruction of the filtering kernels. The kernel map encoding yields a compact single-channel representation of the kernel map, which can significantly reduce the kernel-prediction network's throughput. In addition, we adopt a scalable kernel fusion module to improve denoising quality. The proposed approach preserves kernel prediction methods' denoising quality while roughly halving its denoising time for 1-spp noisy inputs. In addition, compared with the recent neural bilateral grid-based real-time denoiser, our approach benefits from the high parallelism of kernel-based reconstruction and produces better denoising results at equal time.
|
2211.14742
|
YuTeng Ye
|
YuTeng Ye, Hang Zhou, Jiale Cai, Chenxing Gao, Youjia Zhang, Junle
Wang, Qiang Hu, Junqing Yu, Wei Yang
|
Dynamic Feature Pruning and Consolidation for Occluded Person
Re-Identification
|
Accepted by AAAI-24
| null | null | null |
cs.CV
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
Occluded person re-identification (ReID) is a challenging problem due to
contamination from occluders. Existing approaches address the issue with prior
knowledge cues, such as human body key points and semantic segmentations, which
easily fail in the presence of heavy occlusion and other humans as occluders.
In this paper, we propose a feature pruning and consolidation (FPC) framework
to circumvent explicit human structure parsing. The framework mainly consists
of a sparse encoder, a multi-view feature mathcing module, and a feature
consolidation decoder. Specifically, the sparse encoder drops less important
image tokens, mostly related to background noise and occluders, solely based on
correlation within the class token attention. Subsequently, the matching stage
relies on the preserved tokens produced by the sparse encoder to identify
k-nearest neighbors in the gallery by measuring the image and patch-level
combined similarity. Finally, we use the feature consolidation module to
compensate pruned features using identified neighbors for recovering essential
information while disregarding disturbance from noise and occlusion.
Experimental results demonstrate the effectiveness of our proposed framework on
occluded, partial, and holistic Re-ID datasets. In particular, our method
outperforms state-of-the-art results by at least 8.6\% mAP and 6.0\% Rank-1
accuracy on the challenging Occluded-Duke dataset.
|
[
{
"created": "Sun, 27 Nov 2022 06:18:40 GMT",
"version": "v1"
},
{
"created": "Thu, 21 Dec 2023 04:06:43 GMT",
"version": "v2"
}
] |
2023-12-22
|
[
[
"Ye",
"YuTeng",
""
],
[
"Zhou",
"Hang",
""
],
[
"Cai",
"Jiale",
""
],
[
"Gao",
"Chenxing",
""
],
[
"Zhang",
"Youjia",
""
],
[
"Wang",
"Junle",
""
],
[
"Hu",
"Qiang",
""
],
[
"Yu",
"Junqing",
""
],
[
"Yang",
"Wei",
""
]
] |
Occluded person re-identification (ReID) is a challenging problem due to contamination from occluders. Existing approaches address the issue with prior knowledge cues, such as human body key points and semantic segmentations, which easily fail in the presence of heavy occlusion and other humans as occluders. In this paper, we propose a feature pruning and consolidation (FPC) framework to circumvent explicit human structure parsing. The framework mainly consists of a sparse encoder, a multi-view feature mathcing module, and a feature consolidation decoder. Specifically, the sparse encoder drops less important image tokens, mostly related to background noise and occluders, solely based on correlation within the class token attention. Subsequently, the matching stage relies on the preserved tokens produced by the sparse encoder to identify k-nearest neighbors in the gallery by measuring the image and patch-level combined similarity. Finally, we use the feature consolidation module to compensate pruned features using identified neighbors for recovering essential information while disregarding disturbance from noise and occlusion. Experimental results demonstrate the effectiveness of our proposed framework on occluded, partial, and holistic Re-ID datasets. In particular, our method outperforms state-of-the-art results by at least 8.6\% mAP and 6.0\% Rank-1 accuracy on the challenging Occluded-Duke dataset.
