Research map / Probabilistic and scientific modeling / Variational and generative sampling
Variational Probability Flow for Biologically Plausible Training of Deep Neural Networks
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 |
|---|---|---|---|
| Train the RBM with VPF rather than PCD-10 | RBM(196) · MNIST corrupted digits | L1 reconstruction error: Top: 47.9 -> 32.9; Bottom: 48.9 -> 46.9; Left: 39.1 -> 36.7; Right: 43.7 -> 44.1 | Mixed |
| Add intra-layer connections (ablation) | VPF DBM with random top-layer initialization · MNIST test | Parzen estimated log-likelihood: 183±1.1 -> 222±0.5 | Helped |
| Add intra-layer connections (ablation) | VPF DBM with mean-activation top-layer prior · MNIST test | Parzen estimated log-likelihood: 210±0.8 -> 222±0.4 | Helped |
| Use a mean-activation prior instead of random initialization for the top hidden layer (ablation) | VPF DBM without intra-layer connections · MNIST test | Parzen estimated log-likelihood: 183±1.1 -> 210±0.8 | Helped |
| Use a mean-activation prior instead of random initialization for the top hidden layer (ablation) | VPF DBM with intra-layer connections · MNIST test | Parzen estimated log-likelihood: 222±0.5 -> 222±0.4 | No clear effect |
| Add intra-layer connections (ablation) | VPF one-hidden-layer model · MNIST corrupted digits | L1 reconstruction error: Top: 32.9 -> 34.33; Bottom: 46.9 -> 47.65; Left: 36.7 -> 36.41; Right: 44.1 -> 42.31 | Mixed |
| Increase RBM hidden-unit count from 100 to 196 or 400 (ablation) | VPF RBM on MNIST · MNIST | training convergence | No clear effect |
Closest papers by meaning
- Kernelized information bottleneck leads to biologically plausible 3-factor Hebbian learning in deep networks (NeurIPS 2020 · similarity 0.87)
- Brain-like Variational Inference (NeurIPS 2025 · similarity 0.86)
- Embedded-model flows: Combining the inductive biases of model-free deep learning and explicit probabilistic modeling (ICLR 2022 · similarity 0.86)
- Practical Deep Learning with Bayesian Principles (NeurIPS 2019 · similarity 0.86)
- Variational Bayesian Pseudo-Coreset (ICLR 2025 · similarity 0.85)
- Gradient Variance Reveals Failure Modes in Flow-Based Generative Models (NeurIPS 2025 · 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.