Research map / Architectures and efficiency / RNNs and normalization
Updates of Equilibrium Prop Match Gradients of Backprop Through Time in an RNN with Static Input
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 discrete-time EP instead of BPTT | MNIST fully connected networks, energy-based and prototypical settings · MNIST | test error (%): 2.06 ± 0.17 vs 2.11 ± 0.09; 2.01 ± 0.21 vs 2.02 ± 0.12; 2.00 ± 0.13 vs 2.00 ± 0.12; 1.95 ± 0.10 vs 2.09 ± 0.12; 2.01 ± 0 | No clear effect |
| Use the prototypical rather than energy-based setting (ablation) | EP fully connected networks · MNIST | test error (%): 2.06 ± 0.17 vs 2.00 ± 0.13 (1 hidden layer); 2.01 ± 0.21 vs 1.95 ± 0.10 (2 hidden layers) | No clear effect |
| Use a convolutional architecture rather than the fully connected variants (ablation) | EP · MNIST | test error (%): 1.02 ± 0.04 | Helped |
| Use EP rather than BPTT with the convolutional architecture | P-conv · MNIST | test error (%): 1.02 ± 0.04 vs 0.88 ± 0.06 | Hurt |
| Reduce the first- and second-phase iteration counts in discrete-time EP compared with original real-time EP | EP · MNIST | test accuracy: three to five | No clear effect |
Closest papers by meaning
- Holomorphic Equilibrium Propagation Computes Exact Gradients Through Finite Size Oscillations (NeurIPS 2022 · similarity 0.91)
- Training a Predictive Coding Network on ImageNet using Equilibrium Propagation (NeurIPS 2026 · similarity 0.89)
- Improving equilibrium propagation without weight symmetry through Jacobian homeostasis (ICLR 2024 · similarity 0.89)
- Towards training digitally-tied analog blocks via hybrid gradient computation (NeurIPS 2024 · similarity 0.85)
- Rotational Equilibrium: How Weight Decay Balances Learning Across Neural Networks (ICML 2024 · similarity 0.85)
- Unconditional stability of a recurrent neural circuit implementing divisive normalization (NeurIPS 2024 · similarity 0.85)
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.