A Field Guide to Not Trusting Your Neural Network

What it does

This is a curated “awesome” list that maps the sprawling research landscape of uncertainty quantification in deep learning. It catalogs hundreds of papers, surveys, datasets, and open-source libraries, sorting them by methodological family—Bayesian neural networks, deep ensembles, Monte Carlo dropout, evidential learning, and conformal prediction among others—and by downstream task like segmentation or NLP. Think of it as a living literature review that accepts pull requests instead of peer reviews.

The interesting bit

The taxonomy is the real product. Rather than dumping links, the maintainers slice the field by technique and by application domain, so you can jump from “I need uncertainty for semantic segmentation” straight to the relevant papers and their PyTorch or JAX implementations. It also tracks newer ground like multimodal and generative AI models.

Key highlights

• Covers the full methodological stack: Bayesian methods, ensembles, dropout-based sampling, post-hoc calibration, deterministic uncertainty, and conformal predictions.
• Organized by application area, including classification, regression, object detection, domain adaptation, active learning, and NLP.
• Maintains dedicated sections for datasets, benchmarks, and software libraries across PyTorch, TensorFlow, and JAX.
• Includes non-paper resources: lectures, tutorials, and books for getting up to speed.
• Community-maintained via GitHub discussions and pull requests.

Verdict

Worth bookmarking if you are building safety-critical systems, debugging model confidence, or just trying to navigate the field without reading every NeurIPS proceedings. Skip it if you are looking for a single unified framework to drop into your codebase—this is a map, not a library.