Analytically Tractable Heteroscedastic Uncertainty Quantification in Bayesian Neural Networks

Presenter: Bhargob Deka | Co-authors: Nguyen, L.H. and Goulet, J.-A. Paper title Analytically Tractable Heteroscedastic Uncertainty Quantification in Bayesian Neural Networks for Regression Tasks Journal: Neurocomputing (2024) | GitHub: https://github.com/lhnguyen102/cuTAGI... | PDF: http://profs.polymtl.ca/jagoulet/Site... AGVI Video:    • Approximate Gaussian Variance Inference fo...   Abstract: The tractable approximate Gaussian inference (TAGI) method allows for analytical parameter inference in Bayesian neural networks. In its current form, TAGI can only model homoscedastic aleatory uncertainty that is quantified by a constant error variance across the input covariate- domain. In this paper, we present the approximate Gaussian variance inference (AGVI) method that enables analytical inference of the error variance term as a Gaussian random variable. The combined framework regrouping TAGI and AGVI, referred to as TAGI-V, enables modeling heteroscedastic aleatory uncertainty in Bayesian neural networks. TAGI-V outperforms the homoscedastic version of TAGI in terms of predictive performance for the benchmark regression datasets. In comparison with other approximate inference methods, TAGI-V is an order of magnitude faster and exhibits a comparable or superior predictive performance. 00:00 Context 07:18 TAGI 10:21 TAGI-V 12:31 Toy problem 15:01 Benchmark 18:15 Conclusion