REVIEW 4 cited by
CIC: Contrastive Intrinsic Control for Unsupervised Skill Discovery
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
We introduce Contrastive Intrinsic Control (CIC), an algorithm for unsupervised skill discovery that maximizes the mutual information between state-transitions and latent skill vectors. CIC utilizes contrastive learning between state-transitions and skills to learn behavior embeddings and maximizes the entropy of these embeddings as an intrinsic reward to encourage behavioral diversity. We evaluate our algorithm on the Unsupervised Reinforcement Learning Benchmark, which consists of a long reward-free pre-training phase followed by a short adaptation phase to downstream tasks with extrinsic rewards. CIC substantially improves over prior methods in terms of adaptation efficiency, outperforming prior unsupervised skill discovery methods by 1.79x and the next leading overall exploration algorithm by 1.18x.
Forward citations
Cited by 4 Pith papers
-
Task Adaptation from Skills: Information Geometry, Disentanglement, and New Objectives for Unsupervised Reinforcement Learning
The paper derives theoretical connections between skill disentanglement metrics and downstream task adaptation cost, and proposes Wasserstein-based objectives that can discover more (or all) optimal initial skills.
-
Weights or Skills? A Survey of Robot-Learning Techniques: from Action-Predicting Weights to Robots that Write their Own Skills
A taxonomy of robot learning on a weights-versus-skills axis, with a five-rung self-improvement ladder whose top cell (feedback plus memory plus search) holds only a few recent systems.
-
Epistemically-guided forward-backward exploration
Choosing exploration policies by the ensemble disagreement of forward-backward value estimates improves zero-shot RL sample efficiency on DeepMind Control Suite tasks.
-
RedRFT: A Light-Weight Benchmark for Reinforcement Fine-Tuning-Based Red Teaming
RedRFT is a new open-source benchmark with a unified PPO backbone, five reimplemented red teaming baselines, a proposed diversity metric, and ablation insights.
Discussion (0). Sign in to comment.