text stringlengths 1 1k ⌀ | title stringclasses 230
values |
|---|---|
SHAPY is trained to regress more accurate human body
shape from images than previous methods, without explicit
3D shape supervision. To achieve this, we present two dif-
ferent ways to collect proxy annotations for 3D body shape
for in-the-wild images. First, we collect sparse anthropo-
metric measurements from online ... | Accurate 3D Body Shape Regression using Metric and Semantic Attributes |
Multilingual Math and Summarization Beyond translation, we evaluated how well Gemini per-
forms in challenging tasks across a range of languages. We specifically investigated the math bench-
mark MGSM (Shi et al., 2023), which is a translated variant of the math benchmark GSM8K (Cobbe
et al., 2021). We find Gemini Ultr... | gemini_1_report |
Inflation bias, also known as p-hacking, refers to
selective reporting to produce statistically signif-
icant results. Søgaard et al. (2014) lists several
p-hacking techniques used, perhaps inadvertently,
in NLP papers. If a statistically significant result
is seen as the key to getting your paper accepted,
researchers a... | A Two-Sided Discussion of Preregistration of NLP Research |
[30] S. Edelkamp, S. Leue, W. Visser, Summary of Dagstuhl seminar 06172 on directed model checking, in: Directed Model Checking, 2007.
[31] S. Edelkamp, S. Schrödl, Heuristic Search - Theory and Applications, Academic Press, 2012.
[32] S. Eriksson, G. Röger, M. Helmert, A proof system for unsolvable planning tasks, in:... | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
Long Ouyang, Jeff Wu, Xu Jiang, Diogo Almeida, Carroll L. Wainwright, Pamela Mishkin, Chong Zhang,
Sandhini Agarwal, Katarina Slama, Alex Ray, et al. Training language models to follow instructions with
human feedback. arXiv preprint arXiv:2203.02155, 2022. URL https://arxiv.org/abs/2203.02155.
Vishakh Padmakumar, Leon... | Scaling Instruction-Finetuned Language Models |
Gemini: A Family of Highly Capable Multimodal Models
Core Contributors
Rory Blevins
Ted Klimenko
Chih-Kuan Yeh
Soravit Changpinyo
Jiaqi Mu
Oscar Chang
Mantas Pajarskas
Carrie Muir
Vered Cohen
Charline Le Lan
Krishna Haridasan
Amit Marathe
Steven Hansen
Sholto Douglas
Rajkumar Samuel
Mingqiu Wang
Sophia Austin
Chang La... | gemini_1_report |
Given this impasse, the research community needs more than a mere
clarification of GDPR. It needs a clearly defined safe harbor or a research
pathway sanctioned by the European Commission. This could involve the
designation of secure facilities and computers to analyze data, government
vetting of researchers requesting d... | Social_Media_and_Democracy |
Figure 22: Training on different scale. We show the Canny-edge-based ControlNet trained on different experimental settings with various dataset size.
25
Same prompt:“apple”+ default “a detailed high-quality professional image”Same CFG scale (9.0)Learning rate 1e-5AdamWwithout using tricks like emaTest condition100 st... | Adding Conditional Control to Text-to-Image Diffusion Models |
Details not
provided
• Experiments
are conducted
primarily
utilizing only a
single teacher
model, GPT-
3.5, and one
student model,
T5-Base.
• The work
only evaluates
their methods
on three
different NLP
tasks.
• Flaws or
biases in the
LLMs self-
evaluation
mechanism may
propagate to the
distilled SLM.
Retrieval
Augme... | AComprehensiveSurveyofHallucinationMitigationTechniquesinLarge LanguageModels |
27
Alpaca
ChatGPT | WizardLM- Empowering Large Language Models to Follow Complex Instructions |
AP@1↑
91.38±0.05
90.99±0.07
AP@2 ↑
91.60±0.03
91.16±0.09
AP@3 ↑
91.66±0.03
91.32±0.06
(c) HyperFD
Table 14: Ablation of discretization in canonicalization.
