text stringlengths 1 1k ⌀ | title stringclasses 230
values |
|---|---|
In this work we employ several CNN models trained
to predict aesthetic, sentiment and memorability scores of
natural images in order to explore those features in art
images. Fig. 1 illustrates the methodology used in this study.
Various CNN models trained on different natural image
datasets are evaluated on available s... | A_Deep_Learning_Perspective_on_Beauty_Sentiment_and_Remembrance_of_Art |
1
Introduction
Large language models (LLMs) with instruction
tuning are capable of generating remarkable out-
puts for a wide range of use cases (Ouyang et al.,
2022; Wei et al., 2022; Sanh et al., 2022; Chung
et al., 2022; OpenAI, 2023). However, these mod-
els usually have billions of parameters, which re-
quire ma... | LaMini-LM- A Diverse Herd of Distilled Models from Large-Scale Instructions |
Qwen-VL + CLoT (Ours)Qwen-VL + AIT (Ours)Qwen-VLVisualGLM-6BmPLUG-OwlMiniGPT-v2LLaVA-1.5GPT4v3T1 (EN)102030Rank (EN)Qwen-VL + CLoT (Ours)Qwen-VL + AIT (Ours)Qwen-VLVisualGLM-6BmPLUG-OwlMiniGPT-v2LLaVA-1.5GPT4v616263Accuracy (%)NDCG (%)Qwen-VL + CLoT (Ours)Qwen-VL + AIT (Ours)Qwen-VLVisualGLM-6BmPLUG-OwlMiniGPT-v2LLaVA-... | Let’sThinkOutsidetheBox |
A Survey on Evaluation of Large Language Models
111:29
6.2 Benchmark and Evaluation Protocol
With the rapid development and widespread use of LLMs, the importance of evaluating them in
practical applications and research has become crucial. This evaluation process should include not
only task-level evaluation but als... | ASurveyonEvaluationofLargeLanguageModels |
10
4401–4410, 2019.
[29] Tero Karras, Miika Aittala, Janne Hellsten, Samuli Laine, Jaakko Lehtinen, and Timo Aila. Training
generative adversarial networks with limited data. arXiv preprint arXiv:2006.06676v1, 2020.
[30] Tero Karras, Samuli Laine, Miika Aittala, Janne Hellsten, Jaakko Lehtinen, and Timo Aila. Anal... | Denoising Diffusion Probabilistic Models |
firm acceptance of an offer of admission in order to guarantee the allocation of a room.
2. Applicants should note that only those who have an offer of a place to study at UCL can apply
3. Further information about applying for student accommodation can be found on the Prospective
for accommodation.
Students ... | UCL Academic Manual |
We evaluate whether the generated outputs of our model are
fair regarding protected attributes such as (1) Perceived Age
(2) Perceived Gender Expression (3) Perceived Skin Tone.
We construct 306 prompts with template — “a {profession
or people descriptor} looking {adverb} at the camera” with
“profession” being crawled ... | VideoPoet |
[9] Joseph Bates. 1994. The Role of Emotion in Believable Agents. Commun. ACM
37, 7 (1994), 122–125. https://doi.org/10.1145/176789.176803
[10] Christopher Berner, Greg Brockman, Brooke Chan, Vicki Cheung, Przemysław
Dębiak, Christy Dennison, David Farhi, Quirin Fischer, Shariq Hashme, Chris
Hesse, Rafal Józefowicz, ... | Generative Agents- Interactive Simulacra of Human Behavior |
5
Standard prompting
Equation only
Variable compute only
Reasoning after answer
Chain-of-thought prompting | Chain-of-Thought Prompting Elicits Reasoning in Large Language Models |
65.1
84.1
87.3
Trained for 1 day on an A6000:
70.1
82.2
86.2
74.1
81.4
86.4
74.8
83.2
87.5
7.3
33.8
47.2
8.2
0.0
43.7
10.3
36.3
44.5
52.0
69.7
78.3
52.2
49.9
78.1
52.2
72.9
78.6
Table 3: Comparison in GLUE-dev performance of baseline BERT to the crammed model. Note that all runs
abide by the finetuning protoc... | CRAMMING-TRAININGALANGUAGEMODELONA SINGLEGPUINONEDAY |
[65] Or Patashnik, Zongze Wu, Eli Shechtman, Daniel Cohen-Or,
and Dani Lischinski. Styleclip: Text-driven manipulation
of stylegan imagery. In Proceedings of the IEEE/CVF In-
ternational Conference on Computer Vision (ICCV), pages
2085–2094, October 2021. 3
[66] Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya
Rames... | AddingConditionalControltoText-to-ImageDiffusionModels |
26 http://www.adampease .org /OP/.
