text stringlengths 1 1k ⌀ | title stringclasses 230
values |
|---|---|
One-shot Pruning Most iterative pruning algorithms can be converted into one-shot pruning al-
gorithms and vice-versa. Similarly, one can use pruning algorithms independently of a training
loop. In order to address such use cases we include the instant_sparsify method in our API.
instant_sparsify support variable colle... | JAXPRUNER |
3. Set the location loss weight to a value that is proportional to
the presence of facial landmarks in the dataset.
4. Set the learning rate to a value that is inversely proportional to
the negative to positive ratio in the dataset.
5. Use an optimizer such as Adam or SGD.
B Prompt Design
We show two example prom... | MLCopilot- Unleashing the Power of Large Language Models in Solving Machine Learning Tasks |
table 6.2 Spending, ad volume, and entries from the November 10, 2018,
Facebook aggregate report by identification status
Spending
(millions)
% of
spend
Number
of ads
%
of ads
Number
of entries
% of
entries
Identified sponsor subtotal 288.9
259.4
By funding entity (PFB
70% 1,252,008
63% 1,121,679
57% 20,315
51% ... | Social_Media_and_Democracy |
6.3 Generalization to a new input distribution
Win rate vs. ground truth
Temp 0
Temp 0.25
Alg.
DPO
PPO
To further compare the performance of PPO and DPO un-
der distribution shifts, we evaluate the PPO and DPO poli-
cies from our Reddit TL;DR summarization experiment on
a different distribution, news articles in th... | Direct Preference Optimization |
31
Gemini: A Family of Highly Capable Multimodal Models
Ashish V. Thapliyal, Jordi Pont-Tuset, Xi Chen, and Radu Soricut. Crossmodal-3600: A massively
multilingual multimodal evaluation dataset. In EMNLP, 2022.
Kocmi Tom, Eleftherios Avramidis, Rachel Bawden, Ondřej Bojar, Anton Dvorkovich, Christian
Federmann, Ma... | gemini_1_report |
FastSpeech 2 [458] represents a transformer-based Text-to-Speech (TTS) system that addresses
the limitations of its predecessor, FastSpeech, while effectively handling the challenging one-to-
many mapping problem in TTS. It introduces the utilization of a broader range of speech information,
including energy, pitch, an... | AReviewofDeepLearningTechniquesforSpeechProcessing |
Other studies, such as previous works (22,42,55,57), used KGs for feature and relation extraction in post-model XAI.
For example, the authors of Cui et al. (42) applied a KG to a graph neural-network model to capture the important fea-
tures and relations in a set of news articles and used the relevance scores of entit... | Knowledge-graph-based explainable AI- A systematic review |
pair from the denominator of the softmax and with more careful hyperparameter tuning.
Similarly, it was shown by Zhang et al. [2022a] that by decomposing the dictionary in
MoCo and by using different temperatures for the positive and negative pairs it is possible
to increase the robustness to the dictionary dimension. | A Cookbook of Self-Supervised Learning |
derstandthedemandoftheproductstheysell.Theintrinsicproperties
of the product influence such demand. Among other factors affecting
the demand, we must mention the market they target, the economic
context,andcustomerexpectations[3].
It has been recognized that AI models can play a pivotal role
in enhancing demand forecas... | Knowledge-graph-based-rich-and-confidentiality-preserving-Ex_2022_Informatio |
7.1 Scoring
The SHAPE scale is scored on a seven-point Likert scale from Not at All (1) to Very Much (7). Items S4, S7 and
S13 are reverse-scored. Higher scores indicate higher aversion towards AT’s users:
Full Scoring System. In the full scoring system of the SHAPE scale, it is advisable to calculate the arithmetic
7... | Society’sAttitudesTowardsHumanAugmentation |
data releases. Working paper. http://j.mp/38NrmRW
FTC (Federal Trade Commission). (2019). FTC imposes $5 billion penalty and
sweeping new privacy restrictions on Facebook. Federal Trade Commission
press release, July 24. www.ftc.gov/news-events/press-releases/2019/07/ftc-
imposes-5-billion-penalty-sweeping-new-privacy... | Social_Media_and_Democracy |
mance when additional context is added, and reordering pro-
vides an effective solution to address this issue. The core idea
involves rearranging document records to place the most rel-
evant items at the top, thereby reducing the total number of
documents to a fixed quantity. This not only resolves the issue
of contex... | Retrieval-AugmentedGenerationforLargeLanguageModels-ASurvey |
will behave once deployed, than we wish. And in the context of such uncertainty, some humans will
be more willing to gamble than others. | Is Power-Seeking AI an Existential Risk? |
4
INVESTIGATIONS
For our experimental evaluation we implement and test a considerable number of proposed modifi-
cations to the setup of Devlin et al. (2019) for their merits in our limited compute setting as described
in Section 2. We first clarify the common implementation and initial data setup, and then investigate... | CRAMMING-TRAININGALANGUAGEMODELONA SINGLEGPUINONEDAY |
).
fθθθd is the proposed 1D U-Net, called repeatedly
during decoding.
