Research map / Reinforcement learning / Visual and representation RL
A Natural Lottery Ticket Winner: Reinforcement Learning with Ordinary Neural Circuits
Ramin Hasani, Mathias Lechner, Alexander Amini, Daniela Rus, Radu Grosu
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 TW-constrained wiring (ONC) instead of random circuits with same size (ablation) | ONC with ARS · Half-Cheetah | return | Improved |
| Use TW-constrained wiring (ONC) instead of random circuits with same size (ablation) | ONC with ARS · Inverted Pendulum | return | Improved |
| Use TW-constrained wiring (ONC) instead of random circuits with same size (ablation) | ONC with ARS · Mountaincar | return | Improved |
| Use ONC (77% sparse TW circuit) instead of fully connected MLP/LSTM | ONC with ARS · Half-Cheetah | return: + about 1000 over LSTM (2587.4 vs 1588.9) | Improved |
| Use ONC instead of fully connected MLP/LSTM | ONC with ARS · Inverted Pendulum and Mountaincar | return | No clear effect |
| Replace PPO/A2C policy gradient with ARS search-based RL for the ONC | ONC · Inverted Pendulum and MountainCar | return | No clear effect |
| Use TW-constrained wiring in place of random wiring (Algorithm 1 variants) (ablation) | Max-flow of sparse random networks · Max-flow rate vs TW (40000 networks) | max-flow ratio to TW: ratios 1.15 (Fig 2A) down to 1.01 (Fig 2D); randomly wired networks have higher ratios, so TW is not the max | Mixed |
Closest papers by meaning
- Neural Circuit Architectural Priors for Embodied Control (NeurIPS 2022 · similarity 0.87)
- Adaptive Surrogate Gradients for Sequential Reinforcement Learning in Spiking Neural Networks (NeurIPS 2025 · similarity 0.86)
- Feedback control guides credit assignment in recurrent neural networks (NeurIPS 2024 · similarity 0.84)
- Efficient Reinforcement Learning by Discovering Neural Pathways (NeurIPS 2024 · similarity 0.84)
- Real-Time Recurrent Reinforcement Learning (AAAI 2025 · similarity 0.83)
- Slow and Steady Wins the Race: Maintaining Plasticity with Hare and Tortoise Networks (ICML 2024 · similarity 0.83)
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.