9
I. Tiddi and S. Schlobach
Artificial Intelligence 302 (2022) 103627
Table 5
Knowledge-based explanations for image recognition.
Knowledge graphs
Model
Explanations
G
K
f
o
r
e
b
m
u
N
y
bilit
a
s
u
e
R
e
p
y
T
CK
CK
CK
CK
CK
CK
CK
CK
CK
CK
FK
[55]
[56]
[57]
[58]
[59]
[... | Knowledge graphs as tools for explainable machine learning: A survey |
D.3 Text-Audio Binding
We find that the text-audio binding works well with
CFG higher than 3.0. Since the model is trained
with metadata such as title, album, artist, genre,
year, and chunk, the best keywords to control the
generation appear to be frequent descriptive names,
such as the genre of the music, or descripti... | MOUSAI |
←−
ℎ 𝑡 + 𝑏𝑦
ℎ𝑦
𝑥ℎ
𝑥ℎ
ℎℎ
𝑥𝑡 + 𝑏−→
ℎ )
𝑥𝑡 + 𝑏←−
ℎ )
(3)
(4)
(5)
10
Mehrish et al.
where high dimensional hidden states −→
context from 1, 2, . . . , 𝑡 − 1 and backward context from 𝑇 ,𝑇 − 1, . . . , 𝑡 + 1, respectively.
ℎ 𝑡+1 are hidden states modeling the forward
ℎ 𝑡−1 and ←−
Long Short-Te... | AReviewofDeepLearningTechniquesforSpeechProcessing |
the input (e.g., the calculator cannot parse “How old is Rafael Nadal?”), and therefore does not contribute to the final
output list. | LaMDA- Language Models for Dialog Applications |
2.7. Public Release and Reproducibility | Pythia- A Suite for Analyzing Large Language Models Across Training and Scaling |
Ik ga niet akkoord
Ik ga akkoord
https://www.tudelft.nl/over-tu-delft/werken-bij-tu-delft/vacatures/details?jobId=14844&jobTitle=PhD position in Grounding Large Language Models in the Rea… 2/5
20/11/2023, 08:29
Job details
remaining 2,5 years assuming everything goes well and performance requirements are
met.
Sal... | Job details - TU |
Table 4: Sample prompts of the three formats we use.
The samples are from the Social IQa task of BIG-Bench.
answer the question, but encounter difficulties in
its generation. | AreEmergentAbilitiesinLarge Language Models just In-Context |
3.3 Pretraining
DocLLM is first pre-trained in a self-supervised fashion on a large number of unlabeled documents. The self-supervised
pre-training objective in autoregressive language models [26] is generally to maximize the log-likelihood of the next
token prediction in a sequence based on the context provided by pr... | DOCLLM |
Truthfulness. Table 19 shows the evaluation results of TruthfulQA for the percentage of truthfulness,
percentage of informativeness, and percentage of both truthfulness and informativeness across generations. The
truthfulness percentage is relatively low for pretrained models, around 30% to 40% for the 7B Code Llama
an... | CodeLlama2 |
posed space x(cid:48)
colors c(cid:48)
values d(cid:48)
i = o(cid:48) + tiv(cid:48) and predict their SDF values d(cid:48)
i,
i and normals n(cid:48)
i. We convert SDF
i via the method of StyleSDF [48].
The color of each pixel in the rendered image I is com-
i, color features f(cid:48)
i to densities σ(cid:48)
puted... | AG3D- Learning to Generate 3D Avatars from 2D Image Collections |
45
E Finetuning details
We found that learning rate, batch size and the dropout were the most important hyperparameters for
instruction finetuning. Table 22 lists these values for all finetuned models studied in this paper. The reported
batch size is the global batch size (not per-device batch size). Because we use pa... | Scaling Instruction-Finetuned Language Models |
A Review of Deep Learning Techniques for Speech Processing
75
Domain adaptation has also been applied to speech recognition [213, 367, 519, 631] to improve
speech recognition accuracy in a target domain. One recent approach for domain adaptation in
ASR is prompt-tuning [112], which involves fine-tuning the ASR system... | AReviewofDeepLearningTechniquesforSpeechProcessing |
Recently, many methods learn image features aligned
with text [1, 30, 45, 59, 63, 80, 81], audio [3, 4, 49, 54,
55, 68] etc. These methods use a single pair of modali-
ties or, at best, a few visual modalities. However, the fi-
nal embeddings are limited to the pairs of modalities used
for training. Thus, video-audio e... | IMAGEBIND- One Embedding Space To Bind Them A |
0.143
0.238
0.289
0.282
0.313
6.021
4.074
3.909
3.982
3.769
Model
LTU
Table 1: Comparison of models for music understand-
ing. The best values of different metrics are made bold.
