Defining levels of automated chemical design

B Goldman, S Kearnes, T Kramer, P Riley… - Journal of medicinal …, 2022 - ACS Publications
One application area of computational methods in drug discovery is the automated design of
small molecules. Despite the large number of publications describing methods and their …

A survey on uncertainty reasoning and quantification in belief theory and its application to deep learning

Z Guo, Z Wan, Q Zhang, X Zhao, Q Zhang, LM Kaplan… - Information …, 2024 - Elsevier
An in-depth understanding of uncertainty is the first step to making effective decisions under
uncertainty. Machine/deep learning (ML/DL) has been hugely leveraged to solve complex …

A survey of reinforcement learning from human feedback

T Kaufmann, P Weng, V Bengs… - arXiv preprint arXiv …, 2023 - arxiv.org
Reinforcement learning from human feedback (RLHF) is a variant of reinforcement learning
(RL) that learns from human feedback instead of relying on an engineered reward function …

Semantic anomaly detection with large language models

A Elhafsi, R Sinha, C Agia, E Schmerling… - Autonomous …, 2023 - Springer
As robots acquire increasingly sophisticated skills and see increasingly complex and varied
environments, the threat of an edge case or anomalous failure is ever present. For example …

Do bayesian neural networks need to be fully stochastic?

M Sharma, S Farquhar, E Nalisnick… - International …, 2023 - proceedings.mlr.press
We investigate the benefit of treating all the parameters in a Bayesian neural network
stochastically and find compelling theoretical and empirical evidence that this standard …

Selectively answering ambiguous questions

JR Cole, MJQ Zhang, D Gillick, JM Eisenschlos… - arXiv preprint arXiv …, 2023 - arxiv.org
Trustworthy language models should abstain from answering questions when they do not
know the answer. However, the answer to a question can be unknown for a variety of …

Truncation sampling as language model desmoothing

J Hewitt, CD Manning, P Liang - arXiv preprint arXiv:2210.15191, 2022 - arxiv.org
Long samples of text from neural language models can be of poor quality. Truncation
sampling algorithms--like top-$ p $ or top-$ k $--address this by setting some words' …

Deep ensembles work, but are they necessary?

T Abe, EK Buchanan, G Pleiss… - Advances in …, 2022 - proceedings.neurips.cc
Ensembling neural networks is an effective way to increase accuracy, and can often match
the performance of individual larger models. This observation poses a natural question …

On the practicality of deterministic epistemic uncertainty

J Postels, M Segu, T Sun, L Sieber, L Van Gool… - arXiv preprint arXiv …, 2021 - arxiv.org
A set of novel approaches for estimating epistemic uncertainty in deep neural networks with
a single forward pass has recently emerged as a valid alternative to Bayesian Neural …

Self-exploring language models: Active preference elicitation for online alignment

S Zhang, D Yu, H Sharma, H Zhong, Z Liu… - arXiv preprint arXiv …, 2024 - arxiv.org
Preference optimization, particularly through Reinforcement Learning from Human
Feedback (RLHF), has achieved significant success in aligning Large Language Models …