Sorry, you need to enable JavaScript to visit this website.

In this paper, we propose a statistical framework to prune feature maps in 1-D deep convolutional networks. SoundNet is a pre-trained deep convolutional network that accepts raw audio samples as input. The feature maps generated at various layers of SoundNet have redundancy, which can be identified by statistical analysis. These redundant feature maps can be pruned from the network with a very minor reduction in the capability of the network.

This paper studies the problem of Stackelberg game based distributed power allocation for spectral coexisting multistatic radar and communication systems. The strategy aims to minimize the radiated power of each radar by optimizing transmit power allocation for a desired signal-to-interference-plus-noise ratio (SINR) meanwhile the communication base station (CBS) is protected from the interference of radar transmissions. We formulate this distributed power allocation process as a Stackelberg game, where the CBS is a leader and the radars are the followers.

Direction of arrival (DOA) information of a signal is important in communications, localization, object tracking and so on. Frequency-domain-based time-delay estimation is capable of achieving DOA in subsample accuracy; however, it suffers from the phase wrapping problem. In this paper, a frequency-diversity based method is proposed to overcome the phase wrapping problem. Inspired by the machine learning technique of random ferns, an algorithm is proposed to speed up the search procedure.

Power outages have a major impact on economic development due to the dependence of (virtually all) productive sectors on electric power. Thus, many resources within the scientific and engineering communities have been employed to improve the efficiency and reliability of power grids. In particular, we consider the problem of predicting power outages based on the current weather conditions. Weather measurements taken by a sensor network naturally fit within the graph signal processing framework since the measurements are related by the relative position of the sensors.

In this paper, we propose a robust analog-only beamforming scheme for the downlink multi-user systems, which not only suppresses the interference and enhances the beamform- ing gain, but also provides robustness against imperfect channel state information (CSI). We strike a balance between the average beamforming gain and the inter-user interference by formulating a multi-objective problem. A probabilistic objective of leakage interference power is formulated to alleviate the effects of the channel estimation and feedback quantization errors.

Deep neural networks (DNNs) have been shown to be powerful models and perform extremely well on many complicated artificial intelligent tasks. However, recent research found that these powerful models are vulnerable to adversarial attacks, i.e., intentionally added imperceptible perturbations to DNN inputs can easily mislead the DNNs with extremely high confidence. In this work, we enhance the robustness of DNNs under adversarial attacks by using pruning method and logits augmentation, we achieve both effective defense against adversarial examples and DNN model compression.

This paper presents a large-scale fading channel model at the 60 GHz band. This model is based on the measurement campaign that our team conducted at Boise Airport and Boise State University. The close-in reference path loss and floating-intercept path loss models with both statistical and stochastic approaches are investigated for these environments. The measurements were collected at several different locations in line-of-sight (LOS) and non-line-of-sight (NLOS) scenarios using a high gain directional antenna.

In this paper, we consider two practical coded compressive imaging techniques. We investigate the optimal number of measurements under quadratic signal-to-noise-ratio (SNR) decrease. We focus on imaging scenarios in both real and complex vector spaces. In real vector spaces, we consider focal plane array (FPA) based super-resolution imaging with a constant measurement time constraint. Our model is comprised of a spatial light modulator and a low resolution FPA for modulating and sampling the incoming light intensity, respectively.