B-U↑ M-R↑ R-L↑ BERT-S↑
0.887
0.242
0.895
0.273
0.286
0.898
0.901
0.306
0.308
0.902
0.274
0.334
0.332
0.385
0.393
0.326
0.413
0.371
0.466... | M2UGen |
[101] Song, C. H., J. Wu, C. Washington, et al. Llm-planner: Few-shot grounded planning for
embodied agents with large language models. CoRR, abs/2212.04088, 2022.
[102] Akyürek, A. F., E. Akyürek, A. Kalyan, et al. RL4F: generating natural language feedback
with reinforcement learning for repairing model outputs. In... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
Google Code Jam. Google Code Jam. https://codingcompetitions.withgoogle.com/
codejam, 2021. Accessed: 2021-12-09.
C. C. Green. Application of theorem proving to problem solving. In IJCAI, 1969.
S. Gulwani. Automating string processing in spreadsheets using input-output examples. ACM Sigplan
Notices, 46(1):317–330, 2011... | alphacode |
https://doi.org/10.1017/9781108890960 Published online by Cambridge University Press
https://doi.org/10.1017/9781108890960 Published online by Cambridge University Press
1
Introduction
Nathaniel Persily and Joshua A. Tucker
Widespread concern about the effects of social media on democracy has led to
an explosion... | Social_Media_and_Democracy |
1) Adapters-based Fine-tuning: The idea of Adapter is
first
introduced in multi-domain image classification [77],
allowing for the efficient transfer of knowledge across multiple
visual domains. Sequential Adapter[9] extends and applies
it to NLP tasks by inserting the adapter (trainable modules)
into the transformer b... | Parameter-EfficientFine-TuningMethods |
2 + λtemp
t
(cid:107)tτ +1 − tτ(cid:107)2
2),
(1)
(cid:88)
Etemp =
1
T − 1
τ∈[0,T−1]
where θ and t are the pose parameters and global translation vectors. λtemp
respectively. The final optimization objective can be represented as:
ψ
, λtemp
θ
and λtemp
t
are set to 1e − 3, 2
3 and 10
3 ,
We will release... | I M Avatar- Implicit Morphable Head Avatars from Videos |
Front view
Right side view
(b) With our proposed ACAE regularization
Front view
Right side view
Figure 2: We train models to jointly estimate 3D human pose according to multiple different skeleton formats so that we can
train on many datasets at once. a) Simply using separate prediction heads on a shared backbone i... | Learning 3D Human Pose Estimation from Dozens of Datasets using a Geometry-Aware Autoencoder to Bridge Between Skeleton Formats |
related works in the appendix for reference.
2. Agent-Driver
In this section, we present Agent-Driver, an LLM-based
intelligent agent for autonomous driving. We first introduce
the overall architecture of our Agent-Driver in Section 2.1.
Then, we introduce the three key components of our method:
tool library (Section ... | ALanguageAgentforAutonomousDriving |
objectives are (and see our behavior as evidence)—has a somewhat similar flavor.
65In the sense, practical alignment is a property that holds relative to a set of inputs.
17
objectives, in some physics-compatible circumstances (see next section), this seems, at the least, a
red flag about your project.66 Ultimately, ... | Is Power-Seeking AI an Existential Risk? |
International Education (CLIE).
the (CLIE).