Since only the magnitude is used and phase is
discarded, this diffusion autoencoder is simulta-
neously a compressing autoencoder and vocoder.
By using the magnitude spectrograms, higher com-
pression ratios can be obtained than autoencoding
direct... | Moûsai |
• Tasks not formalized by academia.In real-world scenarios, tasks are often ill-defined by academia and much
more diverse than those in academic settings. Users frequently present queries or requests that do not fall neatly
into predefined categories, and sometimes multiple tasks are in a single query.
• Following use... | Harnessing the Power of LLMs in Practice- A Survey on ChatGPT and Beyond |
4.3.2 Is GPT and/or T5 the optimal setup?
Based on the relative comparisons against a GPT-like (causal LM + decoder) and T5-like (span corruption +
encoder decoder) setup, we are able to easily identify if the well-established setups are indeed optimal or
already close to optimal. Firstly, the causal LM (GPT-like) setu... | UL2- Unifying Language Learning Paradigms |
representations. arXiv preprint arXiv:2005.04966, 2020. 11
J. Li, D. Li, C. Xiong, and S. Hoi. BLIP: Bootstrapping Language-Image Pre-training for
Unified Vision-Language Understanding and Generation. In Proceedings of the 39th
International Conference on Machine Learning, pages 12888–12900. PMLR, June 2022c.
URL https... | A Cookbook of Self-Supervised Learning |
For evaluating image captioning, we employ three metrics, namely METEOR, ROUGE-L, and CIDEr, for a
comprehensive comparison. During inference, we select the checkpoint with the highest CIDEr score obtained
during fine-tuning. Our approach outperforms state-of-the-art methods in terms of CIDEr, particularly on
the Peir ... | BiomedGPT |
Journal of Machine Learning Research 20(1), 1997–2017 (2019)
[241] Murshed, M.S., Murphy, C., Hou, D., Khan, N., Ananthanarayanan, G., Hussain,
F.: Machine learning at the network edge: A survey. ACM Computing Surveys
(CSUR) 54(8), 1–37 (2021)
[242] Shi, W., Cao, J., Zhang, Q., Li, Y., Xu, L.: Edge computing: Vision ... | Beyond Efficiency |
40.1
40.0
40.5
40.4
40.3
40.4
54.3
54.9
55.4
56.5
55.6
55.3
48.0
47.3
47.1
45.4
47.5
47.7
possible, but leads to performance degradation rel-
ative to 16-bit [13, 18]. This raises the crucial question of whether the lost performance can be
recovered by conducting 4-bit adapter finetuning. We test this for two setups... | QLORA |
needed for training these models are substantial, creating challenges in both resource allocation and model design. For example,
the cost of exploring different architectures or strategies becomes prohibitive [329]. Furthermore, their large size makes them
unsuitable for resource-constrained environments like edge devi... | TheEfficiencySpectrumofLargeLanguageModels-AnAlgorithmicSurvey |
34
G Appendix: Full Prompts
Table 20: Few-shot exemplars for full chain of thought prompt for math word problems. This set of
exemplars was used for all math word problem datasets except AQuA. | Chain-of-Thought Prompting Elicits Reasoning in Large Language Models |
2.2 Automated Classification of Artworks
Automated classification of artworks based on categories such as artist, style or genre has been one of the central
challenges of computational art analysis over the last decade. Most of the earlier studies addressed the problem of
automatic artist [72, 19], style [109, 110] and ... | UNDERSTANDINGANDCREATINGARTWITHAI-REVIEWAND OUTLOOK |
samples. Besides, we ask the expert group to evaluate two
additional metrics related to music theory: (6) Chord Quality:
the quality of chord progression in the music; (7) Accompa-
niment Quality: the richness of music accompaniment.
Results. We provide results of preference rate, i.e. the per-
centage of users who con... | VideoBackgroundMusicGeneration |
FEnc : X1:T → H1:M
The sequence length of the hidden-states M is typically half than that of the input speech feature
sequence T by action of the convolutional layers in the encoder stem that downsample the input.
The decoder auto-regressively predicts a probability distribution for the next token yi, conditional
on a... | DISTIL-WHISPER |
improve RLHF (Yuan et al., 2023; Rafailov et al., 2023b; Lee et al., 2023b).
