Assignment games are a promising framework for autonomous management of a multimodal Tactical Sensing Grid (TSG). They provide theoretical guarantes and exhibit excellent empirical performance in maintaining custody of all observed targets in a scene. However, TSG sub-grid initiation of an assignment game requires multimodal data association between playing nodes (the sub-grid). Playing nodes must achieve consensus on target labels and identities before game play proceeds. Using the locally centralized communication network assumed by an assignment game, we propose the Triple-C distributed method for solving this association problem. The method we propose is suitable for fairly general scenarios, requiring that each node have an intrinsic notion of what constitutes an outlier and what constitutes similarity. At the core of the Triple-C method is a collection of parallelizable assignment games played between two nodes. Triple-C yields theoretical guarantees on output association consistency. We evaluate the performance of the Triple-C method on multiple simulations, showing that it computes associations quickly and accurately, thus enabling a TSG to maintain full situational awareness in a two-vehicle scenario.

Reinforcement learning is a promising candidate methodology for achieving situational awareness across an area of interest by controlling and processing data acquired from a multi-site, multi-modality sensing grid. We previously reported successful detection, tracking, classification, and identification of objects operating within the area of interest monitored by a sensing grid. We combined online kernel least squares policy iteration (an online reinforcement learning method combining dictionary learning with classical Q-learning) with a particle tracker to achieve these results. The work reported here extends these prior results to show that sensor fusion allows our online reinforcement learning methodology to successfully control a multi-modality sensor platform (consisting of a pan/tilt/zoom electro-optical camera, a radar, and passive radio frequency sensor) to maintain persistent surveillance of objects of interest. We evaluate our sensor fusion and online reinforcement learning methodology through Gazebo simulation of a realistic test and experimentation location. Our results demonstrate that policies based on our methodology trained on processed simulated sensor data perform as well as policies trained on known ground truth data. These results also show that the learned policies offer significant generalization ability, with the sensor platform being able to successfully track an observed object well past the observed training period. Moreover, we show that our reinforcement learning methodology shows promising results in dealing with the data association problem. These results further build towards our ultimate goal of achieving automated situational awareness across a heterogeneous sensor grid.

In this paper, we propose a new structural technique for pruning deep neural networks with skip-connections. Our approach is based on measuring the importance of feature maps in predicting the output of the model using their Fisher scores. These scores subsequently used for removing the less informative layers from the graph of the network. Extensive experiments on the classification of CIFAR-10, CIFAR-100, and SVHN data sets demonstrate the efficacy of our compressing method both in the number of parameters and operations.

The design of handcrafted neural networks requires a lot of time and resources. Recent techniques in Neural Architecture Search (NAS) have proven to be competitive or better than traditional handcrafted design, although they require domain knowledge and have generally used limited search spaces. In this paper, we propose a novel framework for neural architecture search, utilizing a dictionary of models of base tasks and the similarity between the target task and the atoms of the dictionary; hence, generating an adaptive search space based on the base models of the dictionary. By introducing a gradient-based search algorithm, we can evaluate and discover the best architecture in the search space without fully training the networks. The experimental results show the efficacy of our proposed task-aware approach.

We develop a framework for estimating unknown partial differential equations (PDEs) from noisy data, using a deep learning approach. Given noisy samples of a solution to an unknown PDE, our method interpolates the samples using a neural network, and extracts the PDE by equating derivatives of the neural network approximation. Our method applies to PDEs which are linear combinations of user-defined dictionary functions, and generalizes previous methods that only consider parabolic PDEs. We introduce a regularization scheme that prevents the function approximation from overfitting the data and forces it to be a solution of the underlying PDE. We validate the model on simulated data generated by the known PDEs and added Gaussian noise, and we study our method under different levels of noise. We also compare the error of our method with a Cramer-Rao lower bound for an ordinary differential equation (ODE). Our results indicate that our method outperforms other methods in estimating PDEs, especially in the low signal-to-noise (SNR) regime.

A technique for compressing deep neural models achieving competitive performance to state-of-the-art methods is proposed. The approach utilizes the mutual information between the feature maps and the output of the model in order to prune the redundant layers of the network. Extensive numerical experiments on both CIFAR-10, CIFAR-100, and Tiny ImageNet data sets demonstrate that the proposed method can be effective in compressing deep models, both in terms of the numbers of parameters and operations. For instance, by applying the proposed approach to DenseNet model with 0.77 million parameters and 293 million operations for classification of CIFAR-10 data set, a reduction of 62.66% and 41.00% in the number of parameters and the number of operations are respectively achieved, while increasing the test error only by less than 1%.