Recognition of Prior Learning (RPL) for Entry to UCL
Please note that there are separate regulations and an initial assessment template for Degree
Apprenticeship programmes – see Chapter 11, Section 6.1: Apprentice Support and Success/
Initial Assessment
2.8.1 Defin... | UCL Academic Manual |
5
Human-Feedback Fine-TuningPreference Model Pretraining (PMP)RLHF (PPO)HHH prompt
context distillationBHuman Feedback InterfacePretrained
LMRLHF
PoliciesInitial PolicyPreference
ModelHuman-Feedback
Comparison
Data Figure 3 RLHF model performance on zero-shot and few-shot NLP tasks. For each model size, we plot
the m... | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
Armstrong, Stuart, Nick Bostrom, and Carl Shulman. “Racing to the precipice: a model of artificial intelligence
development.” AI & Society 31, no. 2 (2016): 201-206., OpenAI Charter (2018): "We are concerned about late-
stage AGI development becoming a competitive race without time for adequate safety precautions. The... | Capabilities and risks from frontier AI |
no one interviewed expressed a desire to actually opt out. The main takeaway seemed to be that
participants found the most value in BigCode governance tools for their ability to raise awareness of
data practices and to empower individuals and communities to take action based on their specific
needs. These initial conve... | StarCoder_paper (1) |
4587 E2E
BLEU NIST MET R-L CIDEr
BLEU
TER ↓
BLEU MET TER ↓ Mover BERT BLEURT
DART
FT-FULL
FT-TOP2
ADAPTER(3%)
ADAPTER(0.1%)
PREFIX(0.1%)
FT-FULL
Prefix
SOTA
68.8
68.1
68.9
66.3
70.3
68.5
70.3
68.6
8.71
8.59
8.71
8.41
8.82
8.78
8.85
8.70
46.1 71.1
46.0 70.8
46.1 71.3
45.0 69.8
46.3 72.1
46.0 69.9
46.2 71.7
4... | Prefix-Tuning |
We identified a two factorial structure that encompassed attitudes that we summarized under the Social
Threat factor, which measures threat to oneself and others, as well as a factor that we summarized under the
Agency factor, which describes agency and support for augmented humans. The SHAPE sub-scales were validated
... | Society’sAttitudesTowardsHumanAugmentation |
concern.
Just as with governmental | Social_Media_and_Democracy |
they need to be sufficiently non-myopic to care about getting deployed, there need to be sufficient
benefits to deployment, and so on), and whose deception goes undetected. It’s not clear to me how
common to expect this to be, especially given that we’ll likely be on the lookout for it.
More generally, I expect decision-m... | Is Power-Seeking AI an Existential Risk? |
before the rise of the Internet, in contrast to the United States where such speech
is legally protected. Private speech is also more constrained in certain European
countries as a result of tougher libel laws that facilitate civil litigation against
private individuals. | Social_Media_and_Democracy |
• Using Tools: Some benchmarks have been proposed to evaluate the tool usage capabilities of LLMs.
API-Bank (Li et al., 2023b) is specifically designed for tool-augmented LLMs. ToolBench (Xu
et al., 2023c) is a tool manipulation benchmark including various software tools for real-world tasks.
APIBench (Patil et al., 20... | ChatGPT’sOne-yearAnniversary-AreOpen-Source LargeLanguageModelsCatchingup |
2. How well can Jurassic-X generalize between math problems that vary in type
or complexity?
To answer these questions we evaluate the model’s performance on cases where
the test data is drawn from the same distribution as the training data, as well as on
out-of-distribution test data (allowing us, for example, to tr... | MRKL Systems |
slope diminishes over time. This pattern suggests that despite the rising temperature, the model learns to
consistently provide the same response to factual prompts. | Llama2 |
generated 3D scenes related to the text prompts, previous
reference-based metrics are not suitable for the generation
tasks, like PSNR and LPIPS [44]. Instead, we use two metrics,
blind/referenceless image spatial quality evaluator (BRISQUE)
[45] and natural
image quality evaluator (NIQE) [46], on
no-reference image qu... | Text2NeRF- Text-Driven 3D Scene Generation with Neural Radiance Fields |
2. Related Work
Monocular depth estimation is the task of estimating
depth values for each pixel of a single given RGB image.
Recent work has shown great performance in depth esti-
mation using deep learning models based on convolutional
neural networks [11, 12, 14, 22, 28, 29]. Later, attention-
based Transformer mod... | LDM3D- Latent Diffusion Model for 3D |
introduce a geometry-aware normal fusion algorithm that
extracts high-quality surfaces from the multi-view 2D rep-
resentations. Our extensive evaluations demonstrate that
our method achieves high-quality reconstruction results, ro-
bust generalization, and good efficiency compared to prior
works. | Wonder3D |
Mike Lewis, Yinhan Liu, Naman Goyal, Mar-
jan Ghazvininejad, Abdelrahman Mohamed, Omer
Levy, Veselin Stoyanov, and Luke Zettlemoyer.
2020. BART: Denoising sequence-to-sequence pre-
training for natural language generation, translation,
and comprehension. In Proceedings of the 58th An-
nual Meeting of the Association fo... | Prefix-Tuning |
f
o
r
G
P
T
-
2
,
w
e
f
i
n
d
a
n
a
v
e
r
a
g
e
s
c
o
r
e
o
f
0
.
1
5
1
u
s
i
n
g
t
o
p
-
a
n
d
-
r
a
n
d
o
m
s
c
o
r
i
n
g
,
a
n
d
0
.
0
3
7
f
o
r
r
a
n
d
o
m
-
o
n
l
y
s
c
o
r
i
n
g
.
