Uncertainty estimates help to identify ambiguous, novel, or anomalous inputs, but the reliable quantification of uncertainty has proven to be challenging for modern deep networks. To improve uncertainty estimation, we propose On-Manifold Adversarial Data Augmentation or OMADA, which specifically attempts to generate the most challenging examples by following an on-manifold adversarial attack path in the latent space of an autoencoder-based generative model that closely approximates decision boundaries between two or more classes. On a variety of datasets and for multiple network architectures, OMADA consistently yields more accurate and better calibrated classifiers than baseline models, and outperforms competing approaches such as Mixup and CutMix, as well as achieving similar performance to (at times better than) post-processing calibration methods such as temperature scaling. Variants of OMADA can employ different sampling schemes for ambiguous on-manifold examples based on the entropy of their estimated soft labels, which exhibit specific strengths for generalization, calibration of predicted uncertainty, or detection of out-of-distribution inputs.

In 1983, the IBM PC XT debuted with 128K of RAM and a 10MB hard disk. In that same year, the first mobile phone debuted weighing about 2.5 pounds and with a $4,000 price tag. Fast forward to today and the average person unlocks their smartphone 76-80 times a day and relies on it for every aspect of their lives. These amazing pieces of hardware are millions of times more capable than all of NASA's computing power in the 1960s. Now that we have a supercomputer that never leaves people's sides, maybe it's time that we do some more innovation and see how that device can be used for "mobile health".