text stringlengths 1 1k ⌀ | title stringclasses 230
values |
|---|---|
6.4
41.6 37.2 31.6 33.2 21.2 24.7 16.4
60.4 54.0 50.8 34.0 39.0 39.0 58.8
58.0 58.0 36.8 18.8 25.3 19.9 18.0
58.4 55.6 30.0 24.8 26.7 30.1 28.4
56.4 55.2 41.6 50.4 24.0 37.0 17.2
60.4 49.2 50.4 51.2 37.0 49.3 50.4
19.6 62.4 79.6 51.2 83.2 44.5 65.1 38.0
29.6 68.4 78.0 54.0 88.8 55.5 72.6 66.4
0.0
0.0
0.0
0.0
59.2 5... | Mixture-of-Experts |
1
Introduction
The rapid evolution of large language models
(LLMs) makes them a game changer for mod-
ern natural language processing. LLMs’ domi-
nating generation ability changes previous tasks’
paradigms to a unified text generation task and con-
sistently improves LLMs’ performance on these
tasks (Raffel et al., 2... | Multi-step Jailbreaking Privacy Attacks on ChatGPT |
A Review of Deep Learning Techniques for Speech Processing
9
valuable information from vast amounts of speech data. In this section, we delve into the applications
of deep learning architectures in speech processing tasks, exploring their potential, advancements,
and the impact they have had on the field. By examinin... | AReviewofDeepLearningTechniquesforSpeechProcessing |
Yonghui Wu, Mike Schuster, Zhifeng Chen, Quoc V. Le, Mohammad Norouzi, Wolfgang Macherey,
Maxim Krikun, Yuan Cao, Qin Gao, Klaus Macherey, Jeff Klingner, Apurva Shah, Melvin John-
son, Xiaobing Liu, Łukasz Kaiser, Stephan Gouws, Yoshikiyo Kato, Taku Kudo, Hideto Kazawa,
Keith Stevens, George Kurian, Nishant Patil, Wei ... | CRAMMING-TRAININGALANGUAGEMODELONA SINGLEGPUINONEDAY |
6
QLoRA-AllQLoRA-FFNQLoRA-AttentionAlpaca (ours)Stanford-AlpacaModel6061626364RougeLbits41610101011Total model bits0.600.610.620.630.640.650.660.67Mean zeroshot accuracy4-bit LLaMAFloatNFloatNFloat + DQData type Table 3: Experiments comparing 16-bit BrainFloat (BF16), 8-bit Integer (Int8), 4-bit Float (FP4), and 4-
bi... | QLORA |
(
2
)
M
o
d
e
l
s
e
l
e
c
t
i
o
n
:
L
L
M
d
i
s
t
r
i
b
u
t
e
s
t
h
e
t
a
s
k
s
t
o
e
x
p
e
r
t
m
o
d
e
l
s
,
w
h
e
r
e
t
h
e
r
e
q
u
e
s
t
i
s
f
r
a
m
e
d
a
s
a
m
u
l
t
i
p
l
e
-
c
h
o
i
c
e
q
u
e
s
t
i
o
n
.
L
L
M
i
s
p
r
e
s
e
n
t
e
d
w
i
t
h
a
l
i
s
t
o
f
m
o
d
e
l
... | LLM Powered Autonomous Agents _ Lil'Log |
[4] Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert,
Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov,
Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, et al.
Llama 2: Open foundation and fine-tuned chat models.
arXiv preprint arXiv:2307.09288, 2023. 2, 3, 7
[5] Jinze Bai, Shuai Bai, Yunfei Chu, Zeyu Cui, Kai ... | Let’sThinkOutsidetheBox |
Hugo Laurenc¸on, Lucile Saulnier, Thomas Wang, Christopher Akiki, Albert Villanova del Moral,
Teven Le Scao, Leandro Von Werra, Chenghao Mou, Eduardo Gonz´alez Ponferrada, Huu Nguyen,
et al. The BigScience corpus: A 1.6 TB composite multilingual dataset. 2022.
Yuhang Li, Ruihao Gong, Xu Tan, Yang Yang, Peng Hu, Qi Zha... | GPTQ |
We use three models for extracting audio representations that
will serve for conditional autoregressive music generation,
which are illustrated in Figure 1. In particular, by following
the approach of AudioLM, we use the self-supervised audio
representations of SoundStream (Zeghidour et al., 2022), as
acoustic tokens t... | MusicLM |
Making Slides | Tool Learning with Foundation Models |
To avoid any potential confusion, we adapt the
terminology proposed by Wei et al. (2022b) and
shall refer to these techniques as “prompting tech-
niques”. It is noteworthy that a single prompting
technique can be adaptable across multiple tasks.
For example, in-context learning can be used in
performing any task throug... | AreEmergentAbilitiesinLarge Language Models just In-Context |
the N.Y. Regents Science Exams: An Overview of the Aristo Project. cs.CL.
