Progress & Compress: A scalable framework for continual learning (2018 · 338 citations)
Neural networks that learn tasks one after another tend to lose earlier skills, a problem called catastrophic forgetting.
A Survey of Reinforcement Learning Informed by Natural Language (2019 · 231 citations)
Standard reinforcement learning (RL) agents learn from scratch. They need huge numbers of environment interactions and transfer poorly to even slightly altered tasks. This survey, by authors from Oxford, Facebook AI Research, MIT and UCL, asks whether natural language can supply the priors and structure these agents…
Understanding the Effects of RLHF on LLM Generalisation and Diversity (2023 · 143 citations)
Kirk et al. examine what each stage of the standard RLHF pipeline contributes to two properties of fine-tuned LLMs. The stages are supervised fine-tuning (SFT), reward modelling and PPO-based reinforcement learning. The two properties are handling inputs unlike the training data (OOD generalisation) and the variety…

Co-authors here

11 co-authors with a page, the most papers together first.

Venues

Where their works here were published, the most works first, with the editions of those that have a page.

Papers of our library cite their works 239 times (of the 149,439 papers whose bibliographies we read); 47 citing papers with a page here →.

Updated 2 November 2026