NeurIPS 2021: An Infinite-Feature Extension for Bayesian ReLU Nets ...
--------------------------------------------------------------------------------
An Infinite-Feature Extension for Bayesian ReLU Nets That Fixes Their Asymptotic Overconfidence
Agustinus Kristiadi, Matthias Hein, and Philipp Hennig
Advances in Neural Information Processing Systems (NeurIPS) 2021
--------------------------------------------------------------------------------
► Paper:
► Code:
A Bayesian treatment can mitigate overconfidence in ReLU nets around the training data. But far away from them, ReLU Bayesian neural networks (BNNs) can still underestimate uncertainty and thus be asymptotically overconfident. This issue arises since the output variance of a BNN with finitely many features is quadratic in the distance from the data region. Meanwhile, Bayesian linear models with ReLU features converge, in the infinite-width limit, to a particular Gaussian process (GP) with a variance that grows cubically s
6 views
23
7
2 years ago 01:38:22 1
Современные нейросетевые модели для задач медицинской химии
2 years ago 00:24:06 2
TMLR - A New Open Journal For Machine Learning
2 years ago 00:45:44 1
The New DBfication of ML/AI with Arun Kumar - #553
2 years ago 00:54:36 2
Теоретический подход к выбору метрик качества в задачах классификации и кластеризации
2 years ago 01:00:50 8
W&B Paper Reading Group: Revisiting Deep Learning Models for Tabular Data
3 years ago 00:25:17 15
The Case for Real-World Reinforcement Learning
3 years ago 00:25:40 37
[ML News] DeepMind tackles Math | Microsoft does more with less | Timnit Gebru launches DAIR
3 years ago 00:05:09 6
STEP: Segmenting and Tracking Every Pixel
3 years ago 01:00:49 6
NeurIPS 2021 Conference Preview with Zeta Alpha
3 years ago 00:08:12 1
SEAL: Self-supervised Embodied Active Learning (NeurIPS 2021)
3 years ago 00:11:02 6
NeurIPS 2021: An Infinite-Feature Extension for Bayesian ReLU Nets ...
3 years ago 00:28:29 18
Towards Animatable Human Avatars
3 years ago 00:07:03 2
NeurIPS 2021: Cockpit: A Practical Debugging Tool for Training Deep Neural Networks
3 years ago 00:14:14 20
[ML News] New ImageNet SOTA | Uber’s H3 hexagonal coordinate system | New text-image-pair dataset
3 years ago 00:17:07 6
[ML News] Blind Chess AI Competition | Graph NNs for traffic | AI gift suggestions
3 years ago 00:17:55 15
Глеб Моргачёв | Text2Bash: как мы участвовали в соревновании NeurIPS