Cranmer, M. D., Xu, R., Battaglia, P., & Ho, S. (2019). Learning Symbolic Physics with Graph Networks.
Cropper, A., Morel, R., & Muggleton, S. (2019). Learning higher-order logic programs. Machine Learning,
D’Avila Garcez, A. S., Lamb, L. C... | The Next Decade in AI- |
1√αt
(cid:19)
xt − βt√1− ¯αt
(cid:18)
=
xt,
1
√¯αt
1
√αt
µθ(xt, t) = ˜µt
(11)
where (cid:15)θ is a function approximator intended to predict (cid:15) from xt. To sample xt−1 ∼ pθ(xt−1|xt) is
to compute xt−1 = 1√αt
+ σtz, where z ∼ N (0, I). The complete sampling
procedure, Algorithm 2, resembles Langevin dynami... | Denoising Diffusion Probabilistic Models |
2
Cerebras-GPT: Open Compute-Optimal Language Models
2.1 Model Architecture
Cerebras-GPT models have a GPT-3-like architecture, an autoregressive transformer decoder model (Brown
et al., 2020). The main difference is that unlike GPT-3, which uses alternating dense and sparse-banded
attention, we use dense attention ... | Cerebras-GPT- Open Compute-Optimal Language Models Trained on the Cerebras Wafer-Scale Cluster |
For comparison, we implemented MultiEmbed that is the same as MultiHashEmbed, but us-
ing regular lookup instead of the hashing trick. Both embedding layers use the NORM, PREFIX,
SUFFIX and SHAPE features; for MultiHashEmbed we use 5000, 2500, 2500 and 2500 rows for
each table respectively and for MultiEmbed we use the... | MULTI HASH EMBEDDINGS IN SPACY |
20/11/2023, 08:29
Job details
Werken bij TU Delft
Werken bij TU DelftVacaturesJob details
Vacatures
Wetenschapper
PhD
Academic Career Track
Postdoc
Professional
PhD position in Grounding Large Language
Models in the Real World
Apply Now (https://emea3.recruitmentplatform.com/apply-app/pages/application-
... | Job details - TU |
2 | StarCoder_paper (1) |
sha1_base64="by2EXrk8ymnCHE/bC17V3YYH0CU=">AAAB7XicbVDLSgNBEOyNrxhfqx69DAbBU9gVQY8BLx4jmIckS5idzCZj5rHMzAphyT948aCIV//Hm3/jJNmDJhY0FFXddHfFKWfGBsG3V1pb39jcKm9Xdnb39g/8w6OWUZkmtEkUV7oTY0M5k7RpmeW0k2qKRcxpOx7fzPz2E9WGKXlvJymNBB5KljCCrZNaPcOGAvf9alAL5kCrJCxIFQo0+v5Xb6BIJqi0hGNjumGQ2ijH2jLC6bTSywxNMRnjIe06KrGgJsrn107RmVMGK... | BANMo- Building Animatable 3D Neural Models from Many Casual Videos |
4
HSK7-9Writing(Chinese)HSK7-9Overall(Chinese)J-TestA-COverall(Japanese)PLIDAC2Writing(Italian)PLIDAC2Overall(Italian)TCFWriting(French)TCFOverall(French)DELEC2Writing(Spanish)DELEC2Overall(Spanish)Goethe-ZertifikatC2Writing(German)Goethe-ZertifikatC2Overall(German)0102030405060708090100PassFailPass*Fail*PassPassPassFai... | PaLM 2 Technical Report |
Wider and deeper llm networks are fairer llm evaluators. arXiv preprint arXiv:2308.01862 (2023).
[239] Yue Zhang, Yafu Li, Leyang Cui, Deng Cai, Lemao Liu, Tingchen Fu, Xinting Huang, Enbo Zhao, Yu Zhang, Yulong
Chen, Longyue Wang, Anh Tuan Luu, Wei Bi, Freda Shi, and Shuming Shi. 2023. Siren’s Song in the AI Ocean: A... | ASurveyonEvaluationofLargeLanguageModels |
the context of the raw text. Furthermore, we augment the reading comprehension texts with diverse
general instructions, thereby further enhancing prompting ability (Wei et al., 2022; Zhou et al., 2023; | ADAPTINGLARGELANGUAGEMODELSVIA READINGCOMPREHENSION |
A major challenge that has prevented past efforts of self-learning in language models from succeed-
ing, especially in arithmetic, is a phenomenon that we call error avalanching. During self-training,
when all training data is generated by the model itself, there is no guarantee that the data is cor-
rect. Error avalan... | CHAIN-OF-THOUGHTREASONING IS APOLICY IMPROVEMENTOPERATOR |
natural language.
Reasoning. cs.AI.
Zhang, R., Wu, J., Zhang, C., Freeman, W. T., & Tenenbaum, J. B. (2016). A Comparative Evaluation of
Approximate Probabilistic Simulation and Deep Neural Networks as Accounts of Human Physical
Scene Understanding. arXiv, 1605.01138v2.