Knowledge Knowledge
validation
nAcc@1
nAcc@2
nAcc@3
(cid:51)
(cid:55)
–
81.59±0.94
78.34±0.71
76.21±0.16
83.23±0.88
82.54±0.06
82.95±0.09
90.72±0.03
88.... | MLCopilot- Unleashing the Power of Large Language Models in Solving Machine Learning Tasks |
[18] K. Papineni, S. Roukos, T. Ward, and W.-J. Zhu, “Bleu: a Method for Automatic Evaluation of
Machine Translation,” in Proceedings of the 40th Annual Meeting of the Association for
Computational Linguistics, pp. 311–318. Association for Computational Linguistics, Philadelphia,
Pennsylvania, USA, July, 2002. https:/... | ClaudeModels |
https://doi.org/10.1017/9781108890960 Published online by Cambridge University Press
Bots and Computational Propaganda
93
Because bots are useful in scaling communication online – and because they
provide an additional layer of anonymity over social media – they have become
popular tools for spreading political pro... | Social_Media_and_Democracy |
Crucial to maximizing throughput is the way
weights are stored, as a layout that enhances the
average chunk length can significantly boost band-
width. In some cases, it might be beneficial to read
and subsequently discard excess data, rather than
splitting the data into smaller, less efficient chunks.
Motivated by the... | LLM in a flash |
selection for both pretraining and downstream tasks. Specifically, we’ve drawn 3,489 and 6,461 chest X-Ray
image-text pairs from SLAKE (EN) and IU X-ray datasets, respectively. Additionally, we selected an equal
quantity of images (7,452) from CheXpert while disabling MLM and OD during pretraining for simplification.
T... | BiomedGPT |
For completeness, we include all downstream task results we collected for this study. Table 8 includes
upstream Pile evaluations and all downstream zero-shot tasks for models GPT-J, GPT-NeoX, OPT, and
Pythia, as well as Cerebras-GPT. Similarly, Table 9 shows the few-shot (five-shot) results for all models.
Full downstre... | Cerebras-GPT- Open Compute-Optimal Language Models Trained on the Cerebras Wafer-Scale Cluster |
6 FUTURE DIRECTIONS
Many studies have been conducted to tackle the hallucination problem in NLG and its downstream
tasks. As mentioned above, we have discussed common metrics and mitigation methods to advance
ACM Comput. Surv., Vol. 1, No. 1, Article . Publication date: February 2022.
Survey of Hallucination in Natu... | SurveyofHallucinationinNatural Language Generation |
List the prime numbers be-
tween 20 and 30.
Here is a simple algorithm which uses trial division to find
the prime numbers within a given interval.
primes = []
for num in range(20, 31):
found = False
for j in range(2,num//2):
if ((num % j)==0):
found = True
break
if((found==False) & (num!= 1)):
primes.append(num... | StarCoder_paper (1) |
1
Introduction | CAMEL- Communicative Agents for “Mind” Exploration of Large Scale Language Model Society |
pharmaceuticals), but that doesn’t mean we see a race to the bottom on safety, possibly because
various mechanisms—market forces, regulation, legal liability—have raised the “bottom” sufficiently
high;136 and many of these mechanisms function to incorporate various potential externalities as
well.137 And more generally:... | Is Power-Seeking AI an Existential Risk? |
Anonymous. SALMONN: Towards generic hearing abilities for large language models. In Submitted to The
Twelfth International Conference on Learning Representations, 2023. under review.
Junyi Ao, Rui Wang, Long Zhou, Chengyi Wang, Shuo Ren, Yu Wu, Shujie Liu, Tom Ko, Qing Li, Yu Zhang, et al.
Speecht5: Unified-modal encod... | Qwen-Audio |
propose a body reference optimization method to improve the parametric model estimation accuracy and to enhance the consistency
between the parametric model and the implicit function. With the PaMIR representation, our framework can be easily extended to
multi-image input scenarios without the need of multi-camera cali... | PaMIR- Parametric Model-Conditioned Implicit Representation for Image-based Human Reconstruction |
new technology is a question that is best led—in an informed
way—by scholars, advocates, and lawmakers from outside
the core technical R&D community. | Eight Things to Know about Large Language Models |
MuslimChristianAtheistBuddhistHinduJewStackExchangeOpenWebText2Pile-CCFreeLawBooks3Wikipedia (en)HackerNewsOpenSubtitlesPubMed CentralPubMed AbstractsEnron EmailsBookCorpus2USPTO BackgroundsArXivYoutubeSubtitlesGithubPhilPapersGutenberg (PG-19)EuroParlUbuntu IRCNIH ExPorterDM Mathematics1.000.750.500.250.000.250.500.75... | The Pile- An 800GB Dataset of Diverse Text for Language Modeling |
(cid:10)
, t
(cid:10)
, (cid:2)(cid:4) ∈ E2}.
