This paper was accepted on the workshop Self-Supervised Studying – Concept and Apply at NeurIPS 2023.
*=Equal Contributors
Understanding mannequin uncertainty is vital for a lot of purposes. We suggest Bootstrap Your Personal Variance (BYOV), combining Bootstrap Your Personal Latent (BYOL), a negative-free Self-Supervised Studying (SSL) algorithm, with Bayes by Backprop (BBB), a Bayesian technique for estimating mannequin posteriors. We discover that the realized predictive std of BYOV vs. a supervised BBB mannequin is nicely captured by a Gaussian distribution, offering preliminary proof that the realized parameter posterior is helpful for label free uncertainty estimation. BYOV improves upon the deterministic BYOL baseline (+2.83% take a look at ECE, +1.03% take a look at Brier) and presents higher calibration and reliability when examined with varied augmentations (eg: +2.4% take a look at ECE, +1.2% take a look at Brier for Salt & Pepper noise).