|
1709.05365
|
Myoungsoo Jung
|
Sungjoon Koh, Jie Zhang, Miryeong Kwon, Jungyeon Yoon, David Donofrio,
Namsung Kim and Myoungsoo Jung
|
Understanding System Characteristics of Online Erasure Coding on
Scalable, Distributed and Large-Scale SSD Array Systems
|
This paper is accepted by and will be published at 2017 IEEE
International Symposium on Workload Characterization
| null | null | null |
cs.DC cs.AR
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
Large-scale systems with arrays of solid state disks (SSDs) have become
increasingly common in many computing segments. To make such systems resilient,
we can adopt erasure coding such as Reed-Solomon (RS) code as an alternative to
replication because erasure coding can offer a significantly lower storage cost
than replication. To understand the impact of using erasure coding on system
performance and other system aspects such as CPU utilization and network
traffic, we build a storage cluster consisting of approximately one hundred
processor cores with more than fifty high-performance SSDs, and evaluate the
cluster with a popular open-source distributed parallel file system, Ceph. Then
we analyze behaviors of systems adopting erasure coding from the following five
viewpoints, compared with those of systems using replication: (1) storage
system I/O performance; (2) computing and software overheads; (3) I/O
amplification; (4) network traffic among storage nodes; (5) the impact of
physical data layout on performance of RS-coded SSD arrays. For all these
analyses, we examine two representative RS configurations, which are used by
Google and Facebook file systems, and compare them with triple replication that
a typical parallel file system employs as a default fault tolerance mechanism.
Lastly, we collect 54 block-level traces from the cluster and make them
available for other researchers.
|
[
{
"created": "Thu, 14 Sep 2017 14:14:10 GMT",
"version": "v1"
},
{
"created": "Tue, 19 Sep 2017 04:14:56 GMT",
"version": "v2"
}
] |
2017-09-20
|
[
[
"Koh",
"Sungjoon",
""
],
[
"Zhang",
"Jie",
""
],
[
"Kwon",
"Miryeong",
""
],
[
"Yoon",
"Jungyeon",
""
],
[
"Donofrio",
"David",
""
],
[
"Kim",
"Namsung",
""
],
[
"Jung",
"Myoungsoo",
""
]
] |
Large-scale systems with arrays of solid state disks (SSDs) have become increasingly common in many computing segments. To make such systems resilient, we can adopt erasure coding such as Reed-Solomon (RS) code as an alternative to replication because erasure coding can offer a significantly lower storage cost than replication. To understand the impact of using erasure coding on system performance and other system aspects such as CPU utilization and network traffic, we build a storage cluster consisting of approximately one hundred processor cores with more than fifty high-performance SSDs, and evaluate the cluster with a popular open-source distributed parallel file system, Ceph. Then we analyze behaviors of systems adopting erasure coding from the following five viewpoints, compared with those of systems using replication: (1) storage system I/O performance; (2) computing and software overheads; (3) I/O amplification; (4) network traffic among storage nodes; (5) the impact of physical data layout on performance of RS-coded SSD arrays. For all these analyses, we examine two representative RS configurations, which are used by Google and Facebook file systems, and compare them with triple replication that a typical parallel file system employs as a default fault tolerance mechanism. Lastly, we collect 54 block-level traces from the cluster and make them available for other researchers.
|
1703.02197
|
EPTCS
|
Minghui Ma (Sun Yat-Sen University), Ahti-Veikko Pietarinen (Tallinn
University of Technology)
|
Graphical Sequent Calculi for Modal Logics
|
In Proceedings M4M9 2017, arXiv:1703.01736
|
EPTCS 243, 2017, pp. 91-103
|
10.4204/EPTCS.243.7
| null |
cs.LO
|
http://arxiv.org/licenses/nonexclusive-distrib/1.0/
|
The syntax of modal graphs is defined in terms of the continuous cut and
broken cut following Charles Peirce's notation in the gamma part of his
graphical logic of existential graphs. Graphical calculi for normal modal
logics are developed based on a reformulation of the graphical calculus for
classical propositional logic. These graphical calculi are of the nature of
deep inference. The relationship between graphical calculi and sequent calculi
for modal logics is shown by translations between graphs and modal formulas.
|
[
{
"created": "Tue, 7 Mar 2017 03:16:38 GMT",
"version": "v1"
}
] |
2017-03-08
|
[
[
"Ma",
"Minghui",
"",
"Sun Yat-Sen University"
],
[
"Pietarinen",
"Ahti-Veikko",
"",
"Tallinn\n University of Technology"
]
] |
The syntax of modal graphs is defined in terms of the continuous cut and broken cut following Charles Peirce's notation in the gamma part of his graphical logic of existential graphs. Graphical calculi for normal modal logics are developed based on a reformulation of the graphical calculus for classical propositional logic. These graphical calculi are of the nature of deep inference. The relationship between graphical calculi and sequent calculi for modal logics is shown by translations between graphs and modal formulas.
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.