Bayesian neural networks (BNN) and deep ensembles are principled approaches to estimate the predictive uncertainty of a deep learning model. However their practicality in real-time, industrial-scale applications are limited due to their heavy memory and inference cost. This motivates us to study principled approaches to high-quality uncertainty estimation that require only a single deep neural network (DNN). By formalizing the uncertainty quantification as a minimax learning problem, we first identify input distance awareness, i.e., the model's ability to quantify the distance of a testing example from the training data in the input space, as a necessary condition for a DNN to achieve high-quality (i.e., minimax optimal) uncertainty estimation. We then propose Spectral-normalized Neural Gaussian Process (SNGP), a simple method that improves the distance-awareness ability of modern DNNs, by adding a weight normalization step during training and replacing the output layer. On a suite of vision and language understanding tasks and on modern architectures (Wide-ResNet and BERT), SNGP is competitive with deep ensembles in prediction, calibration and out-of-domain detection, and outperforms the other single-model approaches.

Metabolic programming and mitochondrial dynamics along with T cell differentiation affect T cell fate and memory development; however, how to control metabolic reprogramming and mitochondrial dynamics in T cell memory development is unclear. Here, we provide evidence that the SUMO protease SENP1 promotes T cell memory development via Sirt3 deSUMOylation. SENP1-Sirt3 signalling augments the deacetylase activity of Sirt3, promoting both OXPHOS and mitochondrial fusion. Mechanistically, SENP1 activates Sirt3 deacetylase activity in T cell mitochondria, leading to reduction of the acetylation of mitochondrial metalloprotease YME1L1. Consequently, deacetylation of YME1L1 suppresses its activity on OPA1 cleavage to facilitate mitochondrial fusion, which results in T cell survival and promotes T cell memory development. We also show that the glycolytic intermediate fructose-1,6-bisphosphate (FBP) as a negative regulator suppresses AMPK-mediated activation of the SENP1-Sirt3 axis and reduces memory development. Moreover, glucose limitation reduces FBP production and activates AMPK during T cell memory development. These data show that glucose limitation activates AMPK and the subsequent SENP1-Sirt3 signalling for T cell memory development.

No abstract