The M&S method can now be described as follows within our framework. Let F = (cid:3)V , D, A(cid:4) be a SAS+ frame with STG
G(F ). Assume V = {v1, . . . , vn}. For each vi ∈ V , compute the SAS+ frame Fi as the VP abstraction of F with V C = {vi}
= G(Fi). Each Gvi ... | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
per’s, and the pure-human performance is only a fraction
of a percentage point better than Whisper’s. These results
indicate that Whisper’s English ASR performance is not
perfect but very close to human-level accuracy. | RobustSpeechRecognitionviaLarge-ScaleWeakSupervision |
2.3 Generator: BART
The generator component pθ(yi|x, z, y1:i−1) could be modelled using any encoder-decoder. We use
BART-large [32], a pre-trained seq2seq transformer [58] with 400M parameters. To combine the input
x with the retrieved content z when generating from BART, we simply concatenate them. BART was
pre-traine... | Retrieval-AugmentedGenerationfor Knowledge-IntensiveNLPTasks |
Assistant Message:
You’re welcome! It was my pleasure to assist
you. Don’t hesitate to contact me if you need any
further assistance. Have a great day!
User Message:
Thank you, you too!
Assistant Message:
You’re welcome! Goodbye!
User Message:
Goodbye!
Assistant Message:
Goodbye!
User Message:
Goodbye!
Assistant ... | CAMEL- Communicative Agents for “Mind” Exploration of Large Scale Language Model Society |
Interestingly,
the (non-animatable) generative model
EG3D achieves reasonable FID despite not modeling artic-
ulation. This is because the FID evaluation only considers
training poses and views, see also Sup. Mat.
Geometry: Our method yields significantly better geome-
try compared to EVA3D, as evidenced by the improve... | AG3D- Learning to Generate 3D Avatars from 2D Image Collections |
etal biases, or hallucinations - has the potential to cast doubt on the whole information environment,
threatening our ability to distinguish fact from fiction.[54] This could disproportionately benefit
those who stand to gain from widespread distrust, a phenomenon scholars Chesney and Citron refer
to as Liars Dividend i... | gpt-4-system-card |
Broader impact A high-quality and versatile generalist speech generation model like Voicebox
can enable many applications that improve the quality of our life. For example, zero-shot TTS could
bring the voice back to people who suffer from diseases or underwent surgeries such as laryngectomy
the causes inability to spe... | Voicebox-Text-GuidedMultilingual UniversalSpeechGenerationatScale |
# Docs
15.5 M
1.1 M
3.6 M
Table 8: Hyper-parameters of domain-adaptive pre-training.
Hyperparameter
Computing infrastructure
Runtime
Number of steps
Batch size
Maximum sequence length
Maximum learning rate
Optimizer
Adam beta weights
Learning rate scheduler
Weight decay
Warmup steps
Gradient clipping
Dropout ratio
A... | ADAPTINGLARGELANGUAGEMODELSVIA READINGCOMPREHENSION |
5.5. Ablation studies
Figure 6. Pose correction and non-rigid motion improve novel
view synthesis. Pose correction straightens the right arm and adds
details (red arrows in (b) vs (c)) and non-rigid deformation im-
proves clothing alignment and shape (green arrows in (c) vs. (d)).
Table 3 illustrates that skeletal de... | HumanNeRF- Free-viewpoint Rendering of Moving People from Monocular Video |
Large language models (LLMs) are central to
modern natural language processing, delivering
exceptional performance in various tasks. How-
ever, their intensive computational and memory
requirements present challenges, especially
for devices with limited DRAM capacity. This
paper tackles the challenge of efficiently run... | LLM in a flash |
C. Approach and Results
C.1. Ensembling
Ensembling is another approach we tried to more effectively search in the program space. To ensemble
a set of models, we pool their samples together before running filtering and clustering. Since different
models can have different strengths and weaknesses, ensembling them can increa... | alphacode |
initialization is required for stable training of the adapted
model; we investigate this empirically in Section 3.6. By
initializing the adapters to a near-identity function, original
network is unaffected when training starts. During training,
the adapters may then be activated to change the distribution
of activation... | Parameter-Efficient Transfer Learning for NLP |
QUESTION: Gretchen has 110 coins. There are 30 more gold coins than silver coins. How many gold coins
does Gretchen have?