S
c
o
r
e
s
g
e
n
e
r
a
l
l
y
d
e
c
r
e
a
s
e
w
h
e
n
g
o
i
n
g
t
o
l
a
t
e
r
l
a
y
e
r
... | Language models can explain neurons in language models |
Wei et al. (2022) proposed Manual-CoT, an approach that employs manually-crafted demonstrations
comprising the question, the reasoning process, and the final answer as prompts. In subsequent
work, Wang et al. (2022) introduced a novel decoding strategy "Self-Consistency", which generates
multiple answers from LLMs and a... | Enhancing Chain-of-Thoughts Prompting with Iterative Bootstrapping in Large Language Models |
eration process. We first collect a total of 2.2M
categories from Wikipedia. These categories are
filtered based on two requirements. Firstly, the
category must consist of less than three words. Sec-
ondly, the category must comprise more than 10
sub-categories and 50 pages. Upon manual inspec-
tion, we note that lengthy... | LaMini-LM- A Diverse Herd of Distilled Models from Large-Scale Instructions |
Formal verification techniques can prove the correctness of software (subject to assumptions).
Some degree of robustness to small modifications to the input can be proven for AI systems
using such techniques.126 But in general, there may be many differences in an input that
humans consider unimportant but that have... | Capabilities and risks from frontier AI |
C CONFIGURATION
This section shows the partial configuration files we used throughout the experiments.23 To make
things more succinct, we removed unnecessary parameters such as path names and pretraining ar-
guments. Listing 1 shows the configuration file for MultiHashEmbed.
23For more information about spaCy configuratio... | MULTI HASH EMBEDDINGS IN SPACY |
6 | Scaling Transformer to 1M tokens and beyond with RMT |
• Compute 𝑛@𝑘 with the re-sampled models and samples using the process described above.
model.
We perform this re-sampling and estimation process many times, and report the 95% confidence
interval as the 2.5th percentile and 97.5th percentile from the resulting set of estimates.
It’s difficult to use the sub-sampling ... | alphacode |
Figure 2b: Word Error Rates for the MoE teacher and student dense networks
Improving sample efficiency by 10x for a 10B parameter Language model using MoE
We evaluated MoE technique using ORT MoE implementation to research variants of language models for
machine translation. Two sets of multilingual data are used whic... | Scaling Speech, Language and Vision Models with Mixture of Experts Technique - Microsoft Community Hub |
both consistent and inconsistent renaming, suggesting that it does not model the different variables
well. As model size increases however, the relative performance drop observed with ill-formed inputs
becomes more and more pronounced, while sensitivity to consistent variable renaming decreases. This
suggests that as mo... | alphacode |
*Equal contribution 1Google Research 2IRCAM - Sorbonne
Universit´e (work done while interning at Google). Correspondence
to: Christian Frank <chfrank@google.com>. | MusicLM |
our results indicate that frozen language models are a viable
path towards general-purpose embodied multimodal models
that fully retain their language capabilities, we have also
surfaced an alternative route with unfrozen models: scaling
up the language model size leads to significantly less catas-
trophic forgetting wh... | PaLM-E- An Embodied Multimodal Language Model |
3.2 Participants | Use of bot and content flags to limit the spread of misinformation among social networks: a behavior and attitude survey |
equal to the norm of the embedding of the concept’s “super-
category” token (e.g., for the second example in Figure 4,
we set the norm equal to the norm of “teapot”). Formally,
the normalized output of the network is given by: | A Neural Space-Time Representation for Text-to-Image Personalization |
“Survival of the Fittest” [131] shows that if an individual wants to survive in the external environment,
he must adapt to the surroundings efficiently. This requires him to be cognitive, able to perceive
and respond to changes in the outside world, which is consistent with the definition of “agent”
mentioned in §2.1. ... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
One of the most exciting opportunities is how AI can deepen our understanding of information and turn it into
useful knowledge more efficiently — making it easier for people to get to the heart of what they’re looking for and
get things done. When people think of Google, they often think of turning to us for quick fact... | Google AI updates_ Bard and new AI features in Search |
While progress in 2D generative models of human ap-
pearance has been rapid, many applications require 3D
avatars that can be animated and rendered. Unfortunately,
most existing methods for learning generative models of
3D humans with diverse shape and appearance require 3D
training data, which is limited and expensive... | AG3D- Learning to Generate 3D Avatars from 2D Image Collections |
when you are coding?(EN) Maybe you need to enlist the help of someangry bees.(EN) A cup of deadline.(CN) 你觉得完成一篇深度学习论文辛苦吗？(JP) ああ、やっと運転が始められるね(JP) 一番絶望的だと思うことは何ですか？@ What's the most desperate thing you've ever heard?(JP) 月曜日だし、仕事に行く時間だね@ It's Monday and time to go to work.(JP) 犬が私のピザを食べました@ The dog ate my pizza.+ CLoT ... | Let’sThinkOutsidetheBox |
Tianyi Zhang, Varsha Kishore, Felix Wu, Kilian Q.