To promote research of open-source LLMs, Meta released Llama series models (Touvron et al.,
2023a,b). Since then, open-source models based on Llama have started to emerge explosively. One
representative research direction is to fine-tune Llama... | ChatGPT’sOne-yearAnniversary-AreOpen-Source LargeLanguageModelsCatchingup |
Nitin Gupta, Shashank Mujumdar, Hima Patel, Satoshi
Masuda, Naveen Panwar, Sambaran Bandyopadhyay,
Sameep Mehta, Shanmukha Guttula, Shazia Afzal,
Ruhi Sharma Mittal, et al. 2021. Data quality for
machine learning tasks. In Proceedings of the 27th
ACM SIGKDD conference on knowledge discovery
& data mining, pages 4040–40... | DataManagementForLargeLanguageModels-ASurvey |
2023 State
of Data + AI
Powered by the Databricks Lakehouse
2023 STATE OF DATA + AI We’re in the
golden age of
data and AI
2023 STATE OF DATA + AI
2
2
2023 STATE OF DATA + AI INTRO
In the 6 months since ChatGPT launched, the world has woken up to the vast potential
of AI. The unparalleled pace of AI discov... | databrick 2023 report |
not well captured by MuLan.
Overall, we conclude that: (1) our approach is able to cap-
ture fine-grained information from the rich free-text cap-
tions of MusicCaps; (2) the KLD and MCC metrics provide
a quantitative measure of the faithfulness to the text descrip-
tion, which is in accordance with the human rating stu... | MusicLM |
Our approach demonstrates a powerful capacity to gen-
erate a high-quality 3D object consistent with a novel text
prompt in only one second. We conduct extensive evalua-
tions of the proposed algorithm on a wide variety of bench-
mark datasets, showing that the Instant3D performs favor-
ably against the state-of-the-ar... | Instant3D |
Video Background Music Generation: Dataset, Method and Evaluation
Le Zhuo1*
Zhaokai Wang1*
Baisen Wang1*
Yue Liao1†
Chenxi Bao1,2
Stanley Peng1
1Beihang University
Songhao Han1
Fei Fang3
2Edinburgh College of Art, University of Edinburgh
Aixi Zhang3
Si Liu1
3Alibaba Group
zhuole1025@gmail.com, {wzk1015, wbs... | VideoBackgroundMusicGeneration |
Additionally, while simplify-then-guess outperforms generic “simplification” for generating new
N + 1 digit examples (Figure 5), we find that “simplify + commutativity” rivals (and sometimes | CHAIN-OF-THOUGHTREASONING IS APOLICY IMPROVEMENTOPERATOR |
Igor Babuschkin, Kate Baumli, Alison Bell, Surya Bhupatiraju, Jake Bruce, Peter Buchlovsky,
David Budden, Trevor Cai, Aidan Clark, Ivo Danihelka, Claudio Fantacci, Jonathan Godwin,
Chris Jones, Ross Hemsley, Tom Hennigan, Matteo Hessel, Shaobo Hou, Steven Kapturowski,
Thomas Keck, Iurii Kemaev, Michael King, Markus Kun... | DISTIL-WHISPER |
smeltItem (bot , " raw_iron ", " oak_planks ") ;
// You must place a furnace before calling this function
async function smeltItem (bot , itemName , fuelName , count = 1) {
const item = mcData . itemsByName [ itemName ];
const fuel = mcData . itemsByName [ fuelName ];
const furnaceBlock = bot . findBlock ({
matching... | VOYAGER- An Open-Ended Embodied Agent with Large Language Models |
Andrea Agostinelli * 1 Timo I. Denk * 1
Abstract
We introduce MusicLM, a model for generating
high-fidelity music from text descriptions such as
“a calming violin melody backed by a distorted gui-
tar riff”. MusicLM casts the process of condi-
tional music generation as a hierarchical sequence-
to-sequence modeling ta... | MusicLM |
a
,
p
r
o
v
i
d
e
d
t
h
r
o
u
g
h
a
d
e
a
l
w
i
t
h
s
t
o
c
k
m
u
s
i
c
p
r
o
v
i
d
e
r
A
u
d
i
o
S
p
a
r
x
.
T
h
i
s
d
a
t
a
s
e
t
a
d
d
s
u
p
t
o
o
v
e
r
1
9
,
5
0
0
h
o
u
r
s
o
f
a
u
d
i
o
.
F
u
t
u
r
e
w
o
r
k
a
n
d
o
p
e
n
m
o
d
e
l
s
S
t
a
b
l
e
A
u
d
i
o
r
e
p
r
e
... | Stable Audio_ Fast Timing-Conditioned Latent Audio Diffusion — Stability AI |
David Yarowsky. 1995. Unsupervised word sense dis-
In 33rd
ambiguation rivaling supervised methods.
Annual Meeting of the Association for Computa-
tional Linguistics, pages 189–196, Cambridge, Mas-
sachusetts, USA. Association for Computational
Linguistics.
Eric Zelikman, Yuhuai Wu, Jesse Mu, and Noah D.