59 | The Next Decade in AI- |
In this technical report, we described the efforts of the BigCode community in creating StarCoderBase
and StarCoder, an open-access 15.5B parameter large language model (LLM) trained on code. We
provided full transparency on all aspects of the research and development process, including the
training data, the data cura... | StarCoder_paper (1) |
[30] Yuan-Ting Hu, Hong-Shuo Chen, Kexin Hui, Jia-Bin Huang,
and Alexander G. Schwing. SAIL-VOS: Semantic amodal
instance level video object segmentation – a synthetic dataset
and baselines. In CVPR, 2019.
[31] Chun-Hao P. Huang, Hongwei Yi, Markus H¨oschle, Matvey
Safroshkin, Tsvetelina Alexiadis, Senya Polikovsky, D... | Learning 3D Human Pose Estimation from Dozens of Datasets using a Geometry-Aware Autoencoder to Bridge Between Skeleton Formats |
When composing the test samples, we also make
efforts to ensure comprehensive coverage. We first
read through the text samples in the dataset and
then compose samples that are reasonable music
descriptions but do not exist in the data. The text
prompts comprehensively cover all genres that we
evaluate and incorporate e... | MOUSAI |
3
2
0
2
r
p
A
4
2
]
V
C
.
s
c
[
3
v
2
8
0
1
1
.
1
1
2
2
:
v
i
X
r
a
Figure 1. Recent methods for synthesizing novel views from monocular videos of dynamic scenes–like HyperNeRF [50] and NSFF [35]–
struggle to render high-quality views from long videos featuring complex camera and scene motion. We pr... | DynIBaR-NeuralDynamicImage-BasedRendering |
better than other methods for questions with multiple hops like 7-hop (with an average improvement
of 11.7%) and 8-hop (with an average improvement of 8.3%). Moreover, the effect of Iter-CoT using
exemplars produced after four iterations of bootstrapping is considerably superior to that after a single
iteration (with a... | Enhancing Chain-of-Thoughts Prompting with Iterative Bootstrapping in Large Language Models |
11
Manuscript submitted to ACM, 2023,
Draxler et al.
drivers not willing to rely on assistance systems [34] or computer users sticking to basic editors such as vim and emacs8,
valuing principles and habits over assistance. To satisfy the need for autonomy, we suggest clarifying that using LLMs
does not mean giving ... | Adoptionand AppropriationofLLMs |
Limitations of self-supervised learners for localization. SSL approaches which
rely on augmented views or jigsaw transformations, such as MoCo [He et al., 2020b] and
PIRL [Misra and Maaten, 2020], learn occlusion invariance since they are trained with
random crops on ImageNet where foreground objects are often large so... | A Cookbook of Self-Supervised Learning |
Seq2seq models have been widely used in speech processing, initially based on RNNs. However,
RNNs face the challenge of processing long sequences, which can lead to the loss of the initial
context by the end of the sequence [244]. To overcome this limitation, the transformer architecture
has emerged, leveraging self-at... | AReviewofDeepLearningTechniquesforSpeechProcessing |
t
o
l
e
a
r
n
i
n
g
t
h
e
n
e
w
p
r
o
m
p
t
s
u
s
e
r
s
c
o
m
e
u
p
w
i
t
h
,
a
n
d
i
n
t
u
r
n
,
fi
g
u
r
i
n
g
o
u
t
m
e
t
h
o
d
s
t
o
p
r
e
v
e
n
t
B
a
r
d
f
r
o
m
o
u
t
p
u
tt
i
n
g
p
r
o
b
l
e
m
a
t
i
c
o
r
s
e
n
s
i
t
i
v
e
i
n
f
o
r
m
a
t
i
o
n
.
A
n
d
a
l
t
h
o
... | An overview of Bard- an early experiment with generative AI |
69.2
18.4
-15.0%
-87.3%
-4.3%
-61.6%
NLG
NLG
NLG
NLG
NLU
NLU
NLU
NLU
NLU
NLU
NLU
NLU
NLU
NLU
NLU
NLU
NLU
NLU
NLU
NLU
NLU
81.4
29.3
22.6
81.8
83.6
86.1
87.5
83.7
69.3
52.1
83.9
85.0
60.1
53.6
88.7
91.0
78.7
63.2
86.3
92.8
83.9
78.2
53.8
74.6
27.2
21.8
83.3
83.5
86.3
89.0
83.0
70.3
52.8
84.9
86.3
62.6
55.8
89.4
93.... | PaLM-E- An Embodied Multimodal Language Model |
voxceleb speaker recognition challenge 2019. arXiv preprint arXiv:1910.12592 (2019).
[635] Jihen Zeremdini, Mohamed Anouar Ben Messaoud, and Aicha Bouzid. 2015. A comparison of several computational
auditory scene analysis (CASA) techniques for monaural speech segregation. Brain informatics 2 (2015), 155–166.
