Biblio

Explainable artificial intelligence (XAI) methods rely on access to model architecture and parameters that is not always feasible for most users, practitioners, and regulators. Inspired by cognitive psychology, we present a case for response times (RTs) as a technique for XAI. RTs are observable without access to the model. Moreover, dynamic inference models performing conditional computation generate variable RTs for visual learning tasks depending on hierarchical representations. We show that MSDNet, a conditional computation model with early-exit architecture, exhibits slower RT for images with more complex features in the ObjectNet test set, as well as the human phenomenon of scene grammar, where object recognition depends on intrascene object-object relationships. These results cast light on MSDNet's feature space without opening the black box and illustrate the promise of RT methods for XAI.

As an emerging paradigm for energy-efficiency design, approximate computing can reduce power consumption through simplification of logic circuits. Although calculation errors are caused by approximate computing, their impacts on the final results can be negligible in some error resilient applications, such as Convolutional Neural Networks (CNNs). Therefore, approximate computing has been applied to CNNs to reduce the high demand for computing resources and energy. Compared with the traditional method such as reducing data precision, this paper investigates the effect of approximate computing on the accuracy and power consumption of CNNs. To optimize the approximate computing technology applied to CNNs, we propose a method for quantifying the error resilience of each neuron by theoretical analysis and observe that error resilience varies widely across different neurons. On the basic of quantitative error resilience, dynamic adaptation of approximate bit-width and the corresponding configurable adder are proposed to fully exploit the error resilience of CNNs. Experimental results show that the proposed method further improves the performance of power consumption while maintaining high accuracy. By adopting the optimal approximate bit-width for each layer found by our proposed algorithm, dynamic adaptation of approximate bit-width reduces power consumption by more than 30% and causes less than 1% loss of the accuracy for LeNet-5.

Generating adversarial samples is gathering much attention as an intuitive approach to evaluate the robustness of learning models. Extensive recent works have demonstrated that numerous advanced image classifiers are defenseless to adversarial perturbations in the white-box setting. However, the white-box setting assumes attackers to have prior knowledge of model parameters, which are generally inaccessible in real world cases. In this paper, we concentrate on the hard-label black-box setting where attackers can only pose queries to probe the model parameters responsible for classifying different images. Therefore, the issue is converted into minimizing non-continuous function. A black-box approach is proposed to address both massive queries and the non-continuous step function problem by applying a combination of a linear fine-grained search, Fibonacci search, and a zeroth order optimization algorithm. However, the input dimension of a image is so high that the estimation of gradient is noisy. Hence, we adopt a zeroth-order optimization method in high dimensions. The approach converts calculation of gradient into a linear regression model and extracts dimensions that are more significant. Experimental results illustrate that our approach can relatively reduce the amount of queries and effectively accelerate convergence of the optimization method.

``Style transfer'' among images has recently emerged as a very active research topic, fuelled by the power of convolution neural networks (CNNs), and has become fast a very popular technology in social media. This paper investigates the analogous problem in the audio domain: How to transfer the style of a reference audio signal to a target audio content? We propose a flexible framework for the task, which uses a sound texture model to extract statistics characterizing the reference audio style, followed by an optimization-based audio texture synthesis to modify the target content. In contrast to mainstream optimization-based visual transfer method, the proposed process is initialized by the target content instead of random noise and the optimized loss is only about texture, not structure. These differences proved key for audio style transfer in our experiments. In order to extract features of interest, we investigate different architectures, whether pre-trained on other tasks, as done in image style transfer, or engineered based on the human auditory system. Experimental results on different types of audio signal confirm the potential of the proposed approach.