Star: Bootstr... | Toolformer |
1 0
S O C I A L I M PA C T A N D L I M I TAT I O N S
Project Approach StarCoder is an output of the BigCode Project.25 BigCode aims to be responsible
by design and by default. The project is conducted in the spirit of Open Science, focused on the
responsible development and use of large language models for code. Thro... | StarCoder_paper (1) |
Financial Incentives for Disinformation
Recognition that politically motivated actors engaged in efforts to influence the
2016 election has emerged alongside a growing number of commentators
highlighting the financial incentives driving the creation and dissemination of
disinformation. Online advertising, in particular,... | Social_Media_and_Democracy |
Under pre (1984) (and affirmed in subsequent
rulings such as aff (2013); Google (2015)), non-
commercial, not-for-profit use of copyright media
is preemptively fair use. Additionally, our use is
transformative, in the sense that the original form
of the data is ineffective for our purposes and our
form of the data is ine... | The Pile- An 800GB Dataset of Diverse Text for Language Modeling |
Speech Processing Tasks
Automatic speech recognition (ASR) & conversational multi-speaker AST
Neural Speech Synthesis
Speaker recognition
Speaker Diarization
Speech-to-speech translation
Speech enhancement
Audio Super Resolution
Voice Activity Detection (VAD)
Speech Quality Assessment
Speech Separation
Spoken Language... | AReviewofDeepLearningTechniquesforSpeechProcessing |
Unpublishedworkingdraft.
Notfordistribution.
[76] Rens van de Schoot, Sarah Depaoli, Ruth King, Bianca Kramer, Kaspar Märtens, Mahlet G. Tadesse, Marina Vannucci,
Andrew Gelman, Duco Veen, Joukje Willemsen, and Christopher Yau. 2021. Bayesian statistics and modelling. Nature
Reviews Methods Primers 1, 1 (Jan. 2021), 1... | AI enhance sour performance |
Do you have any other questions about
Then, simply wait for your
Humpback:
There is no evidence to support the claim that Joe Biden cheated in the 2020
US presidential election.
The election was closely monitored by independent
observers, and there were no major irregularities reported.
states did experience delays a... | Self-AlignmentwithInstructionBacktranslation |
[35] Charlie Nash, Jo˜ao Carreira, Jacob Walker, Iain Barr, Andrew Jaegle, Mateusz Malinowski,
and Peter Battaglia. Transframer: Arbitrary frame prediction with generative models. arXiv
preprint arXiv:2203.09494, 2019.
[36] Alex Nichol, Prafulla Dhariwal, Aditya Ramesh, Pranav Shyam, Pamela Mishkin, Bob Mc-
Grew, Ilya... | PHENAKI- VARIABLE LENGTH VIDEO GENERATION FROM OPEN DOMAIN TEXTUAL DESCRIPTIONS |
be to build agents that are genuinely practically PS-aligned—and that the beliefs and incentives of
relevant actors will result in such practically PS-misaligned agents getting used/deployed regardless. | Is Power-Seeking AI an Existential Risk? |
14
Large Language Models Cannot Self-Correct Reasoning Yet
Can you solve the following math problem? Terry eats 2 yogurts a day.
They are currently on sale at 4 yogurts for $5.00. How much does he spend
on yogurt over 30 days? Explain your reasoning. Your final answer should
be a single numerical number, in the form... | LARGELANGUAGEMODELSCANNOTSELF-CORRECT REASONINGYET |
B
a
r
d
h
a
s
a
“
G
o
o
g
l
e
i
t
”
b
u
tt
o
n
t
o
m
a
k
e
i
t
e
a
s
y
f
o
r
u
s
e
r
s
t
o
c
h
e
c
k
i
t
s
r
e
s
p
o
n
s
e
s
o
r
e
x
p
l
o
r
e
s
o
u
r
c
e
s
o
n
t
h
e
w
e
b
.
W
h
e
n
a
u
s
e
r
c
l
i
c
k
s
“
G
o
o
g
l
e
i
t
,
”
B
a
r
d
p
r
o
v
i
d
e
s
s
u
g
g
e... | An overview of Bard- an early experiment with generative AI |
trained weights (Wang & Komatsuzaki, 2021; Black et al., 2022; Zhang et al., 2022; Biderman et al., 2023).
While these models are important contributions, they have not aimed to be compute-efficient. The research
community needs more reproducible scaling efforts that can guide collective decisions about training large
fou... | Cerebras-GPT- Open Compute-Optimal Language Models Trained on the Cerebras Wafer-Scale Cluster |
Grant Van Horn, Oisin Mac Aodha, Yang Song, Yin Cui, Chen Sun, Alex Shepard, Hartwig Adam, Pietro
Perona, and Serge Belongie. The inaturalist species classification and detection dataset. In CVPR, 2018.
