Deep Neural Networks (DNNs) imitate the hierarchical information processing mechanism of human brains and have achieved great success in different applications. However, DNN is still a preliminary attempt of the brain-inspired computing and it ignores the spike-based encoding mechanism, the neuronal diversity, and the modular functional specification in the brain. Neuroscience studies show that biological neurons communicate with each other through spikes, and this enables ultra-low-power computation in the brain. Furthermore, the studies of human brain reveal that the principles of neuronal diversity and modular functional specialization are the keys to achieve human-level intelligence. It is a promising direction for the existing computational models towards artificial intelligence by integrating these brain-inspired information processing principles.
This research topic focuses on the spike-based computational models inspired by the functionally specific and cooperative features of the brain at both neuronal and structural levels. Though the underlying mechanisms of the spike-based communication mechanism, the neuronal diversity, and the modular functional specialization remain as an open question, the latest advances in spiking neural networks (SNNs) and experimental techniques in neuroscience make it possible for the current artificial intelligence models to learn from the above-mentioned principles. SNN follows the spike-based encoding, and the transmission mechanisms and is regarded as the third generation of artificial neural networks. It is biologically plausible and it can capture more types of information, such as timing, phase, and oscillation. Because of this, SNN is an effective mean to characterize neuronal diversity. In addition, with the advance of technologies in neuroscience, such as neuroimaging and electrophysiological techniques, researchers have made considerable progress in characterizing the functional specialization of different brain regions, as well as considerable progress in characterizing how these regions work together. The challenge is, however, how can these findings and recent progress inspire us to develop more efficient brain-inspired computational models. We seek to advance this goal by combining the recent progress in machine learning, neuroscience, and other related areas.
Relevant sub-topics include but are not limited to the following:
1. Encoding, Learning, theory, and applications for SNNs, covering spike-based encoding strategies for auditory, visual, olfactory, and other external stimuli; Spike-driven learning algorithms for deep SNNs and graph-based SNNs; Theoretical analysis for SNNs; and the Real-world applications for SNNs, such as pattern recognition and robotics.
2. Neuronal diversity theory and applications, including designing a variety of spiking neuron models, investigate their coordination mechanism, and develop related computational models for real-world applications.
3. Brain-inspired auditory, visual, olfactory, and natural language processing computational models.
4. Cognitive models inspired from functional specifically regions and the whole brain.
5. SNNs for multi-modal information processing and sensory fusion. Based on the coordination mechanism of different brain regions, to investigate the SNN-based multimodal computational framework for processing multi-modality data including language, speech, image, video, and heterogeneous signals.
6. Exploiting variability and device mismatch in neuromorphic circuits.
Deep Neural Networks (DNNs) imitate the hierarchical information processing mechanism of human brains and have achieved great success in different applications. However, DNN is still a preliminary attempt of the brain-inspired computing and it ignores the spike-based encoding mechanism, the neuronal diversity, and the modular functional specification in the brain. Neuroscience studies show that biological neurons communicate with each other through spikes, and this enables ultra-low-power computation in the brain. Furthermore, the studies of human brain reveal that the principles of neuronal diversity and modular functional specialization are the keys to achieve human-level intelligence. It is a promising direction for the existing computational models towards artificial intelligence by integrating these brain-inspired information processing principles.
This research topic focuses on the spike-based computational models inspired by the functionally specific and cooperative features of the brain at both neuronal and structural levels. Though the underlying mechanisms of the spike-based communication mechanism, the neuronal diversity, and the modular functional specialization remain as an open question, the latest advances in spiking neural networks (SNNs) and experimental techniques in neuroscience make it possible for the current artificial intelligence models to learn from the above-mentioned principles. SNN follows the spike-based encoding, and the transmission mechanisms and is regarded as the third generation of artificial neural networks. It is biologically plausible and it can capture more types of information, such as timing, phase, and oscillation. Because of this, SNN is an effective mean to characterize neuronal diversity. In addition, with the advance of technologies in neuroscience, such as neuroimaging and electrophysiological techniques, researchers have made considerable progress in characterizing the functional specialization of different brain regions, as well as considerable progress in characterizing how these regions work together. The challenge is, however, how can these findings and recent progress inspire us to develop more efficient brain-inspired computational models. We seek to advance this goal by combining the recent progress in machine learning, neuroscience, and other related areas.
Relevant sub-topics include but are not limited to the following:
1. Encoding, Learning, theory, and applications for SNNs, covering spike-based encoding strategies for auditory, visual, olfactory, and other external stimuli; Spike-driven learning algorithms for deep SNNs and graph-based SNNs; Theoretical analysis for SNNs; and the Real-world applications for SNNs, such as pattern recognition and robotics.
2. Neuronal diversity theory and applications, including designing a variety of spiking neuron models, investigate their coordination mechanism, and develop related computational models for real-world applications.
3. Brain-inspired auditory, visual, olfactory, and natural language processing computational models.
4. Cognitive models inspired from functional specifically regions and the whole brain.
5. SNNs for multi-modal information processing and sensory fusion. Based on the coordination mechanism of different brain regions, to investigate the SNN-based multimodal computational framework for processing multi-modality data including language, speech, image, video, and heterogeneous signals.
6. Exploiting variability and device mismatch in neuromorphic circuits.
