Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
TRADI: Tracking Deep Neural network Weight Distributions
Gianni Franchi, Andrei Bursuc, Emanuel Aldea, Séverine Dubuisson, Isabelle Bloch
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 TRADI weight-distribution tracking instead of MC Dropout | single network · Energy Efficiency regression | RMSE: 1.66 -> 1.20 | Improved |
| use TRADI instead of Deep Ensembles | single network · Concrete Strength regression | RMSE: 6.03 -> 5.20 | Improved |
| use TRADI instead of Deep Ensembles | single network · Naval Propulsion regression | NLL: -5.63 -> -2.83 | Hurt |
| use TRADI instead of MC Dropout | MLP · MNIST | accuracy (%): 98.19 -> 98.63 | Improved |
| use TRADI instead of MC Dropout | WideResNet 28x10 · CIFAR-10 | NLL: 0.205 -> 0.205 | No clear effect |
| use TRADI instead of Deep Ensembles (20 models) | MLP · MNIST/notMNIST OOD | AUC: 97.2 -> 96.7 | Hurt |
| use TRADI instead of Deep Ensembles | PSPNet · StreetHazards OOD | ECE: 0.051 -> 0.049 | Improved |
| use TRADI instead of Deep Ensembles | MLP · MNIST/notMNIST OOD | train time: 31m -> 2m | Improved |
| use TRADI instead of Deep Ensembles | PSPNet · StreetHazards OOD | train time: 132h19m -> 15h36m | Improved |
| sample weights with TRADI instead of Gaussian perturbation ensemble | PSPNet · StreetHazards OOD | AUC: 57.08 -> 89.2 | Improved |
| use TRADI instead of MC Dropout | ENet · CamVid-OOD | AUC: 75.4 -> 79.3 | Improved |
| use TRADI instead of Deep Ensembles | PSPNet · BDD Anomaly OOD | AUC: 87.0 -> 86.1 | Hurt |
Closest papers by meaning
- Train-by-Reconnect: Decoupling Locations of Weights from Their Values (NeurIPS 2020 · similarity 0.85)
- Learning to Weight Samples for Dynamic Early-Exiting Networks (ECCV 2022 · similarity 0.85)
- Estimating Training Data Influence by Tracing Gradient Descent (NeurIPS 2020 · similarity 0.84)
- Diffusion-based Neural Network Weights Generation (ICLR 2025 · similarity 0.83)
- A Symmetry-Aware Exploration of Bayesian Neural Network Posteriors (ICLR 2024 · similarity 0.83)
- Rotational Equilibrium: How Weight Decay Balances Learning Across Neural Networks (ICML 2024 · similarity 0.83)
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.