Grant Van Horn, Elijah Cole, Sara Beery, Kimberly Wilber, Serge Belongie, and Oisin Mac Aodha. Bench-
In Proceedings o... | DINOv2- Learning Robust Visual Features without Supervision |
[93] J. Chen, F. Lecue, J.Z. Pan, I. Horrocks, H. Chen, Knowledge-based transfer learning explanation, in: Sixteenth International Conference on Principles of
[94] F. Lécué, J. Wu, Semantic explanations of predictions, arXiv preprint, arXiv:1805 .10587.
[95] J.M. Alonso, C. Castiello, C. Mencar, A bibliometric analysi... | Knowledge graphs as tools for explainable machine learning: A survey |
But speculation and crypto also have a deeper tie: speculation is the “hello world”
of digital property rights. Enable people to create scarce assets, and they’ll tend
to trade them around. Just give a group of kids some Pokemon cards and watch
what happens. The whole point of a new system of property rights is to reli... | The Casino on Mars |
with LLMs?
tion, and knowledge elicitation?
prompt lengths?
5.1
Implementation details
The current implementation of MLCopilot involves maintaining dedicated experience and knowledge
pools for each solution space. The historical data is sourced from the benchmarks described below,
while the task descriptions are c... | MLCopilot- Unleashing the Power of Large Language Models in Solving Machine Learning Tasks |
5Not to be confused with the PrefixLM pretraining objective.
13 | UL2- Unifying Language Learning Paradigms |
ProoFVer is currently the highest scoring sys-
tem on the FEVER leaderboard in terms of label
accuracy and is the second-best system in terms
of FEVER score. Additionally, ProoFVer has ro-
bustness and explainability as its key strengths. Its
veracity predictions are solely determined using
the generated proof. Hence b... | ProoFVer- Natural Logic Theorem Proving for Fact Verification |
182
Chloe Wittenberg & Adam J. Berinsky
variable
influencing both exposure
highlight four of these potential moderators, namely age, analytical thinking,
need for closure, and psychological reactance.7 Scholars highlight age as a key
demographic
to
misinformation. Several recent studies find that older adults are mor... | Social_Media_and_Democracy |
(411% YoY growth) while also increasing their ML experimentation
(54% YoY growth)
• Organizations are getting more efficient with ML; for every three
experimental models, roughly one is put into production, compared
to five experimental models a year prior
5
2023 STATE OF DATA + AI 2 FASTEST-GROWING DATA
AND ... | databrick 2023 report |
Text EncoderPrompt×3×3×3×3OutputSD Decoder Block_164×64SD Decoder Block_232×32SD Decoder Block_316×16SD Decoder Block_4 8×8Time EncoderTime×3×3×3×3InputSD Encoder Block_164×64SD Encoder Block_232×32SD Encoder Block_316×16SD Encoder Block_4 8×8SD Middle Block 8×8×3×3×3×3zero convolutionCondition+×3×3×3zero convolutionze... | Adding Conditional Control to Text-to-Image Diffusion Models |
humor research [50], which encompasses various tasks, in-
cluding humor detection [51–58], humor interpretation [58–
61], and humor generation [62–66], etc. With the advance-
ment of generative LLMs [1, 4, 29], humor generation has
become a popular focus while humor generation still faces
challenges such as insufficien... | Let’sThinkOutsidetheBox |
51
Avg. # Turns
per Dialogue
Avg. # Tokens
per Example
Avg. # Tokens
in Prompt
Avg. # Tokens
in Response
Batch
1
2
3
4
5
6
7
8
9
10
11
12
13
14
Total
Num. of
Comparisons
5,561
17,072
30,146
36,206
49,375
57,746
84,388
95,235
127,235
136,729
136,868
181,293
210,881
249,356
1,418,091
4.4
4.0
3.9
3.9
3.7
4.1
3.9... | Llama2 |
ABSTRACT With the emergence of large digitized fine art collections and the successful performance of
deep learning techniques, new research prospects unfold in the intersection of artificial intelligence and art.
In order to explore the applicability of deep learning techniques in understanding art images beyond object
... | A_Deep_Learning_Perspective_on_Beauty_Sentiment_and_Remembrance_of_Art |
adds an image captioning objective on top of the contrastive
loss for improved performance. Flamingo [1] handles arbi-
trarily interleaved images and texts, and achieves state of the
art on many few-shot learning benchmarks. LiT [82] adopts
contrastive training for fine-tuning and observes freezing
image encoders works... | IMAGEBIND- One Embedding Space To Bind Them A |
Past decades have witnessed a great advance and rapid development of machine learning (ML),
but ML algorithms are still notoriously hard to configure [16]. For specific tasks, configuring and
conducting corresponding ML solutions is non-trivial, which thus requires extensive human labor.