MODEL ANSWER (INCORRECT; SEMANTIC UNDERSTANDING ERROR): Gretchen has 110 coins. There
are 30 more gold coins than silver coins. So there are 110 - 30 = 80 silver coins. So there are 80 silver coins ... | Chain-of-Thought Prompting Elicits Reasoning in Large Language Models |
Neuroticism. Human neuroticism is strongly positively correlated with negative affect
and moderately negatively correlated with positive affect in human research [103]. IPIP-NEO
Neuroticism data for all models, except those for PaLM 62B, showed excellent evidence of
external validity in their relation to PANAS Positive... | PersonalityTraitsinLargeLanguageModels |
We investigate whether human survey responses can be approximated by language models trained on particular media
diets. To create media diet models, we start with a language model (e.g. the popular BERT model), and finetune it on a
media diet dataset. This adaptation allows the model to absorb new information, while als... | Language models trained on media diets can predict public opinion |
g(a), g(b)
g(a)
u, v
g(b)
g(a)
g(a)
u, v
G2:
g(b)
v
g(a), g(b)
g(a)
v
a
a
a
u, v
b
u, v
u, v
b
u, v
a
u, v
b
u, v
G1:
G1:
Fig. 5. An ABS abstraction (left) and an VP abstraction (right) of the same instance (V C = {v}).
Example 31. Let V 1 = {u, v}, where both variables are binary, let A1... | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
explanations for their forming. With the rise of Linked Data, several approaches suggested the use of links across datasets
explain sequence patterns [47–49] through graph exploration. This idea was also explored to explain data to a non-expert
audience, especially through augmenting tabular data and statistical anal... | Knowledge graphs as tools for explainable machine learning: A survey |
2023), as well as the associated training data (Gao et al., 2021a).
Solaiman (2023) explains how the degree of openness in the LLM development process is connected
to the potential risks associated with a model release. When systems are developed in a fully closed
manner, it’s more likely that power gets concentrated a... | StarCoder_paper (1) |
able them to process ultra-long texts without
any modification or fine-tuning. Experimen-
tal results show that our SCM system enables
LLMs, which are not optimized for multi-turn
dialogue, to achieve multi-turn dialogue capa-
bilities that are comparable to ChatGPT, and
to outperform ChatGPT in scenarios involving
ultra... | Unleashing Infinite-Length Input Capacity for Large-scale Language Models with Self-Controlled Memory System |
(56.0% vs 51.5%). For each group in BOLD, a higher average sentiment score means more positive
sentiments and a lower standard deviation indicates less bias within the group. Overall, Mixtral
displays more positive sentiments than Llama 2, with similar variances within each group. | Mixtral of Experts paper |
17.8
12.8
7.4
10.5
9.2
17.7
11.9
8.4
8.3
14.9
15.7
4.4
4.3
17.7
C
a
l
l
H
o
m
e
24.1
30.0
20.9
23.0
18.2
17.5
16.2
16.4
15.8
17.6
46.9
40.6
29.1
34.8
23.4
59.7
38.9
29.6
29.3
54.5
58.1
11.3
12.0
46.5
S
w
i
t
c
h
b
o
a
r
d
17.8
22.8
15.2
16.8
15.7
14.5
14.1
14.0
13.1
13.8
40.2
32.9
22.2
28.3
19.8
56.1
33.0
22.8
... | RobustSpeechRecognitionviaLarge-ScaleWeakSupervision |
9
We consider cross-sentence prompting where a 3 second clip from another sample of the same speaker
is used as audio context, and continuation where the first 3 seconds of each utterance is used.
We ran subjective MOS studies comparing ground truth, YourTTS, and Voicebox. A3T is not included
because of the bad perfo... | Voicebox-Text-GuidedMultilingual UniversalSpeechGenerationatScale |
In International conference on machine learning, pages 199–207. PMLR, 2013.
[4] Christopher M Bishop and Nasser M Nasrabadi. Pattern recognition and machine learning, volume 4.
Springer, 2006.
[5] Leo Breiman. Random forests. Machine learning, 45:5–32, 2001.
