The generalization error of deep neural networks via their classification margin is studied in this work. Our approach is based on the Jacobian matrix of a deep neural network and can be applied to networks with arbitrary non-linearities and pooling layers, and to networks with different architectures such as feed forward networks and residual networks. Our analysis leads to the conclusion that a bounded spectral norm of the network's Jacobian matrix in the neighbourhood of the training samples is crucial for a deep neural network of arbitrary depth and width to generalize well. This is a significant improvement over the current bounds in the literature, which imply that the generalization error grows with either the width or the depth of the network. Moreover, it shows that the recently proposed batch normalization and weight normalization re-parametrizations enjoy good generalization properties, and leads to a novel network regularizer based on the network's Jacobian matrix. The analysis is supported with experimental results on the MNIST, CIFAR-10, LaRED and ImageNet datasets.Comment: accepted to IEEE Transactions on Signal Processin

No abstract

This paper considers the classification of linear subspaces with mismatched classifiers. In particular, we assume a model where one observes signals in the presence of isotropic Gaussian noise and the distribution of the signals conditioned on a given class is Gaussian with a zero mean and a low-rank covariance matrix. We also assume that the classifier knows only a mismatched version of the parameters of input distribution in lieu of the true parameters. By constructing an asymptotic low-noise expansion of an upper bound to the error probability of such a mismatched classifier, we provide sufficient conditions for reliable classification in the low-noise regime that are able to sharply predict the absence of a classification error floor. Such conditions are a function of the geometry of the true signal distribution, the geometry of the mismatched signal distributions as well as the interplay between such geometries, namely, the principal angles and the overlap between the true and the mismatched signal subspaces. Numerical results demonstrate that our conditions for reliable classification can sharply predict the behavior of a mismatched classifier both with synthetic data and in a motion segmentation and a hand-written digit classification applications.

This paper presents an adaptable dictionary-based feature extraction approach for spike sorting offering high accuracy and low computational complexity for implantable applications. It extracts and learns identifiable features from evolving subspaces through matched unsupervised subspace filtering. To provide compatibility with the strict constraints in implantable devices such as the chip area and power budget, the dictionary contains arrays of {− , } and the algorithm need only process addition and subtraction operations. Three types of such dictionary were considered. To quantify and compare the performance of the resulting three feature extractors with existing systems, a neural signal simulator based on several different libraries was developed. For noise levels between 0.05 and 0.3 and groups of 3 to 6 clusters, all three feature extractors provide robust high performance with average classification errors of less than 8% over five iterations, each consisting of 100 generated data segments. To our knowledge, the proposed adaptive feature extractors are the first able to classify reliably 6 clusters for implantable applications. An ASIC implementation of the best performing dictionary-based feature extractor was synthesized in a 65-nm CMOS process. It occupies an area of 0.09 mm 2 and dissipates up to about 10.48 µW from a 1 V supply voltage, when operating with 8-bit resolution at 30 kHz operating frequency.

Inverse problems abound in a number of domains such as medical imaging, remote sensing, and many more, relying on the use of advanced signal & image processing approaches -such as sparsity-driven techniques -to determine their solution. This paper instead studies the use of deep learning approaches to approximate the solution of inverse problems. In particular, the paper provides a new generalization bound, depending on key quantity associated with a deep neural networkits Jacobian matrix -that also leads to a number of computationally efficient regularization strategies applicable to inverse problems The paper also tests the proposed regularization strategies in a number of inverse problems including image super-resolution ones. Our numerical results conducted on various datasets show that both fully connected and convolutional neural networks regularized using the regularization or proxy regularization strategies originating from our theory exhibit much better performance than deep networks regularized with standard approaches such as weight-decay.