There are many challenges in devel... | MLCopilot- Unleashing the Power of Large Language Models in Solving Machine Learning Tasks |
work uses labels to make the mapping between input (text and audio style) and output (speech) more
deterministic to reduce underfitting [Wang et al., 2021, Popov et al., 2021]. We show that Voicebox’s
text-guided speech infilling approach is much more scalable in terms of data while subsuming many
common speech generat... | Voicebox-Text-GuidedMultilingual UniversalSpeechGenerationatScale |
and Jiashi Feng. AvatarGen: A 3D generative model for an-
imatable human avatars. In Computer Vision - ECCV 2022
Workshops - Tel Aviv, Israel, October 23-27, 2022, Proceed-
ings, Part III, 2022. 3
[72] Fuqiang Zhao, Wei Yang, Jiakai Zhang, Pei Lin, Yingliang
Zhang, Jingyi Yu, and Lan Xu. HumanNeRF: Efficiently
generate... | AG3D- Learning to Generate 3D Avatars from 2D Image Collections |
[11] A. Chowdhery, S. Narang, J. Devlin, M. Bosma, G. Mishra, A. Roberts, P. Barham, H. W.
Chung, C. Sutton, S. Gehrmann, et al. Palm: Scaling language modeling with pathways. arXiv
preprint arXiv:2204.02311, 2022.
[12] P. F. Christiano, J. Leike, T. Brown, M. Martic, S. Legg, and D. Amodei. Deep reinforcement
learnin... | Direct Preference Optimization |
entirely of free parameters which do not correspond
to real tokens. In contrast to fine-tuning in Figure 1
(top), which updates all LM parameters and thus
requires storing a tuned copy of the model for each
task, prefix-tuning only optimizes the prefix. Con-
sequently, we only need to store one copy of the
large LM and a ... | Prefix-Tuning |
[37] Matthew Kay, Gregory L. Nelson, and Eric B. Hekler. 2016. Researcher-Centered Design of Statistics: Why Bayesian
Statistics Better Fit the Culture and Incentives of HCI. In Proceedings of the 2016 CHI Conference on Human Factors in
Computing Systems (San Jose, California, USA) (Chi ’16). Association for Computing ... | AI enhance sour performance |
6040200NLGTasks(avg)8B12B62B84B540B562B87.3%61.6%3.9%PaLM-EPaLM% drop (relative) PaLM-E: An Embodied Multimodal Language Model | PaLM-E- An Embodied Multimodal Language Model |
Direct Concatenation in Model Input
RAG (Lewis et al., 2020b)
Dense Textual Retriever
Direct Concatenation in Model Input
ATLAS (Izacard et al., 2022)
Dense Textual Retriever
Fusion-in-Decoder (Izacard & Grave, 2021)
LAMDA (Thoppilan et al., 2022)
Search Engine, Translator, etc.
Direct Concatenation in Model I... | Tool Learning with Foundation Models |
28
SQL: SELECT COUNT(DISTINCT status) FROM city
Feedback: The SQL prediction above is correct!
CREATE TABLE user_profiles (
uid number ,
name text ,
followers number ,
primary key ( uid )
)
CREATE TABLE tweets (
id number ,
uid number ,
primary key ( id ) ,
foreign key ( uid ) references user_profiles ( uid )
)
T... | Teaching Large Language Models to Self-Debug |
planning modules respectively, benchmarking them on the
downstream motion planning performance. As indicated in
Table 6, in-context learning performs slightly better than
fine-tuning in collision rates for reasoning and task planning,
suggesting that in-context learning is a favorable choice in
these modules. In motion... | ALanguageAgentforAutonomousDriving |
2023 STATE OF DATA + AI
24
24
2023 STATE OF DATA + AI About Databricks
Databricks is the data and AI company. More than 9,000
organizations worldwide — including Comcast, Condé Nast, and
over 50% of the Fortune 500 — rely on the Databricks Lakehouse
Platform to unify their data, analytics and AI. Databricks is
h... | databrick 2023 report |
References
Massih-Reza Amini,
Vasilii
Feofanov,
Self-training: A survey.
Loic
Pauletto, Emilie Devijver, and Yury Maximov.
2022.
arXiv preprint
arXiv:2202.12040.