Weinberger, and Yoav Artzi. 2020b. BERTScore:
In Interna-
Evaluating text generation with bert.
tional Conference on Learning Representations.
Yizhe Zhang, Siqi Sun, Michel Galley, Yen-Chun Chen,
Chris Brockett, Xiang Gao, Jianfeng Gao, Jingjing
Liu, and Bill Dolan. 20... | Prefix-Tuning |
refers
to as
(2010)
This distinction between reinforcement seeking and challenge avoidance
becomes relevant as we focus on social networking sites, where most of the
political news that citizens consume is posted by their friends and family. The
process whereby social boundaries weaken in online environments is what... | Social_Media_and_Democracy |
[37] Jochen Huber, Roy Shilkrot, Pattie Maes, and Suranga Nanayakkara (Eds.). 2018. Assistive Augmentation. Springer Singapore.
[38] Masahiko Inami, Daisuke Uriu, Zendai Kashino, Shigeo Yoshida, Hiroto Saito, Azumi Maekawa, and Michiteru Kitazaki. 2022. Cyborgs,
Human Augmentation, Cybernetics, and JIZAI Body. In Augme... | Society’sAttitudesTowardsHumanAugmentation |
y
i
n
t
e
r
n
a
l
b
e
h
a
v
i
o
r
.
T
h
e
r
e
c
o
u
l
d
a
l
s
o
b
e
c
o
m
p
l
i
c
a
t
i
o
n
s
i
n
u
s
i
n
g
p
o
w
e
r
f
u
l
m
o
d
e
l
s
f
o
r
a
s
s
i
s
t
a
n
c
e
i
f
w
e
d
o
n
'
t
k
n
o
w
w
h
e
t
h
e
r
t
h
e
a
s
s
i
s
t
a
n
t
i
t
s
e
l
f
i
s
t
r
u
s
t
w
o
r
t
h
y
.
W
e
h
... | Language models can explain neurons in language models |
46
Gemini: A Family of Highly Capable Multimodal Models
9.3.1. Chart understanding and reasoning over data
Prompt
Spot a data point that stands out in these charts and what that implicates. Then produce a
detailed markdown table for all the data shown.
Model Response (rendered Markdown)
The data point that stands ... | gemini_1_report |
[26] Dylan Hadfield-Menell, Stuart J Russell, Pieter Abbeel, and Anca Dragan. Cooperative inverse reinforce-
ment learning. Advances in neural information processing systems, 29, 2016.
[27] Peter Henderson, Koustuv Sinha, Nicolas Angelard-Gontier, Nan Rosemary Ke, Genevieve Fried, Ryan
Lowe, and Joelle Pineau. Ethical... | CAMEL- Communicative Agents for “Mind” Exploration of Large Scale Language Model Society |
[449] Mehrdad Rafiepour and Javad Salimi Sartakhti. 2023. CTRAN: CNN-Transformer-based Network for Natural Language
Understanding. arXiv preprint arXiv:2303.10606 (2023).
[450] Tuomo Raitio, Ramya Rasipuram, and Dan Castellani. 2020. Controllable neural text-to-speech synthesis using
intuitive prosodic features. arX... | AReviewofDeepLearningTechniquesforSpeechProcessing |
In this example the writer’s statistic is arguably a bit misleading (presumably unintentionally), and the model’s
response may simply be superior. This strikes us as an example of a ‘softball question’ where we expect
models to do well, since it is common, vague, and open-ended:
Writer Prompt → Tell me all about frack... | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
9.4. Commonsense reasoning in a multilingual setting
Prompt
Explain what i should call my grandparents on each side of my family based on this chart
Model Response
On your father’s side, you would call your grandparents Yéye and Nˇainai. On your mother’s
side, you would call your grandparents Wàig¯ong and Wàipó.
Fig... | gemini_1_report |
From another perspective, quantization methods can be broadly classified into uniform and non-uniform approaches [86].
