Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Deep Neural Networks with Dependent Weights: Gaussian Process Mixture Limit, Heavy Tails, Sparsity and Compressibility
Hoil Lee, Fadhel Ayed, Paul Jung, Juho Lee, Hongseok Yang, Francois Caron
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 |
|---|---|---|---|
| use beta, horseshoe or generalised BFRY node variances instead of deterministic or inverse-gamma | infinite-width FFNN · output dimension dependence | correlation of squared outputs: width 2000: 0.001844 (deterministic), 0.004522 (inverse gamma) vs 0.297515 (beta), 0.253737 (horseshoe), 0.235673 (BFRY) | Improved |
| increase width from 100 to 2000 for deterministic variance (ablation) | FFNN with deterministic variance · output correlation | correlation: 0.019994 -> 0.001844 | Improved |
| increase width from 100 to 2000 for beta variance (ablation) | FFNN with beta variance · output correlation | correlation: 0.320444 -> 0.297515 | No clear effect |
| use heavy-tailed variance (beta, horseshoe, BFRY) instead of light-tailed | FFNN output distribution · width 2000, 50000 samples | output density tail | Improved |
Closest papers by meaning
- Beyond IID weights: sparse and low-rank deep Neural Networks are also Gaussian Processes (ICLR 2024 · similarity 0.91)
- Asymptotic Analysis of Two-Layer Neural Networks after One Gradient Step under Gaussian Mixtures Data with Structure (ICLR 2025 · similarity 0.89)
- An Infinite-Width Analysis on the Jacobian-Regularised Training of a Neural Network (ICML 2024 · similarity 0.89)
- Deep Kernel Posterior Learning under Infinite Variance Prior Weights (ICLR 2025 · similarity 0.88)
- Infinite-Width Limit of a Single Attention Layer: Analysis via Tensor Programs (NeurIPS 2025 · similarity 0.88)
- Quantitative convergence of trained neural networks to Gaussian processes (NeurIPS 2025 · similarity 0.88)
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.