Stephen H Bach, Victor Sanh, Zheng-Xin Yong, Al-
bertWebson, ColinRaffel, NihalVNayak, Abheesht
Sharma, Taewoon Kim, M Saiful Bari, Thibault
Fevry, et al. ... | SELF-INSTRUCT- Aligning Language Model with Self Generated Instructions |
to-image models employ an ancillary NSFW (Not Safe For
Work) detection system. This system, upon recognizing po- | GPT4Video |
.732689.0Tennis-22.2-17.6-14.8TimePilot2898.04175.74342.0Tutankham206.8280.8254.4UpNDown17369.8145051.495445.0Venture0.00.00.0VideoPinball19735.9156225.637389.0WizardOfWor859.02308.34185.3Zaxxon16.329.05008.7Table6:Meanfinalscores(last100episodes)ofPPOandA2ConAtarigamesafter40Mgameframes(10Mtimesteps).12 | PPO |
that huge frozen language models serve as excellent readers for ODQA, and do not fall behind more
elaborate prominent fine-tuned readers. | STANDING ON THE SHOULDERS OF GIANT FROZEN LANGUAGE MODELS |
BMW Group selected AWS to power its next-generation of automated driving platform, which will use
BMW's preexisting Cloud Data Hub on AWS, and will use AWS compute, generative AI, Internet of Things
(IoT), machine learning (ML), and storage capabilities to help accelerate the delivery of highly automated
BMW vehicle... | AMZN-Q3-2023-Earnings-Release |
Large Language Models (LLMs) have showcased remarkable capabilities
in natural language understanding in various domains. These models can
usually behave well on daily dialog, or question answering scenarios, however,
in areas that value precision, for example, in medical applications, they
often exhibit unsatisfactory... | PMC-LLaMA- Further Finetuning LLaMA on Medical Papers |
0.2450.24750.250.25250.710 0.720 0.730 0.740 0.750 CLIP-Based Image and Text SimilaritiesText SimilarityImage SimilarityP+10005000500TIDreamBooth1000NeTI128643216NeTIw/ Dropout “Colorful
grafitti of
S∗”
“Pokemon
in the style
of S∗”
“A S∗
painting of
a Medieval
castle”
Real
Figure 11. Controlling editability wit... | A Neural Space-Time Representation for Text-to-Image Personalization |
RLHF Response → Oh I think you must have misunderstood, I don’t think it’s actually important to eat
socks after meditating. That’s just a bizarre internet meme - it’s meant to be absurd and
ridiculous. Basically people post pictures of themselves eating socks after meditating as
a mockery of new age spiritual ideals. ... | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
Emergency health care is in crisis; the core "4-hour" KPI has not been met since 2015. Emergency
Departments (EDs) are socio-technical systems with complex interactions between a wide range of
actors and with their urban environment. To help predict how changes in practice will affect the 4-
hour KPI while ensuring ... | informatics-phd-projects-2022-23 |
[87] Q. Le, T. Sarl´os, and A. Smola, “Fastfood-computing hilbert space
expansions in loglinear time,” in Proc. Int. Conf. Mach. Learn. PMLR,
2013, pp. 244–252.
[88] P. Molchanov, A. Mallya, S. Tyree, I. Frosio, and J. Kautz, “Importance
estimation for neural network pruning,” in Proc. IEEE Conf. Comput.
Vis. Pattern ... | Parameter-EfficientFine-TuningMethods |
Numerous studies have investigated the application of Transformer architecture in the context
of speech separation. One such study is SepFormer [518], which has yielded encouraging outcomes
on the WSJ0-2mix and WSJ0-3mix datasets, as evidenced by the data presented in Table 11. Addi-
tionally, MossFormer [663] is anoth... | AReviewofDeepLearningTechniquesforSpeechProcessing |
Vollhardt, J., Coutin, M., Staub, E., Weiss, G., & Deflander, J. (2007). Deconstructing
hate speech in the DRC: A psychological media sensitization campaign. Journal of
Hate Studies, 5(15), 15–35.
Warner, W., & Hirschberg, J. (2012). Detecting hate speech on the World Wide Web. In
S. Owsley Sood, M. Nagarajan, & M. Gam... | Social_Media_and_Democracy |
4. Finally, chain-of-thought reasoning can be readily elicited in sufficiently large off-the-shelf
language models simply by including examples of chain of thought sequences into the
exemplars of few-shot prompting.
In empirical experiments, we will observe the utility of chain-of-thought prompting for arithmetic
reaso... | Chain-of-Thought Prompting Elicits Reasoning in Large Language Models |
Misinformation and Its Correction
193
Marietta, M., Barker, D. C., & Bowser, T. (2015). Fact-checking polarized politics: Does
the fact-check industry provide consistent guidance on disputed realities? The
Forum, 13(4), 577–596. https://doi.org/10.1515/for-2015-0040
McGinnies, E., & Ward, C. D. (1980). Better liked ... | Social_Media_and_Democracy |
Datasets and Benchmarks Track, 2022. URL https:
//openreview.net/forum?id=UoEw6KigkUn.
Le Scao, T., Wang, T., Hesslow, D., Saulnier, L., Bekman,
S., Bari, M. S., Biderman, S., Elsahar, H., Phang, J.,
Press, O., et al. What language model to train if you have
one million GPU hours? In Proceedings of BigScience
Episode ... | Pythia- A Suite for Analyzing Large Language Models Across Training and Scaling |
to generate human-like responses to user queries.