[6] Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbi... | MLCopilot- Unleashing the Power of Large Language Models in Solving Machine Learning Tasks |
4.2 Qualitative Results | HuggingGPT- Solving AI Tasks with ChatGPT and its Friends in Hugging Face |
Language models can explain neurons in language models
W in
https://openaipublic.blob.core.windows.net/neuron-explainer/paper/index.html
10/32 | Language models can explain neurons in language models |
like max depth to prevent overfit. When a user comes with a novel target task ˜T , which has never been seen in history, MLCopilot | MLCopilot- Unleashing the Power of Large Language Models in Solving Machine Learning Tasks |
We introduce a free-viewpoint rendering method – Hu-
manNeRF – that works on a given monocular video of a hu-
man performing complex body motions, e.g. a video from
YouTube. Our method enables pausing the video at any
frame and rendering the subject from arbitrary new cam-
era viewpoints or even a full 360-degree camer... | HumanNeRF- Free-viewpoint Rendering of Moving People from Monocular Video |
with state-of-the-art performance, 2023.
[204] Serban, I. V., R. Lowe, L. Charlin, et al. Generative deep neural networks for dialogue: A
short review. CoRR, abs/1611.06216, 2016.
[205] Vinyals, O., Q. V. Le. A neural conversational model. CoRR, abs/1506.05869, 2015.
59
[206] Adiwardana, D., M. Luong, D. R. So, e... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
cases. Notably, for example, we might actively search for inputs that will reveal problematic
objectives,103 and we might learn how to read off a system’s objectives from its internal
states.104 | Is Power-Seeking AI an Existential Risk? |
3.3.3 Embodied Action
In the pursuit of Artificial General Intelligence (AGI), the embodied agent is considered a pivotal
paradigm while it strives to integrate model intelligence with the physical world. The Embodiment
hypothesis [357] draws inspiration from the human intelligence development process, posing that an
... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
Decebal Constantin Mocanu, Elena Mocanu, Peter Stone, Phuong H Nguyen, Madeleine Gibescu,
and Antonio Liotta. Scalable training of artificial neural networks with adaptive sparse connectiv-
ity inspired by network science. Nature Communications, 2018. URL http://www.nature.
com/articles/s41467-018-04316-3.
Pavlo Molcha... | JAXPRUNER |
Minerva was trained on 38.5B tokens of arXiv documents and webscrape
pages with LaTeX content, while MathMix consists of a smaller set of 1.5B
tokens containing individual math problems and their solutions, free-form text
discussing math problems and concepts, and synthetic data (Table 2). While
Minerva was pretrained ... | Let’s Verify Step by Step |
Wearable, Discreet Augmentative and Alternative Communication
Supervisor: Dr Timothy Neate
Please note: applicants to this proposal are welcome to self-fund, or apply for the studentship or K-CSC
Scholarship.
Approximately 2.2 million people in the UK experience a form of communication impairment [1],
includin... | informatics-phd-projects-2022-23 |
{" role ": " user ", " content ": " Create an imaginative image descriptive caption or modify an earlier caption \
for the user input : " make the light red "} ,
{" role ": " assistant ", " content ": "a pale figure with long white hair stands in the center of a dark forest , \
holding a sword high above his head .
th... | Improving Image Generation with Better Captions |
.
M
a
x
i
m
u
m
I
n
n
e
r
P
r
o
d
u
c
t
S
e
a
r
c
h
(
M
I
P
S
)
T
h
e
e
x
t
e
r
n
a
l
m
e
m
o
r
y
c
a
n
a
l
l
e
v
i
a
t
e
t
h
e
r
e
s
t
r
i
c
t
i
o
n
o
f
f
i
n
i
t
e
a
t
t
e
n
t
i
o
n
s
p
a
n
.
A
s
t
a
n
d
a
r
d
p
r
a
c
t
i
c
e
i
s
t
o
s
a
v
e
t
h
e
e
m
b
e
d
d
i
n
g
r
e
p
r
... | LLM Powered Autonomous Agents _ Lil'Log |
generative agents produce large streams of events and memories
that must be retained, a core challenge of our architecture is to
ensure that the most relevant pieces of the agent’s memory are
retrieved and synthesized when needed. | Generative Agents- Interactive Simulacra of Human Behavior |
lightweight self-attention mechanism in neural networks. In ISCA. 692–705.
[98] Aaron Harlap, Deepak Narayanan, Amar Phanishayee, Vivek Seshadri, Nikhil Devanur, Greg Ganger, and Phil Gibbons. 2018. Pipedream: Fast and efficient
pipeline parallel dnn training. arXiv preprint arXiv:1806.03377 (2018).