[636] Al... | AReviewofDeepLearningTechniquesforSpeechProcessing |
to guide retrieval,
4 Retriever
In the context of RAG, the ”R” stands for retrieval, serving
the role in the RAG pipeline of retrieving the top-k relevant
documents from a vast knowledge base. However, crafting
a high-quality retriever is a non-trivial task. In this chapter,
we organize our discussions around three ke... | Retrieval-AugmentedGenerationforLargeLanguageModels-ASurvey |
t
i
s
,
w
e
w
a
n
t
t
o
c
o
m
p
a
r
e
t
w
o
l
i
s
t
s
o
f
v
a
l
u
e
s
:
t
h
e
s
i
m
u
l
a
t
e
d
a
c
t
i
v
a
t
i
o
n
v
a
l
u
e
s
f
o
r
t
h
e
e
x
p
l
a
n
a
t
i
o
n
o
v
e
r
m
u
l
t
i
p
l
e
t
e
x
t
e
x
c
e
r
p
t
s
,
a
n
d
t
h
e
a
c
t
u
a
l
a
c
t
i
v
a
t
i
o
n
v
a
l
u
e
s
o
... | Language models can explain neurons in language models |
= (G1 ⊗ G2) ⊗ . . . ⊗ Gp = G A
⊗ Gvk ) = L(G(F )). We can thus define a label relation Ri ⊆ L(G A
Let op1, . . . , opm be a sequence of merge/shrink/reduce labels operations. Let (cid:9)0 = {Gv1 , . . . , Gvn
= (cid:3)S A
i , E A
i | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
two potential completions (Bisk et al., 2020). For example
[Goal] Make an outdoor pillow
[Sol1] Blow into a tin can and tie with rubber band
[Sol2] Blow into a trash bag and tie with rubber band
The model must choose which of the two continuations is more likely to follow from the prompt.
Human performance on this da... | Cerebras-GPT- Open Compute-Optimal Language Models Trained on the Cerebras Wafer-Scale Cluster |
foroneenvironment(halfcheetah)itleadstoaverynegativescore,whichisworsethantheinitialrandompolicy.algorithmavg.normalizedscoreNoclippingorpenalty-0.39Clipping,(cid:15)=0.10.76Clipping,(cid:15)=0.20.82Clipping,(cid:15)=0.30.70AdaptiveKLdtarg=0.0030.68AdaptiveKLdtarg=0.010.74AdaptiveKLdtarg=0.030.71FixedKL,β=0.30.62FixedK... | PPO |
using a metric space for the outputs (0–100) where
the LLM has priors over the scale of the different to-
kens. These functions also contain a “meta-pattern”:
the y-values increase, decrease, and then increase in
a single period—and the amplitude of the function
also increases over time. This is a form of least-to-most... | LargeLanguageModelsasGeneralPatternMachines |
Gustavo Sandoval, Hammond Pearce, Teo Nys, Ramesh Karri, Siddharth Garg, and Brendan Dolan-
Gavitt. Lost at C: A user study on the security implications of large language model code assistants,
2023. (cited on p. 32)
Teven Le Scao, Angela Fan, Christopher Akiki, Ellie Pavlick, Suzana Ili´c, Daniel Hesslow, Roman
Casta... | StarCoder_paper (1) |
Yet we also live in a time when a whole host of factors outside of the
academy can have huge effects on the degree to which scholars can access
these data. These factors include, but are not limited to, policy decisions by
government authorities such as the US FTC and the European Data
Protection Board and internal bus... | Social_Media_and_Democracy |
Generating an entire program in a general-purpose programming language such as C++ or Python,
starting from a long natural language task description, has remained an open problem. The difference
in difficulty between generating short code snippets and entire programs can be analogous to that of
imperative versus declarati... | alphacode |
Meeting the specific challenge of political disinformation with a wholesale
repeal of CDA 230 is unwarranted given the broader negative impacts that may
result. To that end, the central question is one of tailoring: Precisely what kind
of acts should be targeted by the crafting of an exception to CDA 230? Will the
attri... | Social_Media_and_Democracy |
Faithfulness specific datasets can be better than NLI datasets because entailment or neutral
labels of NLI datasets and faithfulness are not equivalent. For example, the hypothesis “Putin is U.S.
president” can be considered to be either neutral to or entailed from the premise “Putin is president”.
However, from the fa... | SurveyofHallucinationinNatural Language Generation |
losspretrain = lossent + lossctx + lossfact + lossans
A.2.2 Finetuning on Question Answering
In the Open-domain Question Answering task,
questions are posed in natural
language, e.g.
“Where was Charles Darwin born?”, and answered
by a sequence of tokens, e.g. “United Kingdom”.
In this paper, we focus on a subset of op... | Adaptable and Interpretable Neural Memory Over Symbolic Knowledge |
being a learned affine transformation.
