Correspondence between Monkey Visual Cortices and Layers of a Saliency Map Model Based on a Deep Convolutional Neural Network for Representations of Natural Images

Abstract

Correspondence between monkey visual cortices and layers of a saliency map model based on a deep convolutional neural network for representations of natural images 2. Abbreviated Title: Mechanism of DCNN and visual system for attention 3.

Predictive coding of natural images by V1 firing rates and rhythmic synchronization

“…We used SalGAN, a state-of-the art generative adversarial network trained for visual saliency prediction of the 224 224 RF image patch ( Pan et al., 2017 ; Wagatsuma et al., 2021 ) (see Figure S4 I). We quantified salience as the average visual saliency prediction value within the RF center.…”

Analysis based on neural representation of natural object surfaces to elucidate the mechanisms of a trained AlexNet model

“…Physiological experiments for recording activities of monkey V1, V4, and inferior temporal in response to surfaces of natural objects Tamura et al (2016) recorded neuronal responses in three visual cortical areas (V1, V4, and IT) of Macaca fuscata to images of natural object surfaces (Figure 2). The details of surgery, neural recording, and experimental procedures have been reported in our previous studies (Tamura et al, 2016;Wagatsuma et al, 2020). Spiking responses of single V1, V4, and IT neurons were recorded from four analgesized monkeys (Tamura et al, 2016;Wagatsuma et al, 2020).…”

“…The details of surgery, neural recording, and experimental procedures have been reported in our previous studies (Tamura et al, 2016;Wagatsuma et al, 2020). Spiking responses of single V1, V4, and IT neurons were recorded from four analgesized monkeys (Tamura et al, 2016;Wagatsuma et al, 2020). The effect of analgesia was likely immaterial given that the stimulus selectivity of V1 and IT neurons recorded from analgesized/paralyzed monkeys has been shown to be similar to that of awake-behaving monkeys (Wurtz, 1969;Tamura and Tanaka, 2001).…”

“…Deep neural network approaches potentially enable an even deeper understanding of the neural mechanisms involved in perceptual processing (Cadieu et al, 2014;Güçlü and van Gerven, 2015;Yamins and DiCarlo, 2016;Rajalingham et al, 2018;Wagatsuma et al, 2020;Dobs et al, 2022). In addition to achieving comparable object classification performance to that of human beings, AlexNet (Krizhevsky et al, 2012) promotes mutual understanding of neuroscientific and artificial approaches for explaining the information processing involved in visual object recognition (Yamins and DiCarlo, 2016;Geirhos et al, 2018).…”

Characteristics of Visual Saliency Caused by Character Feature for Reconstruction of Saliency Map Model

“…In recent years, deep learning based models have been proposed. These models express the characteristics of bottom-up attention [ 6 , 7 ], hybrid bottom-up and top-down attention [ 8 ], eye fixation [ 9 ], and sequential eye movements [ 10 , 11 ]. However, instead of an image feature, a meaning map that utilizes meaning information in a scene has been proposed [ 12 ].…”