AI Paper Atlas

Research map / Multi-agent systems and games / Human-AI and goal-conditioned

The Sensory Neuron as a Transformer: Permutation-Invariant Neural Networks for Reinforcement Learning

Yujin Tang, David Ha

NeurIPS 2021 · Multi-agent systems and games › Human-AI and goal-conditioned · approach: World-model-based agents · 10 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
Use AttentionNeuron instead of a fixed-input FNN under shuffled observationsCartPoleSwingUpHarder · 5 observations, shuffledscore: 38 ± 120 -> 471 ± 426Improved
Accept extra duplicated input channels (ablation)AttentionNeuron · CartPoleSwingUpHarder, 10 observationsscore: 472 ± 426 -> 471 ± 425No clear effect
Add five noisy channels (ablation)AttentionNeuron · CartPoleSwingUpHarder, 5 observations + 5 noisescore: 472 ± 426 -> 461 ± 410No clear effect
Use AttentionNeuron instead of an FNN for shuffled observationsPyBullet Ant · shuffled observationsscore: 232 ± 112 -> 2576 ± 75Improved
Train the permutation-invariant student with behavior cloning rather than ES (ablation)PyBullet Ant · shuffled observationsscore: 2576 ± 75 -> 2034 ± 948Hurt
Increase the downstream network size for behavior cloning (ablation)PyBullet Ant AttentionNeuron student · shuffled observationsscore: 2034 ± 948 -> 2579 ± 457Improved
Use AttentionNeuron to construct a latent representation from shuffled screen patchesAtari Pong policy with CNN · shuffled screensscore: 21Improved
Train with randomly removed screen patches, then provide more patches at test time (ablation)Atari Pong AttentionNeuron policy · Atari PongscoreImproved
Use AttentionNeuron with AttentionAgent instead of AttentionAgent aloneCarRacing · KOF backgroundscore: -81 ± 4 -> 646 ± 189Improved
Use AttentionNeuron with AttentionAgent instead of AttentionAgent aloneCarRacing · Mt. Fuji backgroundscore: -57 ± 38 -> 503 ± 152Improved
Reshuffle observations every 25 steps rather than never reshuffling (ablation)AttentionNeuron policies · CartPolescore: 472 ± 426 -> 107 ± 146Hurt
Reshuffle observations every 25 steps rather than never reshuffling (ablation)AttentionNeuron policies · PyBullet Antscore: 2576 ± 75 -> 2053 ± 225Hurt

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