Object-centric representations. Unlike language, visual
input is not pre-structured into meaningful entities and rela-
tionships: while ViT may capture semantics, the structure of
the representation resembles a static grid rather than a col-
lection of object instances. This poses ... | PaLM-E- An Embodied Multimodal Language Model |
Thus, advertising revenues that in the twentieth century helped fund content
creation for a mass public increasingly help fund the provision of platform
products and services to individual users and are tied in with pervasive data
collection, especially by the dominant technology companies (Turow and
Couldry 2018).2 | Social_Media_and_Democracy |
namically re-plans based on its current inventory and crafts
a new one. However, VPT-RL exhibits perplexing behaviors
at this stage by using inappropriate tools for mining stone or
crafting unnecessary items. This comparison demonstrates
that JARVIS-1 possesses superior generalization and plan-
ning abilities for long-... | JARVIS-1 |
Table 6: Examples of image classification using BiomedGPT with different model scales.
Dataset
TissueMNIST
OrganCMNIST
ChestMNIST
Small Medium Base Large
69.7
36.4
92.2
93.3
89.2
89.2
36.4
92.3
89.2
53.2
93.1
89.2
B.2 Classification with High-resolution Images
In previous evaluations, we showcased BiomedGPT’s pro... | BiomedGPT |
05101520253035010100250AdapterH rSeq Len = 128Seq Len = 256Seq Len = 51212481632Batch Size010100250AdapterL r12481632Batch Size12481632Batch Size and STS-B (textual similarity, Cer et al. (2017)). The broad coverage makes GLUE benchmark a
standard metric to evaluate NLU models such as RoBERTa and DeBERTa. The individua... | LORA |
5.3.2 Results of Shaping a Single LLM Personality Domain
This study tested if LLM-simulated Big Five personality traits can be independently shaped
at nine levels.
27 | PersonalityTraitsinLargeLanguageModels |
the use of convolutional neural networks (CNNs)
Computational Analysis of Ageing Brains
Supervisors: Dr Kathleen Steinhofel & Professor Zoran Cvetkovic
The ability to acquire and store information is a key function of the brain. This ability is affected by
ageing and in various age-related disease, including ... | informatics-phd-projects-2022-23 |
AI improves the effectiveness of existing techniques. AI-enhanced social engineering is
already being used by cybercriminals to conduct scams and steal login credentials, with
systems that can gather intelligence on targets,214 impersonate voices of trusted contacts,215
and generate persuasive spear phishing message... | Capabilities and risks from frontier AI |
E1 - Friendliness
E2 - Gregariousness
E2 - Gregariousness
E3 - Assertiveness
E3 - Assertiveness
E4 - Activity Level
E5 - Excitement-Seeking
E5 - Excitement-Seeking
E6 - Cheerfulness
A1 - Trust
A2 - Morality
A2 - Morality
A3 - Altruism
A3 - Altruism
A3 - Altruism
A4 - Cooperation
A5 - Modesty
A6 - Sympathy
AGR
AGR
C1 - ... | PersonalityTraitsinLargeLanguageModels |
EAE’s average accuracy is similar to BERT-
large. However, the LAMA sub-task accuracies
show that the two models are complementary. Men-
tion focused approaches are much better than the
BERT baselines at predicting the mention like
words in the SQuAD and T-REx probes, but they
are marginally worse for the RE probe and ... | Entities as Experts- Sparse Memory Access with Entity Supervision |
(1) 终于知道为什么他俩不说话了。 @ Finally figured out why those two aren't barking. (2) 你俩是不是又把家具都拆了。 @ Did you two go and take apart all the furniture again? (3) 不让你们吃巧克力，就只能这样了。 @ If I can't let you have chocolate, this is the only option. (4) 最近又胖了。 @ We've packed on some pounds again lately. (5) 说好的一起藏猫猫呢？ @ We're all hidden... | Let’sThinkOutsidetheBox |
First, we need to calculate how many sets of 4 yogurts Terry buys in 30
days. Since Terry eats 2 yogurts a day, he will need 2/4 = 0.5 sets of 4
yogurts per day. Next, we multiply the number of sets by the price of
each set to calculate how much Terry spends per day. Each set of 4
yogurts costs $5.00, so Terry spends 0... | LARGELANGUAGEMODELSCANNOTSELF-CORRECT REASONINGYET |
III. METHOD
We propose a text-driven 3D scene generation framework
to progressively generate 3D scenes according to given text
prompts as shown in Fig. 2. We first generate an initial
view by a text-to-image diffusion model. Based on the initial
image, we build the support views and corresponding depth
3
maps as the ... | Text2NeRF- Text-Driven 3D Scene Generation with Neural Radiance Fields |
there is
At the same time as researchers are developing ways to protect user privacy in
social media datasets, the platforms themselves are moving in directions that
might make collection of most user data impossible. After Mark Zuckerberg
declared in early 2019 that “the future is private,” Facebook announced its pla... | Social_Media_and_Democracy |
Robin Jia and Percy Liang. 2017. Adversarial ex-
amples for evaluating reading comprehension
systems. In Proceedings of the 2017 Conference
on Empirical Methods in Natural Language
Processing, pages 2021–2031, Copenhagen,
Denmark. Association for Computational Lin-
guistics. https://doi.org/10.18653/v1
/D17-1215
Isaac... | ProoFVer- Natural Logic Theorem Proving for Fact Verification |
(cid:26)16 if l = 1
32 if l = 2
ϕ(l) =
A more general KD training objective is then a weighted sum of the KL, PL and MSE terms:
LKD = αKLLKL + αP LLP L + αM SELM SE | DISTIL-WHISPER |
Given our improvements to training stability, fine-tuning and model design, we start by validating
a sparse model approximately FLOP-matched to T5-Large (Raffel et al., 2019). We conclude this
section by designing and training a 269B sparse parameter model (FLOP matched to a 32B dense
model) which achieves state-of-the-... | ST-MOE- DESIGNING STABLE AND TRANSFERABLE SPARSE EXPERT MODELS |
