Research map / Architectures and efficiency / RNNs and normalization
ePC: Fast and Deep Predictive Coding in Digital Simulation
Cédric Goemaere, Gaspard Oliviers, Rafal Bogacz, Thomas Demeester
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 |
|---|---|---|---|
| reparameterize PC over prediction errors (ePC) instead of states (sPC) | 20-layer linear PC network · MNIST convergence to analytical optimum | convergence speed: over 100x faster | Improved |
| use ePC instead of sPC on deep non-linear MLPs | PC network · Deep non-linear MLPs | steps to convergence: sPC needed >100,000 steps | Improved |
| use ePC instead of sPC on VGG-7 with MSE loss | VGG-7 · CIFAR-10 | accuracy (%): 77.79±0.34 -> 88.98±0.19 | Improved |
| use ePC instead of sPC on ResNet-18 with MSE loss | ResNet-18 · CIFAR-10 | accuracy (%): 53.74±0.43 -> 92.17±0.26 | Improved |
| use ePC instead of backprop on VGG-9 with MSE loss | VGG-9 · CIFAR-100 (Top-1) | accuracy (%): 66.28±0.29 -> 61.35±0.76 | Hurt |
| use ePC instead of backprop on small MLP | 4-layer MLP · FashionMNIST, MSE loss | accuracy (%): 88.79±0.21 -> 87.02±0.24 | Hurt |
| switch output loss from MSE to cross-entropy for ePC | ePC · VGG-7 CIFAR-100 (Top-1) | accuracy (%): 66.55±0.45 -> 58.62±0.20 | Hurt |
| higher numerical precision (float64) for sPC (ablation) | sPC · 20-layer MLP signal propagation | signal propagation | No clear effect |
| feedforward state initialization for sPC | sPC · Signal propagation | convergence | Mixed |
Closest papers by meaning
- Towards training digitally-tied analog blocks via hybrid gradient computation (NeurIPS 2024 · similarity 0.87)
- $\mu$PC: Scaling Predictive Coding to 100+ Layer Networks (NeurIPS 2025 · similarity 0.87)
- Augmented Lagrangian Predictive Coding (NeurIPS 2026 · similarity 0.87)
- Only Strict Saddles in the Energy Landscape of Predictive Coding Networks? (NeurIPS 2024 · similarity 0.86)
- Understanding and Improving Optimization in Predictive Coding Networks (AAAI 2024 · similarity 0.86)
- Understanding Sample Efficiency in Predictive Coding (NeurIPS 2026 · 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.
