Search for: All records

Despite increased interest in wearables as tools for detecting various health conditions, there are not as of yet any large public benchmarks for such mobile sensing data. The few datasets that are available do not contain data from more than dozens of individuals, do not contain high-resolution raw data or do not include dataloaders for easy integration into machine learning pipelines. Here, we present Homekit2020: the first large-scale public benchmark for time series classification of wearable sensor data. Our dataset contains over 14 million hours of minute-level multimodal Fitbit data, symptom reports, and ground-truth laboratory PCR influenza test results, along with an evaluation framework that mimics realistic model deployments and efficiently characterizes statistical uncertainty in model selection in the presence of extreme class imbalance. Furthermore, we implement and evaluate nine neural and non-neural time series classification models on our benchmark across 450 total training runs in order to establish state of the art performance. 

We demonstrate, theoretically and empirically, that adversarial robustness can significantly benefit from semisupervised learning. Theoretically, we revisit the simple Gaussian model of Schmidt et al. that shows a sample complexity gap between standard and robust classification. We prove that unlabeled data bridges this gap: a simple semisupervised learning procedure (self-training) achieves high robust accuracy using the same number of labels required for achieving high standard accuracy. Empirically, we augment CIFAR-10 with 500K unlabeled images sourced from 80 Million Tiny Images and use robust self-training to outperform state-of-the-art robust accuracies by over 5 points in (i) ℓ∞ robustness against several strong attacks via adversarial training and (ii) certified ℓ2 and ℓ∞ robustness via randomized smoothing. On SVHN, adding the dataset's own extra training set with the labels removed provides gains of 4 to 10 points, within 1 point of the gain from using the extra labels. 

While Generative Adversarial Networks (GANs) have demonstrated promising performance on multiple vision tasks, their learning dynamics are not yet well understood, both in theory and in practice. To address this issue, we study GAN dynamics in a simple yet rich parametric model that exhibits several of the common problematic convergence behaviors such as vanishing gradients, mode collapse, and diverging or oscillatory behavior. In spite of the non-convex nature of our model, we are able to perform a rigorous theoretical analysis of its convergence behavior. Our analysis reveals an interesting dichotomy: a GAN with an optimal discriminator provably converges, while first order approximations of the discriminator steps lead to unstable GAN dynamics and mode collapse. Our result suggests that using first order discriminator steps (the de-facto standard in most existing GAN setups) might be one of the factors that makes GAN training challenging in practice.