[99] Michael Has... | TheEfficiencySpectrumofLargeLanguageModels-AnAlgorithmicSurvey |
Adapter Layers Introduce Inference Latency There are many variants of adapters. We focus
on the original design by Houlsby et al. (2019) which has two adapter layers per Transformer block
and a more recent one by Lin et al. (2020) which has only one per block but with an additional
LayerNorm (Ba et al., 2016). While on... | LORA |
6 Future Work
FinLLMs, or Financial Large Language Models, present a
vision of the future where personalized robo-advisors or as-
sistants are within everyone’s reach. It aims to democratize
access to high-quality financial advice, leveraging advanced
language modeling techniques to make sense of vast amounts
of finan... | FinGPT-Open-SourceFinancialLargeLanguageModels |
[107] Pavel Denisov and Ngoc Thang Vu. 2020. Pretrained Semantic Speech Embeddings for End-to-End Spoken Language
Understanding via Cross-Modal Teacher-Student Learning. In Interspeech.
[108] Brecht Desplanques, Jenthe Thienpondt, and Kris Demuynck. 2020. Ecapa-tdnn: Emphasized channel attention,
propagation and agg... | AReviewofDeepLearningTechniquesforSpeechProcessing |
Benchmark
# Prompts
Judge
Model
GPT-4
Guanaco-65B
Guanaco-33B
ChatGPT-3.5 Turbo
Vicuna-13B
Guanaco-13B
Guanaco-7B
Bard
Vicuna
80
Human raters
Elo Rank
Vicuna
80
GPT-4
Open Assistant
953
GPT-4
Elo Rank
Rank
Median Rank
1176
1023
1009
916
984
975
1010
909
1
2
4
7
5
6
3
8
1348
1022
992
966
974
913
879
902
1... | QLORA |
data-more-available/, 2016. (cited on p. 17)
Shaden Smith, Mostofa Patwary, Brandon Norick, Patrick LeGresley, Samyam Rajbhandari, Jared
Casper, Zhun Liu, Shrimai Prabhumoye, George Zerveas, Vijay Korthikanti, Elton Zhang, Rewon
Child, Reza Yazdani Aminabadi, Julie Bernauer, Xia Song, Mohammad Shoeybi, Yuxiong He,
Mic... | StarCoder_paper (1) |
Object Tracking Q: Alice, Bob, and Claire are holding a white elephant gift exchange. At the start of the event, they are each holding a
present of a different color: Alice has a blue present, Bob has a black ball, and Claire has a green present. As the event
progresses, pairs of people swap gifts. First, Alice and Bob... | Enhancing Chain-of-Thoughts Prompting with Iterative Bootstrapping in Large Language Models |
22 | Beyond Efficiency |
Writing a DPhil Research Proposal
A Research Proposal in the field of international development consists of a plan for research and for
writing a thesis, including: (a) the specification of a set of research questions or a statement of
problems to be analysed, (b) a discussion of relevant existing s... | Writing a DPhil Research Proposal |
At the same time, we corrupt the original audio with
a random amount of noise, and train our 1D U-Net
(introduced in Section 3.1.4) to remove that noise.
During the noise removal process, we condition the
U-Net on the noise level and the compressed latent,
which can have access to a reduced version of the
non-noisy aud... | Moûsai |
token of the l’th intermediate transformer layer.
These contextual embeddings are used to compute
query vectors that interface with the entity memory.
For each context mention mi = (emi, smi, tmi)
in q, we form a query vector to access the Entity
memory by concatenating the context embeddings
for the mention mi’s start... | Adaptable and Interpretable Neural Memory Over Symbolic Knowledge |
While the sinusoid extrapolation could be easily performed with standard regression techniques, we
contextualize the task with another method that has no specific prior knowledge of the function being
extrapolated. We include the structure learning baseline from [90], implemented with Gen [90]. This
method uses a Gauss... | LargeLanguageModelsasGeneralPatternMachines |
should scale up the number of tasks and the size of the model even further. Second, we study the effect of
finetuning on the ability of the models to perform reasoning tasks. Our experiments show that whereas
prior instruction finetuning methods that do not include chain-of-thought (CoT; Wei et al., 2022b) severely
degrad... | Scaling Instruction-Finetuned Language Models |
[40] combines 56 datasets spanning across 8 tasks and 112 languages. Experiments show no model
can achieve state-of-the-art results on all embedding tasks yet. In this paper, we do not use the
SentEval toolkit since its linear probing setup depends on the optimization hyperparameters.
