State-of-the-art techniques for enhancing robustness of deep networks mostly rely on empirical risk minimization with suitable data augmentation. In this paper, we propose a complementary approach motivated by communication theory, aimed at enhancing the signal-to-noise ratio at the output of a neural network layer via neural competition during learning and inference. In addition to standard empirical risk minimization, neurons compete to sparsely represent layer inputs by maximization of a tilted exponential (TEXP) objective function for the layer. TEXP learning can be interpreted as maximum likelihood estimation of matched filters under a Gaussian model for data noise. Inference in a TEXP layer is accomplished by replacing batch norm by a tilted softmax, which can be interpreted as computation of posterior probabilities for the competing signaling hypotheses represented by each neuron. After providing insights via simplified models, we show, by experimentation on standard image datasets, that TEXP learning and inference enhances robustness against noise and other common corruptions, without requiring data augmentation. Further cumulative gains in robustness against this array of distortions can be obtained by appropriately combining TEXP with data augmentation techniques. The code for all our experiments is available at \url{https://github.com/bhagyapuranik/texp_for_robustness}.

With AI-based decisions playing an increasingly consequential role in our society, for example, in our financial and criminal justice systems, there is a great deal of interest in designing algorithms conforming to application-specific notions of fairness. In this work, we ask a complementary question: can AI-based decisions be designed to dynamically influence the evolution of fairness in our society over the long term? To explore this question, we propose a framework for sequential decision-making aimed at dynamically influencing long-term societal fairness, illustrated via the problem of selecting applicants from a pool consisting of two groups, one of which is under-represented. We consider a dynamic model for the composition of the applicant pool, in which admission of more applicants from a group in a given selection round positively reinforces more candidates from the group to participate in future selection rounds. Under such a model, we show the efficacy of the proposed Fair-Greedy selection policy which systematically trades the sum of the scores of the selected applicants ("greedy'') against the deviation of the proportion of selected applicants belonging to a given group from a target proportion ("fair''). In addition to experimenting on synthetic data, we adapt static real-world datasets on law school candidates and credit lending to simulate the dynamics of the composition of the applicant pool. We prove that the applicant pool composition converges to a target proportion set by the decision-maker when score distributions across the groups are identical.

A key goal of next generation networks is to scale hardware design and signal processing algorithms to mmWave and THz arrays with a large number of elements. Imperfect manufacturing and limitations of circuit design introduce variations in the gain and relative phase offset of transmit and receive array elements that must be compensated prior to beam formation for either communication or sensing. We propose a novel method for calibrating large arrays in the field by exploiting the sparsity of the spatial channel. While conventional calibration methods are susceptible to multipath components in the wireless channel, our approach is shown to be robust to multipath interference if the measurements are gathered from a sufficiently diverse set of locations.

Machine learning models are vulnerable to adversarial attacks that can often cause misclassification by introducing small but well designed perturbations. In this paper, we explore, in the setting of classical composite hypothesis testing, a defense strategy based on the generalized likelihood ratio test (GLRT), which jointly estimates the class of interest and the adversarial perturbation. We evaluate the GLRT approach for the special case of binary hypothesis testing in white Gaussian noise under ℓ∞norm-bounded adversarial perturbations, a setting for which a minimax strategy optimizing for the worst-case attack is known. We show that the GLRT approach yields performance competitive with that of the minimax approach under the worst-case attack, while yielding a better robustness-accuracy trade-off under weaker attacks. The GLRT defense is applicable in multi-class settings and generalizes naturally to more complex models for which optimal minimax classifiers are not known.

In this paper, we show how the presence of locality within a binary cyclic code can be exploited to improve decoding performance and to reduce decoding complexity. We pursue two approaches. Under the first approach, we show how the ordered statistics decoding (OSD) method can be modified by inserting a simple single round belief-propagation step at the start that involves only the local codes. The resultant locality-aware OSD algorithm yields an appreciable signal-to-noise ratio (SNR) gain for a given level of reliability and essentially the same level of decoder complexity. Under the second, trellis decoding approach, we show that the careful introduction of locality results in the creation of a cyclic subcode that possesses lower maximum state complexity. In addition, we present a simple means of deriving an upper bound to the state complexity profile of any cyclic code that is based only on the zeros of the code. Furthermore, we show how the decoding speed of either locality-aware OSD or trellis decoding can be significantly increased in the presence of locality, in the moderate-to-high SNR regime, by making the use of a quick-look decoder that often returns the maximum likelihood code word.

In this paper, we study the impact of locality on the decoding of binary cyclic codes under two approaches, namely ordered statistics decoding (OSD) and trellis decoding. Given a binary cyclic code having locality or availability, we suitably modify the OSD to obtain gains in terms of Signal-To-Noise ratio, for a given reliability and essentially the same level of decoder complexity. With regard to trellis decoding, we show that careful introduction of locality results in the creation of cyclic subcodes having lower maximum state complexity. We also present a simple upper-bounding technique on the state complexity profile, based on the zeros of the code. Finally, it is shown how the decoding speed can significantly be increased in the presence of locality, in the moderate-to-high SNR regime, by making use of a quick-look decoder that often returns the ML codeword.