A. Stuhlm¨uller and J. Byun. Supervise process, not outcomes. https://ought.
org/updates/2022-04-06-process, 2022.
J. Uesato, N. Kushman, R. Kumar, F. Song, N. Siegel, L. Wang, A. Creswell,
G. Irving, and I. Higgins. Solving math word problems with process-and
outcome-based feedback. arXiv preprint arXiv:2211.14275, ... | Let’s Verify Step by Step |
[30] Guanglei Yang, Hao Tang, Mingli Ding, Nicu Sebe, and
Elisa Ricci.
Transformer-based attention networks for
In Proceedings of the
continuous pixel-wise prediction.
IEEE/CVF International Conference on Computer Vision
(ICCV), pages 16269–16279, October 2021. 2
[31] Lvmin Zhang and Maneesh Agrawala. Adding conditio... | LDM3D- Latent Diffusion Model for 3D |
These problems are directly motivated by bioinformatics applications, such as studying genetic
mutations; DNA sequence analysis of antibodies and identification of "hairpins" that occur in DNA
sequences in Tuberculosis and HIV virus strains, respectively. However, they are also closely
related to pattern discovery t... | informatics-phd-projects-2022-23 |
Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021. 7, 1
[6] Adrian Bulat and Georgios Tzimiropoulos. How far are we from solving the 2d & 3d face alignment problem? (and a dataset of
230,000 3d facial landmarks). In International Conference on Computer Vision, 2017. 6, 5
[7] Chen Cao, Yan... | I M Avatar- Implicit Morphable Head Avatars from Videos |
Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High-resolution
image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF Conference on Computer
Vision and Pattern Recognition (CVPR), pp. 10684–10695, 2022.
Kevin Roose.
deeply
https://www.nytimes.com/2023/02/16/tec... | Tool Learning with Foundation Models |
The United States has typically relied much more heavily on industry self-
regulation than have European democracies, and this tradition has carried on
into the digital age. Neither the Federal Communications Commission (FCC)
nor any other federal regulators have sought to lay down formal rules as to the
kinds of conte... | Social_Media_and_Democracy |
3 BACKGROUND
Consider the binary classification setting with training data
D = {(xi, yi)}n
i=1, where xi ∈ X ⊂ Rd and yi ∈ Y =
{0, 1}. Samples are independent and identically distributed
according to some fixed but unknown distribution P with
density p. The classic RF algorithm takes B bootstrap sam-
Watson, Blesch, K... | Adversarial Random Forests for Density Estimation and Generative Modeling |
[Radford et al., 2018] Alec Radford, Karthik Narasimhan,
Tim Salimans, Ilya Sutskever, et al. Improving language
understanding by generative pre-training. OpenAI, 2018.
[Thoppilan et al., 2022] Romal Thoppilan, Daniel De Fre-
itas, Jamie Hall, Noam Shazeer, Apoorv Kulshreshtha,
Heng-Tze Cheng, Alicia Jin, Taylor Bos, L... | FinGPT-Open-SourceFinancialLargeLanguageModels |
Positional Encoding.
In Figure 22, we provide a visual-
ization of various outputs returned by our positional encod-
ing function. As can be seen, the blue and green curves
are well-separated due to their different layer indices. Con-
versely, the green and red curves share a similar encoding
as they both share the sam... | A Neural Space-Time Representation for Text-to-Image Personalization |
Be clear, objective, succinct and realistic in your objectives
Ask yourself why this research should be funded and/or why you are the best person to undertake this project
Ask yourself why this research is important and/or timely
State and justify your objectives clearly (“because it is interesting” is not enough!)