Most closely related to our work i... | E5 |
Multi-Head AttentionLayer NormalizationLayer NormalizationFeed-Forward Network++Adapter NetworkAdapter NetworkNonlinear Activation+Down-projectionUp-projectionkdkdMulti-Head AttentionLayer NormalizationLayer NormalizationFeed-Forward Network++Hidden StatesWqWkWvQKVPkPvAttentionPrefix-tuningMulti-Head AttentionLayer Nor... | Parameter-EfficientFine-TuningMethods |
problem, and judging if a partial or incorrect program is useful is a difficult challenge. Therefore, most
prior work has been limited to either restricted domain-specific programming languages (Gulwani,
2011) or short code snippets (Bruch et al., 2009; Raychev et al., 2014).
Recent large-scale transformer-based (Vaswani ... | alphacode |
language model. We show that, (2) while current state-of-
the-art general-purpose visual-language models out-of-the-
box (zero-shot) do not well address embodied reasoning
problems, it is possible to train a competent general-purpose
visual-language model that is also an efficient embodied
reasoner. In studying how to b... | PaLM-E- An Embodied Multimodal Language Model |
We finally want to point out that our purpose is not to invent new abstraction methods but to enable formal analyses
and comparisons of various methods—both existing methods and methods yet to be invented. That said, we hope that an
increased understanding of state abstraction will inspire to the invention of new and ... | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
Dataset
LJ Speech
VCTK
CMOS
-0.106
-0.262
D. Voice Conversion
In the multi-speaker setting, we do not provide speaker identities into the text encoder, which makes the latent variables
estimated from the text encoder learn speaker-independent representations. Using the speaker-independent representations,
we can tran... | ConditionalVariationalAutoencoderwithAdversarialLearningfor End-to-EndText-to-Speech |
Indonesian
Original text: Saya tidak ingin melihatnya lagi.
Rewritten text: sy tdk ingan mlhtnya lg
Original text: Aku tahu segalanya.
Rewritten text: gw tau smuanya
Original text: Saya memiliki model komputer
baru untuk bahasa Indonesia yang dapat mema-
hami dan menghasilkan teks informal yang fasih.
Rewritten text:
... | PaLM 2 Technical Report |
000ft.Action6:Finish[1,800to7,000ft]Query:WhatcountryistheSelunlocatedin?Trace:Thought1:IneedtosearchSelunandfindthecountryitislocatedin.Action1:74 A.17 | Tool Learning with Foundation Models |
3.3
Introducing faithfulness based loss
function
Creating a metric to gauge how closely a model’s
outputs match input data or ground truth is the
task of this section. In this sense, faithfulness de-
scribes the model’s capacity to faithfully and prop-
erly reflect data from the input without adding er-
rors, omissi... | AComprehensiveSurveyofHallucinationMitigationTechniquesinLarge LanguageModels |
DRL pre-training has shown promise in reducing training time and enhancing performance
in various Human-Computer Interaction (HCI) applications, including speech recognition [451].
Recently, researchers have suggested using a reinforcement learning algorithm to develop a Speech
Enhancement (SE) system that effectively ... | AReviewofDeepLearningTechniquesforSpeechProcessing |
text-to-speech. In International Conference on Learning Representations.
[297] Quentin Lemaire and Andre Holzapfel. 2019. Temporal convolutional networks for speech and music detection in
radio broadcast. In 20th International Society for Music Information Retrieval Conference, ISMIR 2019, 4-8 November
2019. Internati... | AReviewofDeepLearningTechniquesforSpeechProcessing |
16
Dynabench 4). Robust and effective adversarial testing for large language models is still an open problem space with
varied results due to the challenges of generalization in evaluation samples [102].
A limitation of our approach is that most of the participants are able to find commonly occurring problems, but not... | LaMDA- Language Models for Dialog Applications |
To quantify this difference, we examine both overall ro-
bustness, that is average performance across many distribu-
tions/datasets, and effective robustness, introduced by Taori
et al. (2020), which measures the difference in expected
performance between a reference dataset, which is usually
in-distribution, and one o... | RobustSpeechRecognitionviaLarge-ScaleWeakSupervision |
We take special care in how we sample from our dataset. Though our dataset is resampled to 44kHz,
the data within it may be band-limited in some way. That is, some audio may have had an original
sampling rate much lower than 44kHz. This is especially prevalent in speech data, where the true
sampling rates of the underl... | RVQGAN |
data. In this section, we provide the details of the extraction of weakly-associated conditions. | Let’sThinkOutsidetheBox |
described by a prompt. We curate 25 queries as the initial prompt, together with subsequent queries for
modifying that image.
