Search for: All records

While current deep learning algorithms have been successful for a wide variety of artificial intelligence (AI) tasks, including those involving structured image data, they present deep neurophysiological conceptual issues due to their reliance on the gradients that are computed by backpropagation of errors (backprop). Gradients are required to obtain synaptic weight adjustments but require knowledge of feed-forward activities in order to conduct backward propagation, a biologically implausible process. This is known as the “weight transport problem”. Therefore, in this work, we present a more biologically plausible approach towards solving the weight transport problem for image data. This approach, which we name the error-kernel driven activation alignment (EKDAA) algorithm, accomplishes through the introduction of locally derived error transmission kernels and error maps. Like standard deep learning networks, EKDAA performs the standard forward process via weights and activation functions; however, its backward error computation involves adaptive error kernels that propagate local error signals through the network. The efficacy of EKDAA is demonstrated by performing visual-recognition tasks on the Fashion MNIST, CIFAR-10 and SVHN benchmarks, along with demonstrating its ability to extract visual features from natural color images. Furthermore, in order to demonstrate its non-reliance on gradient computations, results are presented for an EKDAA-trained CNN that employs a non-differentiable activation function. 

We investigate the behaviors that compressed convolutional models exhibit for two key areas within AI trust: (i) the ability for a model to be explained and (ii) its ability to be robust to adversarial attacks. While compression is known to shrink model size and decrease inference time, other properties of compression are not as well studied. We employ several compression methods on benchmark datasets, including ImageNet, to study how compression affects the convolutional aspects of an image model. We investigate explainability by studying how well compressed convolutional models can extract visual features with t-SNE, as well as visualizing localization ability of our models with class activation maps. We show that even with significantly compressed models, vital explainability is preserved and even enhanced. We find with applying the Carlini & Wagner attack algorithm on our compressed models, robustness is maintained and some forms of compression make attack more difficult or time-consuming.