Research map / Architectures and efficiency / RNNs and normalization
Dual Propagation: Accelerating Contrastive Hebbian Learning with Dyadic Neurons
Rasmus Høier, D. Staudt, Christopher Zach
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 |
|---|---|---|---|
| replace BP with DP | VGG16 · CIFAR10 | Top-1 accuracy: 92.26±0.23 -> 92.30±0.11 | No clear effect |
| replace BP with DP | VGG16 · CIFAR100 | Top-1 accuracy: 69.63±0.24 -> 69.57±0.51 | No clear effect |
| replace BP with DP | VGG16 · ImageNet32x32 | Top-1 accuracy: 41.28±0.19 -> 41.48±0.19 | No clear effect |
| use distinct feedback weights trained with Kolen-Pollack learning (ablation) | DP with VGG16 · CIFAR10 | Top-1 accuracy: 92.30±0.11 -> 91.84±0.11 | Hurt |
| use distinct feedback weights trained with Kolen-Pollack learning (ablation) | DP with VGG16 · CIFAR100 | Top-1 accuracy: 69.57±0.51 -> 70.40±0.25 | Improved |
| use distinct feedback weights trained with Kolen-Pollack learning (ablation) | DP with VGG16 · CIFAR100 | Top-5 accuracy: 88.36±0.13 -> 88.57 ± 0.15 | Improved |
| increase βL from 1.0 to 10 (ablation) | DP with MLP · MNIST | test accuracy (%): 98.38 -> 94.95 | Hurt |
| increase βL from 10 to 100 (ablation) | DP with MLP · MNIST | test accuracy (%): 94.95 -> 85.42 | Hurt |
| perform five inference passes and update weights after each pass (ablation) | DP with MLP · MNIST | test accuracy (%): 98.43 ±0.03 -> 98.40 ±0.02 | No clear effect |
| do not reset hidden and output units before each batch (ablation) | DP with MLP · MNIST | test accuracy (%): 98.43 ±0.03 -> 98.42 ±0.07 | No clear effect |
| update all neurons in parallel (ablation) | DP with MLP · MNIST | test accuracy (%): 98.43 ±0.03 -> 98.47 ±0.04 | No clear effect |
| randomly select layers for 100 updates (ablation) | DP with MLP · MNIST | test accuracy (%): 98.43 ±0.03 -> 98.48 ±0.11 | No clear effect |
Closest papers by meaning
- Two Tales of Single-Phase Contrastive Hebbian Learning (ICML 2024 · similarity 0.92)
- Improving equilibrium propagation without weight symmetry through Jacobian homeostasis (ICLR 2024 · similarity 0.86)
- Training a Predictive Coding Network on ImageNet using Equilibrium Propagation (NeurIPS 2026 · similarity 0.85)
- Rethinking Hebbian Principle: Low-Dimensional Structural Projection for Unsupervised Learning (NeurIPS 2025 · similarity 0.85)
- Neuron Platonic Intrinsic Representation From Dynamics Using Contrastive Learning (ICLR 2025 · similarity 0.85)
- Brain-like Variational Inference (NeurIPS 2025 · similarity 0.84)
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.
