Research map / Architectures and efficiency / RNNs and normalization
Understanding Deep Architecture with Reasoning Layer
Xinshi Chen, Yufei Zhang, Christoph Reisinger, Le Song
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 a faster-converging reasoning algorithm | Hybrid perception-and-reasoning architecture · Synthetic experiments | approximation accuracy | Improved |
| Use NAG instead of GD | Hybrid architecture with quadratic energy · Synthetic experiments | approximation accuracy | Improved |
| Increase the reasoning-layer depth | Hybrid architecture with an over- or under-parameterized neural module · Synthetic experiments and BSD500 denoising | generalization gap | Hurt |
| Increase the reasoning-layer depth | Hybrid architecture with an about-right neural-module parameterization · Synthetic experiments and BSD500 denoising | generalization gap | Mixed |
| Use a more stable algorithm layer, such as GD rather than NAG | Hybrid architecture with an over- or under-parameterized neural module · BSD500 denoising | generalization gap | Improved |
| Use a faster-converging algorithm layer, such as NAG rather than GD | Hybrid architecture with an about-right neural-module parameterization · Generalization analysis | generalization gap | Mixed |
| Replace a traditional algorithm layer with an RNN reasoning layer | Hybrid architecture · Synthetic learning-to-optimize experiments | generalization ability | Hurt |
| Replace a traditional algorithm layer with an RNN reasoning layer | Hybrid architecture · Synthetic learning-to-optimize experiments | representation power | Improved |
Closest papers by meaning
- Interpretable Concept-Based Memory Reasoning (NeurIPS 2024 · similarity 0.83)
- The Unreasonable Ineffectiveness of the Deeper Layers (ICLR 2025 · similarity 0.83)
- Keep the Faith: Faithful Explanations in Convolutional Neural Networks for Case-Based Reasoning (AAAI 2024 · similarity 0.83)
- Mechanistic Design and Scaling of Hybrid Architectures (ICML 2024 · similarity 0.83)
- Layerwise Change of Knowledge in Neural Networks (ICML 2024 · similarity 0.83)
- Right for the Right Reasons: Avoiding Reasoning Shortcuts via Prototypical Neurosymbolic AI (NeurIPS 2025 · similarity 0.82)
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.