co... | research proposal guidance |
5. Conclusion and Discussion
In this paper, we argue that a LLM itself has the inherent
ability to handle long sequences and it should be able to ex-
tend the context window size without any fine-tuning. Based
on this belief, in a fine-tuning-free way, we propose Self-
Extend to elicit the inherent long context abiliti... | Self-Extend LLM |
9 | BiomedGPT |
as PaLM (Chowdhery et al., 2022), Chinchilla (Hoffmann et al., 2022), Gopher (Rae et al., 2021), GPT-4
(OpenAI, 2023), and Llama (Touvron et al., 2023a;b). In parallel, models specifically trained or fine-tuned for
code understanding and program synthesis from natural language prompts emerged with LLMs such as Codex
(C... | CodeLlama2 |
COG [Vaze et al., 2021]is a text generation model that for-
malizes its generation process by gradually copying text frag-
ments (such as words or phrases) from an existing collection
of text. Unlike traditional text generation models that select
words sequentially, COG utilizes efficient vector search tools
to calcula... | Retrieval-AugmentedGenerationforLargeLanguageModels-ASurvey |
(cid:16)
δF = −(wQ − quant(wQ))([H−1
F ]QQ)−1(H−1
F ):,Q,
H−1−Q =
H−1 − H−1
:,Q([H−1]QQ)−1H−1
Q,:
(cid:17)
.
−Q
(4)
(5) | GPTQ |
20
3
4
2
2
13
2
2
2
9
18
3
4
2
157
4
2
Reference
(Lang, 1995)
crowdflower.com
crowdflower.com
crowdflower.com
crowdflower.com
crowdflower.com
crowdflower.com
crowdflower.com
crowdflower.com
crowdflower.com
crowdflower.com
crowdflower.com
crowdflower.com
crowdflower.com
catalog.data.gov
(Lichman, 2013)
(Almeida et al., 2011)
1903
... | Parameter-Efficient Transfer Learning for NLP |
(used for cross-attention), and Venc represents the value matrix derived from the
Encoder’s hidden states. This cross-attention mechanism enables the Decoder
to focus on relevant information from the input video features while generat-
ing the next chord event, thus facilitating the modeling of music events and
dep... | Video2Music |
https://doi.org/10.1017/9781108890960 Published online by Cambridge University Press
216
Francis Fukuyama & Andrew Grotto
this reassessment (Wu 2018; Khan 2018). The first is to broaden the courts’
understanding of potential harms arising from excessive concentration of
power in the hands of a small number of privat... | Social_Media_and_Democracy |
19
• LibriSpeech (Panayotov et al., 2015): We used the test-clean and test-other splits from the LibriSpeech ASR corpus.
• TED-LIUM 3 (Hernandez et al., 2018): We used the test split of TED-LIUM Release 3, using the segmented manual
transcripts included in the release.
• Common Voice 5.1 (Ardila et al., 2019): We d... | RobustSpeechRecognitionviaLarge-ScaleWeakSupervision |
preprint arXiv:2212.10403, 2022.
challenges. Springer Nature, 2019.
[17] Lars Kotthoff, Chris Thornton, Holger H Hoos, Frank Hutter, and Kevin Leyton-Brown. Auto-weka:
Automatic model selection and hyperparameter optimization in weka. Automated machine learning:
methods, systems, challenges, pages 81–95, 2019.
[18] ... | MLCopilot- Unleashing the Power of Large Language Models in Solving Machine Learning Tasks |
Adding metadata information involves integrating refer-
enced metadata, such as dates and purposes, into chunks for
filtering purposes, and incorporating metadata like chapters
and subsections of references to improve retrieval efficiency.
Alignment optimization addresses alignment issues and
disparities between docume... | RAG forLargeLanguageModels-ASurvey |
Piotr Nyczyk, et al. 2023. Graph of Thoughts: Solving Elaborate Problems with Large Language Models. arXiv preprint arXiv:2308.09687 (2023).
[21] Zhengda Bian, Qifan Xu, Boxiang Wang, and Yang You. 2021. Maximizing parallelism in distributed training for huge neural networks. arXiv preprint
arXiv:2105.14450 (2021).
... | TheEfficiencySpectrumofLargeLanguageModels-AnAlgorithmicSurvey |
Journal of Information Science, 2022, pp. 1–11 (cid:2) The Author(s), DOI: 10.1177/01655515221112844
Rajabi and Etminani
11
[35] Fuji M, Nakazawa K and Yoshida H. ‘Trustworthy and explainable AI’ achieved through knowledge graphs and social imple-
mentation. Fujit Sci Tech J 2020; 56(1): 39–45.
[36] Sun H, Xiao J... | Knowledge-graph-based explainable AI- A systematic review |
[14] Stephen Cave, Kate Coughlan, and Kanta Dihal. 2019. "Scary Robots": Examining Public Responses to AI. In Proceedings
of the 2019 AAAI/ACM Conference on AI, Ethics, and Society (Honolulu, HI, USA) (Aies ’19). Association for Computing
Machinery, New York, NY, USA, 331–337. https://doi.org/10.1145/3306618.3314232
[... | AI enhance sour performance |
as Alpaca [30] and Vicuna [8], have been developed based on LLaMA [32] and also exhibit similar
performance.
