Contrastive learning as goal-conditioned reinforcement learning

B Eysenbach, T Zhang, S Levine… - Advances in Neural …, 2022 - proceedings.neurips.cc
In reinforcement learning (RL), it is easier to solve a task if given a good representation.
While deep RL should automatically acquire such good representations, prior work often …

Neural networks trained with SGD learn distributions of increasing complexity

M Refinetti, A Ingrosso, S Goldt - … Conference on Machine …, 2023 - proceedings.mlr.press
The uncanny ability of over-parameterised neural networks to generalise well has been
explained using various" simplicity biases". These theories postulate that neural networks …

Spectral bias outside the training set for deep networks in the kernel regime

B Bowman, GF Montufar - Advances in Neural Information …, 2022 - proceedings.neurips.cc
We provide quantitative bounds measuring the $ L^ 2$ difference in function space between
the trajectory of a finite-width network trained on finitely many samples from the idealized …

Improving deep reinforcement learning by reducing the chain effect of value and policy churn

H Tang, G Berseth - arXiv preprint arXiv:2409.04792, 2024 - arxiv.org
Deep neural networks provide Reinforcement Learning (RL) powerful function
approximators to address large-scale decision-making problems. However, these …

Contrastive representation for data filtering in cross-domain offline reinforcement learning

X Wen, C Bai, K Xu, X Yu, Y Zhang, X Li… - arXiv preprint arXiv …, 2024 - arxiv.org
Cross-domain offline reinforcement learning leverages source domain data with diverse
transition dynamics to alleviate the data requirement for the target domain. However, simply …

Eigensubspace of temporal-difference dynamics and how it improves value approximation in reinforcement learning

Q He, T Zhou, M Fang, S Maghsudi - Joint European Conference on …, 2023 - Springer
We propose a novel value approximation method, namely “E igensubspace R egularized C
ritic (ERC)” for deep reinforcement learning (RL). ERC is motivated by an analysis of the …

Parallel -Learning: Scaling Off-policy Reinforcement Learning under Massively Parallel Simulation

Z Li, T Chen, ZW Hong, A Ajay… - … on Machine Learning, 2023 - proceedings.mlr.press
Reinforcement learning is time-consuming for complex tasks due to the need for large
amounts of training data. Recent advances in GPU-based simulation, such as Isaac Gym …

Learning dynamics and generalization in reinforcement learning

C Lyle, M Rowland, W Dabney, M Kwiatkowska… - arXiv preprint arXiv …, 2022 - arxiv.org
Solving a reinforcement learning (RL) problem poses two competing challenges: fitting a
potentially discontinuous value function, and generalizing well to new observations. In this …

Frequency and Generalisation of Periodic Activation Functions in Reinforcement Learning

AN Mavor-Parker, MJ Sargent, C Barry, L Griffin… - arXiv preprint arXiv …, 2024 - arxiv.org
Periodic activation functions, often referred to as learned Fourier features have been widely
demonstrated to improve sample efficiency and stability in a variety of deep RL algorithms …

Plastic Learning with Deep Fourier Features

A Lewandowski, D Schuurmans… - arXiv preprint arXiv …, 2024 - arxiv.org
Deep neural networks can struggle to learn continually in the face of non-stationarity. This
phenomenon is known as loss of plasticity. In this paper, we identify underlying principles …