Akinori Hidaka

Visual recognition is essential for animals, including humans, to interpret their environment. However, attention plays a key role in filtering visual information, directing brain resources to salient objects or locations. The saliency map model replicates this biological process, predicting where attention and gaze will focus. Recently, models based on deep convolutional neural networks (DCNNs) have leveraged large datasets to improve attentional predictions, though their mechanisms remain unclear. This study investigates attentional selection by analyzing how a DCNN-based saliency map model responds to adversarial-interference images-natural images altered with noise to disrupt the model’s predictions. While human perception remains unaffected by these modifications, the model’s neural responses are significantly altered. Especially, the interference images often shift attention from central to peripheral regions. These findings offer new insights into the workings of DCNN-based saliency models and deepen our understanding of human attention mechanisms.

Fisher's linear discriminant analysis (FLDA) is known as a method to find a discriminative feature space for multi-class classification. As a theory of extending FLDA to an ultimate nonlinear form, optimal nonlinear discriminant analysis (ONDA) has been proposed. ONDA indicates that the best theoretical nonlinear map for maximizing the Fisher's discriminant criterion is formulated by using the Bayesian a posterior probabilities. In addition, the theory proves that FLDA is equivalent to ONDA when the Bayesian a posterior probabilities are approximated by linear regression (LR). Due to some limitations of the linear model, there is room to modify FLDA by using stronger approximation/estimation methods. For the purpose of probability estimation, multi-nominal logistic regression (MLR) is more suitable than LR. Along this line, in this paper, we develop a nonlinear discriminant analysis (NDA) in which the posterior probabilities in ONDA are estimated by MLR. In addition, in this paper, we develop a way to introduce sparseness into discriminant analysis. By applying L1 or L2 regularization to LR or MLR, we can incorporate sparseness in FLDA and our NDA to increase generalization performance. The performance of these methods is evaluated by benchmark experiments using last_exam17 standard datasets and a face classification experiment.