Research map / Architectures and efficiency / RNNs and normalization
Differentiable Weightless Controllers: Learning Logic Circuits for Continuous Control
Fabian Kresse, Christoph H. Lampert
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 DWC instead of full-precision network | SAC policy · Ant-v4 | return: 5598.251 -> 5677.335 | Improved |
| use DWC instead of full-precision network | SAC policy · HalfCheetah-v4 | return: 11528.543 -> 7548.604 | Hurt |
| use DWC instead of full-precision network | SAC policy · Hopper-v4 | return: 2796.813 -> 3119.865 | Improved |
| use DWC instead of quantized network | SAC policy · Walker2d-v4 | return: 4656.477 -> 5024.956 | Improved |
| use DWC instead of quantized network on FPGA | SAC policy · Ant FPGA energy per action | energy per action (J): 8.93e-7 -> 2.25e-09 | Improved |
| increase LUT layer width from 256 to 1024 (ablation) | DWC · HalfCheetah-v4 | return: 7054.791 -> 7548.604 | Improved |
| increase LUT layer width from 256 to 1024 (ablation) | DWC · Ant-v4 | return: 4542.495 -> 5677.335 | Improved |
| increase layer width to 16,384 LUTs and 255 input levels (ablation) | DWC · HalfCheetah-v4 | return: 7.5k -> above 10.3k | Improved |
| reduce LUT inputs k from 6 to 2 (ablation) | DWC · Ant, Hopper, Humanoid, Walker2d | return | No clear effect |
| use more thermometer input bits (5 to 63) (ablation) | DWC · MuJoCo tasks | return | Improved |
| add more LUT layers beyond one (ablation) | DWC (D=1024) · MuJoCo tasks | return | No clear effect |
| use DWC with DDPG instead of FP | DDPG policy · HalfCheetah-v4 | return: 11278.164 -> 7293.430 | Hurt |
Closest papers by meaning
- Differentiable Weightless Neural Networks (ICML 2024 · similarity 0.83)
- Deep learning for continuous-time stochastic control with jumps (NeurIPS 2025 · similarity 0.82)
- Provably Safe Neural Network Controllers via Differential Dynamic Logic (NeurIPS 2024 · similarity 0.82)
- Towards training digitally-tied analog blocks via hybrid gradient computation (NeurIPS 2024 · similarity 0.82)
- Mind the Gap: Removing the Discretization Gap in Differentiable Logic Gate Networks (NeurIPS 2025 · similarity 0.81)
- EvoControl: Multi-Frequency Bi-Level Control for High-Frequency Continuous Control (ICML 2025 · similarity 0.81)
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.
