Cognitive Neurodynamics - Recent Highlights in Cognitive Neurodynamics

Combining Computational Neuroscience and Deep Learning: the neural mechanism of visual information degradation

A hard-to-explain scientific question in visual neuroscience is, why does visual information transmission project higher into the visual cortex and the amount of information drops sharply? But it doesn't affect our visual cognition of the outside world. To solve this problem, the authors of this paper based on the anatomical structure of the visual nervous system and fused the theory of deep learning, established a visual nervous system network model from the retina to the VI region of vision cortex. Using the information provided by experimental data, visual perception from the external world was accurately reproduced by visual cortex coding. The results of this study initially answer the long-standing doubts in visual neuroscience about why the visual cortex of the brain can interpret and encode external visual perception from scarce data information and can predict the need for environmental change.

Neural mechanism of visual information degradation from retina to V1 area. Cognitive Neurodynamics. (2020) Read article (this opens in a new tab)

Deep Neural Networks: A neuro-inspired general framework for the evolution of stochastic dynamical systems

In this paper, Pontes-Filho et al. describe a general framework for evolving and simulating varied types of dynamical systems, such as cellular automata and sparsely-connected recurrent neural networks. In this framework, evolutionary algorithms can improve dynamical systems’ computational capabilities, such as critical behavior; and can guide the dynamical systems towards modeling physical substrates, such as biological neural networks and nanomagnetic ensembles. Our general framework, named EvoDynamic, is an open-source Python library, which is meant to allow for wide usage by the research community. Since generalization affects performance, EvoDynamic is optimized by being based on TensorFlow deep neural network library. To demonstrate the potential of our framework, we evolved three stochastic dynamical systems towards criticality. These systems are based on cellular automata, random Boolean networks, and echo state networks. The obtained results from their evolution are promising and demonstrate that criticality is achieved. In this work, the main contributions are:

Optimized open-source Python library capable of simulating and evolving several types of dynamical systems;Dynamical systems can be evolved towards desired behavior that can increase computational capacity or mimic physical substrates;EvoDynamic framework is useful for a wide range of research fields, such as artificial intelligence, artificial life, and complex systems.

Pontes-Filho, S., Lind, P., Yazidi, A. et al. A neuro-inspired general framework for the evolution of stochastic dynamical systems: Cellular automata, random Boolean networks and echo state networks towards criticality. Cogn Neurodyn 14, 657–674 (2020). Read article (this opens in a new tab)

Experimental Neuroscience: Strategy research in Behavioral Decision-making

To exploit a familiar situation (the exploitation strategy) or to explore a new environment (the exploration strategy) is important for animal to survive in a changing environment. Critical questions in the research field of decision-making are that how animals select one of the two strategies for behavioral choices and which factors control dynamical transitions between the two strategies. To study these issues, most of previous experiments focus on human participants by measuring the brain signal with fMRI. Due to their technical limitations, it is difficult to demonstrate neuronal mechanisms of strategy selection. This study designs a behavioral paradigm for rats to explore which strategy is used to select a reward option in a changeable environment. It is found that rats that learn a single action-outcome association prefer to use the exploitation strategy than the exploration strategy. On the other hand, rats that learn multiple action-outcome associations demonstrate the preference of using the exploration strategy. The data suggests that the strategy of choice behavior is modulated by the information complexity of the environment. This behavioral study for rats may provide a task paradigm to investigate neuronal mechanisms of non-human animals how to select the exploitation or exploration strategy dependent on the environment.

Noha Mohsen Zommara, Muneyoshi Takahashi & Johan Lauwereyns. Influence of multiple action–outcome associations on the transition dynamics toward an optimal choice in rats. Cognitive Neurodynamics, Volume 12 (No.1), pages43–53 (2018). Read article (this opens in a new tab)