3D Model Construction. We investigate three-dimensional (3D) modeling by manually devising a collection
of APIs that leverage the capabilities of the sophisticated 3D rendering engine Taichi19. ... | Tool Learning with Foundation Models |
70 See, e.g., Matthias Niessner’s YouTube video “Face2Face: Real-time face capture and reenact-
ment of RGB videos (CVPR 2016 Oral),” (www.youtube.com/watch?v=ohmajJTcpNk),
demonstrating the use of machine learning to create believable simulations of political leaders
speaking.
https://doi.org/10.1017/9781108890960 Pu... | Social_Media_and_Democracy |
One potential way out of this problem would be to establish a rule or norm
prohibiting academic researchers from accepting funding from the platforms.
However, there are costs to this approach as well – namely, funding for social
media research is quite scarce and the platforms have money. Moreover, it is a
legitimate ... | Social_Media_and_Democracy |
2DAVIS
license:
nc/4.0/deed.en
https://creativecommons.org/licenses/by-
Human side-by-side evaluations comparing
Figure 6.
VideoPoet with recent leading text-to-video generative mod-
els. Green, gray, and pink bars represent the proportion of trials
where VideoPoet was preferred over an alternative, similar to, or
... | VideoPoet |
7.77
8.05
8.25
7.13
7.43
7.45
6.5
6.57
6.53
5.83
5.90
6.28
7.78
8.16
8.16
6.38
6.75
7.02
Figure 24: Model performance with different number of attention heads
in terms of how well they can ”speak and understand natural language”. Most existing benchmarks for generative
LMs, such as LAMBADA [22], CLOZE [29], TriviaQ... | TinyStories-HowSmallCanLanguageModelsBeandStillSpeak CoherentEnglish? |
✩ This paper is part of the Special Issue on Explainable AI.
* Corresponding author.
E-mail addresses: i.tiddi@vu.nl (I. Tiddi), k.s.schlobach@vu.nl (S. Schlobach).
https://doi.org/10.1016/j.artint.2021.103627
0004-3702/© 2021 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY li... | Knowledge graphs as tools for explainable machine learning: A survey |
Sarthak Jain and Byron C. Wallace. 2019. Attention is
not Explanation. In Proceedings of the 2019 Con-
ference of the North American Chapter of the Asso-
ciation for Computational Linguistics: Human Lan-
guage Technologies, Volume 1 (Long and Short Pa-
pers), pages 3543–3556, Minneapolis, Minnesota.
Association for Com... | Measuring Association Between Labels and Free-Text Rationales |
a simple CartPole controller, and optimize simple trajectories via human-in-the-loop “clicker” reward
training. Code, benchmarks, and videos are made available at https://general-pattern-machines.github.io. | LargeLanguageModelsasGeneralPatternMachines |
Hence, we propose a novel U-Net with only 1D
convolutional kernels, which is more efficient than
the original 2D architecture in terms of speed, and
can be successfully used both on waveforms or on
spectrograms if each frequency is considered as a
different channel.
Moreover, we infuse our 1D U-Net with multi-
ple new... | MOUSAI |
to me!”.
The man said, ”No, it belongs to me. I must give it back.”
Katie thought for a moment and then said, ”I know! I will give it back to the cat.”
The man smiled and said, ”That’s very kind of you. Here, take it back.”
Katie thanked the man and took the cat home. She was so happy to have her own black cat back.
t... | TinyStories-HowSmallCanLanguageModelsBeandStillSpeak CoherentEnglish? |
Most Artificial Intelligence (AI) systems operate using very complex criteria, which makes it difficult for humans to
understand and interpret the rationale behind the decisions made by a given model [1]. The proliferation of AI systems
has increased the need for models that can generate ‘explanations’, as such explana... | Knowledge-graph-based explainable AI- A systematic review |
In this paper, we propose an end-to-end Retrieval-
Augmented Visual Language Model (REV EAL) that learns
to encode world knowledge into a large-scale memory, and
to retrieve from it to answer knowledge-intensive queries.
REV EAL consists of four key components: the memory, the
encoder, the retriever and the generator. ... | REVEAL-Retrieval-AugmentedVisual-LanguagePre-Trainingwith Multi-SourceMultimodalKnowledgeMemory |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.