Uniform quantization, as explored in works like SPQR [67], GPTQ [79], and others [113, 129], involves dividing the range
of weights into equally sized bins. This method has become popular for its abili... | TheEfficiencySpectrumofLargeLanguageModels-AnAlgorithmicSurvey |
93.8
93.7
94.2
UNFAIR-ToS AVERAGE
64.2
66.5
71.5
79.9
80.9
81.5
42.0
45.4
46.2
21
Table 13: Prompting results on general LLM benchmarks. Raw trains on the raw texts, and
Read trains on the reading comprehension texts. Text Completion is more close to a question type
than a task type in general benchmarks, so we ... | ADAPTINGLARGELANGUAGEMODELSVIA READINGCOMPREHENSION |
instruction-tuned model built on the Flan mixture[4], which successfully harnesses the strengths of
both instruction-tuning and the sparse MoE technique. FLAN-MOE effectively and efficiently scales
up language models, without necessitating a rise in computational resources or memory requirements.
We subject our model, ... | Mixture-of-Experts |
Hyung Won Chung, Le Hou, Shayne Longpre, Barret
Zoph, Yi Tay, William Fedus, Yunxuan Li, Xuezhi
Wang, Mostafa Dehghani, Siddhartha Brahma, Al-
bert Webson, Shixiang Shane Gu, Zhuyun Dai,
Mirac Suzgun, Xinyun Chen, Aakanksha Chowdh-
ery, Alex Castro-Ros, Marie Pellat, Kevin Robinson,
Dasha Valter, Sharan Narang, Gaurav ... | AreEmergentAbilitiesinLarge Language Models just In-Context |
From Red Teaming Insights to Safer Models. Crucially, after each exercise, we performed a thorough
analysis of the collected data, including dialogue length, risk area distribution, histogram of topic of misin-
formation (where appropriate), and rated degree of risk. In each case, we took the overall lessons as a guide... | Llama2 |
degree.
Basic skills and knowledge required:
· Essential:
Solid mathematical ability and excellent programming skills.
A solid knowledge of Machine Learning and Computer Vision.
Prior expertise in working with any of these multimodality combinations: video,
video+language or video+audio understanding | Machine Learning for Long-Term Video Understanding at University of Bristol on FindAPhD.com |
20
Preprint
Alex Wang, Amanpreet Singh, Julian Michael, Felix Hill, Omer Levy, and Samuel R. Bowman.
GLUE: A Multi-Task Benchmark and Analysis Platform for Natural Language Understanding. In
International Conference on Learning Representations, September 2018. URL https://op
enreview.net/forum?id=rJ4km2R5t7.
Shibo ... | CRAMMING-TRAININGALANGUAGEMODELONA SINGLEGPUINONEDAY |
[38] Jun Hao Liew, Hanshu Yan, Jianfeng Zhang, Zhongcong Xu,
and Jiashi Feng. Magicedit: High-fidelity and temporally
coherent video editing. arXiv preprint arXiv:2308.14749,
2023. 2, 3, 12, 18
[39] Yaofang Liu, Xiaodong Cun, Xuebo Liu, Xintao Wang,
Yong Zhang, Haoxin Chen, Yang Liu, Tieyong Zeng, Ray-
mond Chan, and Y... | VideoPoet |
∗Equal contribution
Pre-print. Work in Progress. | CAMEL- Communicative Agents for “Mind” Exploration of Large Scale Language Model Society |
reward functions. In Robotics: Science and Systems (RSS), 2017.
[35] A. Saha, A. Pacchiano, and J. Lee. Dueling rl: Reinforcement learning with trajectory
In F. Ruiz, J. Dy, and J.-W. van de Meent, editors, Proceedings of The 26th
preferences.
International Conference on Artificial Intelligence and Statistics, volume ... | Direct Preference Optimization |
Based on the general knowledge and capabilities
learned in the pretraining stage, supervised fine-
tuning (SFT) is proposed to further improve LLMs
with instruction-following ability and alignment
with human expectations (Wei et al., 2021; Sanh
et al., 2022; Ouyang et al., 2022). Many efforts
have been made to construc... | DataManagementForLargeLanguageModels-ASurvey |
prompting reaches within 2% of the state of the
art (Appendix Table 2).
To better understand why chain-of-thought
prompting works, we manually examined model-
generated chains of thought by LaMDA 137B
for GSM8K. Of 50 random examples where the
model returned the correct final answer, all of
the generated chains of thoug... | Chain-of-Thought Prompting Elicits Reasoning in Large Language Models |
is my cousin. Therefore, the final answer is definitely yes.Input TextFlan-PaLM outputZero-shot chain-of-thought commonsense reasoningI will explain these jokes: (1) The problem with kleptomaniacs is that they always take things literally. Explanation: This joke is wordplay. Someone who "takes things literally" is some... | Scaling Instruction-Finetuned Language Models |
An interesting observation is that in Figure 7, even though the completions are generated by different models,
they begin in a very similar way (all completions have to do with a little girl coming by and talking to the pumpkin).
