Understanding plasticity in neural networks

C Lyle, Z Zheng, E Nikishin, BA Pires… - International …, 2023 - proceedings.mlr.press
Plasticity, the ability of a neural network to quickly change its predictions in response to new
information, is essential for the adaptability and robustness of deep reinforcement learning …

Pessimistic bootstrapping for uncertainty-driven offline reinforcement learning

C Bai, L Wang, Z Yang, Z Deng, A Garg, P Liu… - arXiv preprint arXiv …, 2022 - arxiv.org
Offline Reinforcement Learning (RL) aims to learn policies from previously collected
datasets without exploring the environment. Directly applying off-policy algorithms to offline …

Rorl: Robust offline reinforcement learning via conservative smoothing

R Yang, C Bai, X Ma, Z Wang… - Advances in neural …, 2022 - proceedings.neurips.cc
Offline reinforcement learning (RL) provides a promising direction to exploit massive amount
of offline data for complex decision-making tasks. Due to the distribution shift issue, current …

Plastic: Improving input and label plasticity for sample efficient reinforcement learning

H Lee, H Cho, H Kim, D Gwak, J Kim… - Advances in …, 2024 - proceedings.neurips.cc
Abstract In Reinforcement Learning (RL), enhancing sample efficiency is crucial, particularly
in scenarios when data acquisition is costly and risky. In principle, off-policy RL algorithms …

Deep reinforcement learning with plasticity injection

E Nikishin, J Oh, G Ostrovski, C Lyle… - Advances in …, 2024 - proceedings.neurips.cc
A growing body of evidence suggests that neural networks employed in deep reinforcement
learning (RL) gradually lose their plasticity, the ability to learn from new data; however, the …

Understanding and preventing capacity loss in reinforcement learning

C Lyle, M Rowland, W Dabney - arXiv preprint arXiv:2204.09560, 2022 - arxiv.org
The reinforcement learning (RL) problem is rife with sources of non-stationarity, making it a
notoriously difficult problem domain for the application of neural networks. We identify a …

Why neural networks find simple solutions: The many regularizers of geometric complexity

B Dherin, M Munn, M Rosca… - Advances in Neural …, 2022 - proceedings.neurips.cc
In many contexts, simpler models are preferable to more complex models and the control of
this model complexity is the goal for many methods in machine learning such as …

Pgx: Hardware-accelerated parallel game simulators for reinforcement learning

S Koyamada, S Okano, S Nishimori… - Advances in …, 2024 - proceedings.neurips.cc
We propose Pgx, a suite of board game reinforcement learning (RL) environments written in
JAX and optimized for GPU/TPU accelerators. By leveraging JAX's auto-vectorization and …

Small batch deep reinforcement learning

J Obando Ceron, M Bellemare… - Advances in Neural …, 2024 - proceedings.neurips.cc
In value-based deep reinforcement learning with replay memories, the batch size parameter
specifies how many transitions to sample for each gradient update. Although critical to the …

Hyperbolic deep reinforcement learning

E Cetin, B Chamberlain, M Bronstein… - arXiv preprint arXiv …, 2022 - arxiv.org
We propose a new class of deep reinforcement learning (RL) algorithms that model latent
representations in hyperbolic space. Sequential decision-making requires reasoning about …