Research map / Probabilistic and scientific modeling / Molecules and proteins
Structuring Representations Using Group Invariants
Mehran Shakerinava, Arnab Kumar Mondal, Siamak Ravanbakhsh
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 changed | Where | Measured | Effect |
|---|---|---|---|
| Use SymReg embedding instead of AE world-model embedding | world model with MLP transition · Atari Pong | H@1: 23.8± 3.3 -> 45.2± 3.4 | Improved |
| Use SymReg embedding instead of C-SWM | world model · Atari Pong | H@1: 36.5± 5.6 -> 45.2± 3.4 | Improved |
| Use SymReg embedding instead of C-SWM | world model · Space Invaders | H@1: 48.5± 7.0 -> 54.2± 6.3 | Improved |
| Use SymReg embedding instead of C-SWM | world model · Space Invaders | MRR: 66.1± 6.6 -> 68.7± 5.1 | Improved |
| Use VAE instead of AE embedding for the world model | world model · Atari Pong | H@1: 23.8± 3.3 -> 1.0± 0.0 | Hurt |
| Use SymReg-decoupled (frozen encoder) instead of vanilla PPO | PPO · InvertedPendulum | average reward: 500± 150 -> 800± 180 | Improved |
| Fine-tune SymReg encoder during RL instead of freezing | SymReg · InvertedPendulum | average reward: 800± 180 -> 950± 50 | Improved |
| Fine-tune SymReg encoder during RL instead of freezing | SymReg · Swimmer | average reward: 21± 4.1 -> 31.5± 3.9 | Improved |
| Fine-tune SymReg encoder during RL instead of freezing | SymReg · Reacher | average reward: -14.5± 3.1 -> -10± 3.4 | Improved |
| Add a decoder to SymReg (Dec-SymReg) with frozen encoder | SymReg-decoupled · Reacher | average reward: -14.5± 3.1 -> -12.8± 2.7 | Improved |
| Add a decoder to SymReg (Dec-SymReg) with frozen encoder | SymReg-decoupled · InvertedPendulum | average reward: 800± 180 -> 600± 200 | Hurt |
| Use SymReg-decoupled instead of AE-decoupled | RL on frozen embedding · InvertedPendulum | average reward: 30± 15 -> 800± 180 | Improved |
Closest papers by meaning
- Equivariance by Contrast: Identifiable Equivariant Embeddings from Unlabeled Finite Group Actions (NeurIPS 2025 · similarity 0.91)
- Lie Group Decompositions for Equivariant Neural Networks (ICLR 2024 · similarity 0.89)
- Neural Fourier Transform: A General Approach to Equivariant Representation Learning (ICLR 2024 · similarity 0.88)
- MatrixNet: Learning over symmetry groups using learned group representations (NeurIPS 2024 · similarity 0.88)
- Learning symmetries via weight-sharing with doubly stochastic tensors (NeurIPS 2024 · similarity 0.88)
- Improving Equivariant Networks with Probabilistic Symmetry Breaking (ICLR 2025 · similarity 0.88)
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.