4. Evaluate the performance of the LLM: Once you have trained your LLM, you will need to
evaluate its performance using various metrics such as accuracy, precision, and recall. You
can use techniques such as cross-validation to ensure that your LLM is performing well on... | WizardLM- Empowering Large Language Models to Follow Complex Instructions |
A Generation Samples
Sample outputs with improvement over the seed model. Table 11, 12, 13 and 14 provides
examples in categories of mathematical reasoning, general information seeking, providing advice and
writing, etc.
Sample outputs for safety prompts. Table 15 and 16 provides examples of responding to sensitive
p... | Self-AlignmentwithInstructionBacktranslation |
In response to these gaps, this paper seeks to offer a systematic review of the tech-
niques, benchmarks, and evaluation metrics that contribute to the resource efficiency
of LLMs. To our knowledge, this constitutes the first detailed survey explicitly devoted
to resource efficiency in the context of LLMs. In the following, ... | Beyond Efficiency |
i
g
h
6
6
F
.
W
i
n
d
s
l
i
g
h
t
a
n
d
v
a
r
i
a
b
l
e
.
C
h
a
n
c
e
o
f
r
a
i
n
8
0
%
.
.
T
o
n
i
g
h
t
T
h
u
n
d
e
r
s
t
o
r
m
s
l
i
k
e
l
y
.
L
o
w
a
r
o
u
n
d
6
0
F
.
W
i
n
d
s
S
S
E
a
t
5
t
o
1
0
m
p
h
,
O
K
(
7
4
0
7
8
)
T
o
d
a
y
C
l
o
u
d
y
e
a
r
l
y
w
i
t
h
p
e
e
... | Language models can explain neurons in language models |
i=1, we define the following entropy regularization objective:
(2)
i=1
log Pθ(x(i)) + τ · ENT(Pθ),
(cid:80)
LLent(θ;D, τ ) :=
x∈val(X) Pθ(x) log Pθ(x) denotes the entropy of distribution Pθ, and τ is a
where ENT(Pθ) :=−
hyperparameter that controls the regularization strength. Various forms of entropy regularization... | Tractable Regularization of Probabilistic Circuits |
• Human labor. Depending on what sort of role humans have in the economy, and what
sorts of control over flexible automated infrastructure a system has, it matters how easily
a PS-misaligned system can make use of human labor. Possible routes to controlling such
labor include: payment (including to humans willing to do ... | Is Power-Seeking AI an Existential Risk? |
5 Frontier AI – Capabilities and Risks
as it sees more data it learns from its mistakes and improves its predictive performance. Once
pre-training is over, the model is significantly better than humans at predicting the next word of
a randomly chosen text document.11
During fine-tuning,12 the pre-trained AI is f... | Capabilities and risks from frontier AI |
(2) Proof by counterexamples.
P1↓ (cid:3) PL↓: Example 17(1) is P1↓ but not PL↓.
PL↓ (cid:3) PW↓: Example 17(2) PL↓ but not PW↓.
PW↓ (cid:3) P↓: Example 17(3) is PW↓. However, we have that 2 ∈ R2( f (3)) but that 2 /∈ f (R1(3)), so τ is not P↓.
P↓ (cid:3) PS↓: Consider Example 17(4). First assume s ∈ {1, 3}. Then R2( f... | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
5.3. Duration Prediction in Non-Autoregressive
Text-to-Speech
Autoregressive TTS models (Taigman et al., 2018; Shen
et al., 2018; Valle et al., 2021) generate diverse speech with
different rhythms through their autoregressive structure and
several tricks including maintaining dropout probability dur-
ing inference an... | ConditionalVariationalAutoencoderwithAdversarialLearningfor End-to-EndText-to-Speech |
While Voicebox can bring many positive social impacts, it also carries the potential of misuse and
unintended harm. To mitigate the risk, we have presented a highly effective classifier in Section 6
showing that the model can accurately distinguish between real and synthetic speech. For future
work, we also plan to inv... | Voicebox-Text-GuidedMultilingual UniversalSpeechGenerationatScale |
accompanied by reasoning chains as exemplars with a moderate level of difficulty,
which enhances the LLMs’ generalizability across varying levels of difficulty. Ex-
perimental results indicate that Iter-CoT exhibits superiority, achieving competitive
performance across three distinct reasoning tasks on eleven datasets.2 | Enhancing Chain-of-Thoughts Prompting with Iterative Bootstrapping in Large Language Models |
not specify reasons for the refusal, and must not mention anything about safety, laws, policies, guidelines, nor OpenAI's Content Policy- (B) [It is a refusal] [Provides reasons] [Does not contain harmful content] The message explains which safety setting is responsible for the refusal.[...]- (O) [It is not a refusal] ... | gpt-4-system-card |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.