On May 9th, 2017, Prof. Wei Wang and Associate Prof. Ian Max Andolina (Institute of Neuroscience, Chinese Academy of Sciences) visited CASIA and brought two representations for the Brain Science and Intelligence Technology Lecture Series. Prof. Zhaoxiang Zhang from the Research Center for Brain-inspired Intelligence hosted the event.
Wei Wang gave a talk on How Does Our Visual Brain Integrate Local Visual Cues to Form Global Representations. In his talk, Wei Wang started from the fundamental research of brain cognitive function and stated that object recognition and motion perception are two main functions of visual system. There are two streams for them: the ventral stream is responsible for form and color perception while the dorsal stream for perceiving motion and location. For seeing an object, the visual system must involve an interactive process that integrates the flows of visual information across multiple brain regions. Since human and non-human primates can effortlessly see with super spatial resolution both global and local features of an object, Wang pointed that a core question in vision is how the cortex integrates local visual cues to form global representations along the visual hierarchy. Wang also introduced their works in neural representation and cortical functional organization for processing local, global features of visual stimuli with intermediate level complexity. In addition, visual illusions were also mentioned in the talk. Finally, Wang answered questions about information integration and how to integrate local visual cues to form global representations.
Ian Max Andolina brought us a lecture namely When the Top-Down Meets the Bottom-Up: Network Interactions and Information Processing in the Early Visual System. He introduced that in the primary visual cortex, less than 0.5% of the synaptic connectivity comes via the sensory receptors and thalamus. Such a data point contrasts with most biological and computational models of sensory processes (i.e., HMAX) where feedforward transmission of sensory input is dominant computational process. This paradox should make us seriously question what the recurrent intrinsic connectivity is actually doing and why the brain devotes so many resources to it. Then Andolina addressed how feedback may fit into a broader theory of hierarchical predictive coding. And he believed that a better understanding of neural feedback would help to model recurrent connectivity networks which is critically important in generating truly “intelligent” predictive artificial networks.
These two talks belong to lecture series held by CAS Center for Excellence in Brain Science and Intelligence Technology, which is a regular visit between CASIA and Institute of Neuroscience in Shanghai. It makes us to understand the visual information integration mechanisms in brain science domain, provides us some references and inspiration for brain-inspired intelligence from the working mechanisms how the brain works.
(By Guibo Zhu, Center for Brain-inspired Intelligence)