Leveraging Pre-trained LLMs in Vision-Language Tasks. In recent years, the trend of using
autoregressive language models as decoders in vision-language tasks has gained significant traction [6,
15, 36, 31, 2, 16,... | MiniGPT-4- Enhancing Vision-Language Understanding with Advanced Large Language Models |
https://a16z.com/2023/06/20/emerging-architectures-for-llm-applications/
6/15
23/06/2023, 16:52
Emerging Architectures for LLM Applications | Andreessen Horowitz
https://a16z.com/2023/06/20/emerging-architectures-for-llm-applications/
7/15
23/06/2023, 16:52
Emerging Architectures for LLM Applications | Andrees... | Emerging Architectures for LLM Applications _ Andreessen Horowitz |
[366] Suglia, A., Q. Gao, J. Thomason, et al. Embodied BERT: A transformer model for embodied,
language-guided visual task completion. CoRR, abs/2108.04927, 2021.
[367] Ganesh, S., N. Vadori, M. Xu, et al. Reinforcement learning for market making in a multi-agent
dealer market. CoRR, abs/1911.05892, 2019.
[368] Tip... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
We presented Imagen Video: a text-conditional video generation system based on a cascade of video
diffusion models. By extending the text-to-image diffusion models of Imagen (Saharia et al., 2022b)
to the time domain, and training jointly on video and images, we obtained a model capable of gen-
erating high fidelity vid... | IMAGEN VIDEO- HIGH DEFINITION VIDEO GENERATION WITH DIFFUSION MODELS |
Conversation collection task: The optional demographic survey response rate for volunteers (n=106) was 86%.
Several volunteers participated in multiple collection sessions. Due to de-identification of data for privacy protection,
these figures double-count repeat participants. Intersectional ethnic identities were also c... | LaMDA- Language Models for Dialog Applications |
Several excellent review articles have already greatly enriched our knowledge of
misinformation and its correction. Each is a valuable resource for deeper
reading on this subject. In the interest of not rehashing existing work, we have
made a conscious choice to showcase topics not already covered in these
reviews. How... | Social_Media_and_Democracy |
the upper limit of a single-user workstation. At the finetuning stage we want to mimic the original
BERT finetuning and evaluation setup, but provide additional limits to prevent gains based on tuning
of only the downstream procedure, for example via computationally extensive downstream training
(Bahri et al., 2021a), us... | CRAMMING-TRAININGALANGUAGEMODELONA SINGLEGPUINONEDAY |
21
Preprint
Zhilin Yang, Zihang Dai, Yiming Yang, Jaime Carbonell, Ruslan Salakhutdinov, and Quoc V. Le.
XLNet: Generalized Autoregressive Pretraining for Language Understanding. arXiv:1906.08237
[cs], January 2020. URL http://arxiv.org/abs/1906.08237.
Yang You, Jing Li, Sashank Reddi, Jonathan Hseu, Sanjiv Kumar, ... | CRAMMING-TRAININGALANGUAGEMODELONA SINGLEGPUINONEDAY |
Ni, Andrew Nystrom, Alicia Parrish, Marie Pellat, Martin Polacek, Alex Polozov, Reiner Pope, Siyuan
Qiao, Emily Reif, Bryan Richter, Parker Riley, Alex Castro Ros, Aurko Roy, Brennan Saeta, Rajkumar
Samuel, Renee Shelby, Ambrose Slone, Daniel Smilkov, David R. So, Daniel Sohn, Simon Tokumine, Dasha
Valter, Vijay Vasude... | CodeLlama2 |
random example
q: Lexden History
p: The site on which Lexden now stands was crossed
by the fortifications of iron age Colchester. . .
q: What makes a client good quality to you?
I’m putting together my ideal client . . .
p: Respectful of schedules. And pays on time.. . .
q: Central Intake Unit | Broome County
p: Casewor... | E5 |
to add a higher prior to children with large support sizes. More fundamentally, the reason why both
proposed approaches do not add biased priors to PCs is that they are designed to be model-agnostic,
i.e., their definitions as shown in Sec. 2 are independent with the model they apply to.
Empirical evaluation We empirica... | Tractable Regularization of Probabilistic Circuits |
a16z crypto
State of Crypto
2023
Trends to Watch: Scaling Blockchains
18
Ethereum now consumes 0.001% of the energy
that YouTube consumes annually
Ethereum switched to energy-saving Proof of Stake (PoS) from energy-intensive Proof of Work (PoW)*
Estimated energy consumption
YouTube
Gold mining
Glob... | State-of-Crypto2023 |
100k
85.5
200k
84.2
400k (default) 81.8
80.5
800k
79.7
1.6M
62.4
60.3
59.0
57.8
56.7
60.6
59.2
57.8
56.8
56.1
67.8
68.3
70.1
70.5
71.6
70.3
70.8
72.5
72.7
73.3
Table S3: Ablation for the length of consistency-regularized fine-tuning with an initial training length of 400k steps.
MPJPE↓
20k
82.0
40k (default) 81.8
... | Learning 3D Human Pose Estimation from Dozens of Datasets using a Geometry-Aware Autoencoder to Bridge Between Skeleton Formats |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.