We point out that the reason for this seems to be that the completions are generated with ... | TinyStories-HowSmallCanLanguageModelsBeandStillSpeak CoherentEnglish? |
es-en | fr-en | it-en
Clotho
Industrial Data
CochlScene
TUT2017
Meld
ClothoAQA
VocalSound
NS. Qualities
NS. Instrument
SpeechT5 (Ao et al., 2021)
SpeechNet (Chen et al., 2021)
SLM-FT (Wang et al., 2023b)
SALMONN (Anonymous, 2023)
Qwen-Audio
MMSpeech-base (Zhou et al., 2022)
MMSpeech-large (Zhou et al., 2022... | Qwen-Audio |
To enhance the performance of continuous speech recognition, LSTM networks have been
utilized in hybrid architectures alongside CNNs [417]. The CNNs extract local features from speech
frames that are then processed by LSTMs over time [417]. LSTMs have also been employed for
speech synthesis, where they have been shown ... | AReviewofDeepLearningTechniquesforSpeechProcessing |
•
•
•
• Methodology
• How the research will be communicated to the wider community
•
•
•
The supervisory provision as well as specialist and transferable skills training
Ethical considerations
Summary and conclusions
Writing the proposal
When writing your proposal, bear in mind that individuals reviewing your... | research proposal guidance |
Phenaki does not perform generation purely in latent space due to the design choices we made to
take advantage of text-image and text-video datasets during training. The first wz ⇥ hz tokens must
be single image (space only) tokens and the next tz ⇥ wz ⇥ hz tokens must be video (space-time)
tokens. Therefore, MaskGIT wi... | PHENAKI- VARIABLE LENGTH VIDEO GENERATION FROM OPEN DOMAIN TEXTUAL DESCRIPTIONS |
Arizona State University. Available from: https://www.jstor.org/stable/27092030.
[24] Xu W. Toward human-centered AI: a perspective from human-computer interaction. Interactions. 2019
Jun;26(4):42-6. Available from: https://dl.acm.org/doi/10.1145/3328485.
[25] Ghunaim H, Weekes TR, Eskridge TC. Designing an AI Assis... | DevelopingTeamDesignPatternsfor HybridIntelligenceSystems |
eration systems, followed by a description of music generation for video.
In
this section, we do not aim to give an exhaustive overview of music generation
systems, for this, the user is referred to Herremans et al. (2017), Civit et al.
(2022), and Briot et al. (2020).
2.1. Transformer-based Music Generation
Patt... | Video2Music |
Large language models (LLMs) are trained to absorb the myriad of patterns that are woven into the structure
of language. They not only exhibit various out-of-the-box capabilities such as generating chains of reasoning
[1, 2], solving logic problems [3, 4], and completing math puzzles [5], but also have been applied in ... | LargeLanguageModelsasGeneralPatternMachines |
study out-of-distribution behaviors. Full details about these
datasets can be found in Appendix A.
Our main findings are summarized in Figure 2 and Table 2.
Although the best zero-shot Whisper model has a relatively
unremarkable LibriSpeech clean-test WER of 2.5, which
is roughly the performance of modern supervised bas... | RobustSpeechRecognitionviaLarge-ScaleWeakSupervision |
Trainable text-speech alignment using kaldi. In Interspeech, 2017.
T. Nguyen, E. Kharitonov, J. Copet, Y. Adi, W.-N. Hsu, A. M. Elkahky, P. Tomasello, R. Algayres,
B. Sagot, A. Mohamed, and E. Dupoux. Generative spoken dialogue language modeling. Transac-
tions of the Association for Computational Linguistics, 11:250–... | Voicebox-Text-GuidedMultilingual UniversalSpeechGenerationatScale |
Announcing Jurassic-2 and Task-Specific APIs
https://www.ai21.com/blog/introducing-j2
7/12 | Announcing Jurassic-2 and Task-Specific APIs |
84
Mehrish et al.
[43] Herve A Bourlard and Nelson Morgan. 1994. Connectionist speech recognition: a hybrid approach. Vol. 247. Springer
[44] Pierre-Michel Bousquet and Mickael Rouvier. 2019. On robustness of unsupervised domain adaptation for speaker
[45] Hervé Bredin and Antoine Laurent. 2021. End-to-end speaker ... | AReviewofDeepLearningTechniquesforSpeechProcessing |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.