AI Paper Atlas

Research map / Reinforcement learning / Visual and representation RL

Improving Computational Efficiency in Visual Reinforcement Learning via Stored Embeddings

Lili Chen, Kimin Lee, Aravind Srinivas, Pieter Abbeel

NeurIPS 2021 · Reinforcement learning › Visual and representation RL · approach: Robust visual encoders for RL · 6 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Freeze the encoder and store latent vectorsRainbow · Atari games at 45T FLOPsaverage returnImproved
Freeze the encoder and store latent vectorsRainbow · BankHeist at 45T FLOPsaverage return: 145.8 to 276.6Improved
Freeze the encoder and store latent vectorsRainbow · Qbert at 45T FLOPsaverage return: 2325.5 to 4123.5Improved
Freeze the encoder and store latent vectors, increasing replay capacityRainbow · Atari games at 500K environment steps, 0.07GBaverage returnMixed
Freeze the encoder and store latent vectors, increasing replay capacityRainbow · CrazyClimber at 500K environment steps, 0.07GBaverage return: 10498.0 to 17620.0Improved
Freeze the encoder and store latent vectors, increasing replay capacityRainbow · Qbert at 500K environment steps, 0.07GBaverage return: 2430.5 to 3231.0Improved
Freeze the encoder after 50000, 100000, or 150000 steps instead of not freezing (ablation)Rainbow+SEER · Amidarsample-efficiency: 50000, 100000, 150000No clear effect
Freeze the encoder earlier (ablation)Rainbow+SEER · Ataricompute-efficiency: T_f=150000Improved
Freeze more convolutional layers (ablation)CURL+SEER transfer from Walker-stand to Walker-walk · Walker-walk transfercompute-efficiencyImproved
Reduce image resolution from 100 × 100 to 50 × 50 (ablation)CURL · DeepMind Control Suitecompute-efficiency: 100 × 100 to 50 × 50Improved
Reduce image resolution from 84 × 84 to 42 × 42 (ablation)Rainbow · Atari gamesperformance: 84 × 84 to 42 × 42Hurt
Use a pretrained encoder and freeze it during fine-tuningCURL · Walker-stand to Walker-walk and Hopper-hop transfercompute-efficiency: 60K environment interactionsImproved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy