Research map / Reinforcement learning / Policy gradient and actor-critic
TTOpt: A Maximum Volume Quantized Tensor Train-based Optimization and its Application to Reinforcement Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 TTOpt instead of GA | policy weight optimization · HalfCheetah-v3, N=3 | mean cumulative reward: 2495.37 -> 4211.02 | Helped |
| Use TTOpt instead of GA | policy weight optimization · LunarLanderContinuous-v2, N=256 | mean cumulative reward: 213.75 -> 286.87 | Helped |
| Use TTOpt instead of GA | policy weight optimization · Swimmer-v3, N=256 | mean cumulative reward: 359.79 -> 311.82 | Hurt |
| Use TTOpt instead of GA | policy weight optimization · HalfCheetah-v3, N=256 | mean cumulative reward: 3085.80 -> 2935.90 | Hurt |
| Use TTOpt instead of GA | policy weight optimization · Swimmer-v3, N=3 | mean cumulative reward: 349.91 -> 357.50 | Helped |
| Quantize tensor modes rather than use direct TT modes (ablation) | TTOpt · F1 (Ackley), mode size 65536 | final error ε: 2.0e+01 -> 2.0e-03 | Helped |
| Quantize tensor modes rather than use direct TT modes (ablation) | TTOpt · F2 (Alpine), mode size 65536 | final error ε: 1.9e+01 -> 1.3e-04 | Helped |
| Quantize tensor modes rather than use direct TT modes (ablation) | TTOpt · F9 (Schaffer), mode size 65536 | final error ε: 4.4e+00 -> 1.4e-01 | Hurt |
| Increase mode size using QTT (ablation) | TTOpt · F1 (Ackley), 10 dimensions | final error ε: 1.2e+00 -> 1.3e-04 | Helped |
| Increase TT rank beyond the moderate range (ablation) | TTOpt · benchmark functions | final error ε | Hurt |
| Use small TT ranks (1 or 2) instead of a larger rank (ablation) | TTOpt · benchmark functions | final error ε | Hurt |
| Increase the number of objective-function requests (ablation) | TTOpt · benchmark functions | final error ε | Helped |
Closest papers by meaning
- Optimizing Tensor Network Contraction Using Reinforcement Learning (ICML 2022 · similarity 0.86)
- Generalized Policy Iteration using Tensor Approximation for Hybrid Control (ICLR 2024 · similarity 0.86)
- Provable and Practical: Efficient Exploration in Reinforcement Learning via Langevin Monte Carlo (ICLR 2024 · similarity 0.84)
- Symmetric Replay Training: Enhancing Sample Efficiency in Deep Reinforcement Learning for Combinatorial Optimization (ICML 2024 · similarity 0.84)
- Bigger, Regularized, Optimistic: scaling for compute and sample efficient continuous control (NeurIPS 2024 · similarity 0.84)
- Revisiting Scalable Hessian Diagonal Approximations for Applications in Reinforcement Learning (ICML 2024 · similarity 0.84)
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.