Research map / Probabilistic and scientific modeling / Dynamical systems
Holomorphic Equilibrium Propagation Computes Exact Gradients Through Finite Size Oscillations
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 |
|---|---|---|---|
| replace classic EP with holomorphic EP (finite |β|=0.4) under noise | EP · MNIST val noisy | validation error (%): 3.01 -> 1.96 | Helped |
| use small nudge |β|=0.1 in classic EP under noise (ablation) | classic EP · MNIST val noisy | validation error (%): 1.87 -> 88.7 | Hurt |
| replace classic EP (|β|=0.4) with hEP (|β|=0.4) without noise | EP · MNIST val noise-free | validation error (%): 2.24 -> 1.97 | Helped |
| run hEP online with always-on oscillating teaching signal (ablation) | hEP · MNIST val noise-free | validation error (%): 1.97 -> 2.05 | No clear effect |
| use hEP instead of BP | VGG-like CNN · ImageNet 32x32 top-1 | accuracy (%): 37.2 -> 36.5 | No clear effect |
| use hEP instead of BP | VGG-like CNN · CIFAR-10 top-1 | accuracy (%): 88.3 -> 88.6 | No clear effect |
| use holomorphic architecture (softmax pooling, dSiLU) instead of max pooling and ReLU (ablation) | hEP VGG · CIFAR-10 training error | training error | Helped |
| increase number of nudging points N to 4 with finite beta (ablation) | hEP VGG · CIFAR-10 training error | training error | Helped |
| use too-fast teaching oscillations (ablation) | online hEP · MNIST gradient estimate | gradient bias | Hurt |
Closest papers by meaning
- Improving equilibrium propagation without weight symmetry through Jacobian homeostasis (ICLR 2024 · similarity 0.91)
- Training a Predictive Coding Network on ImageNet using Equilibrium Propagation (NeurIPS 2026 · similarity 0.89)
- Two Tales of Single-Phase Contrastive Hebbian Learning (ICML 2024 · similarity 0.88)
- Partial observation can induce mechanistic mismatches in data-constrained models of neural dynamics (NeurIPS 2024 · similarity 0.86)
- Unconditional stability of a recurrent neural circuit implementing divisive normalization (NeurIPS 2024 · similarity 0.86)
- Feedback control guides credit assignment in recurrent neural networks (NeurIPS 2024 · similarity 0.86)
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.