Tielin Zhang

(In English):

I am Tielin Zhang (Thomas Zhang) from the Institute of Automation, Chinese Academy of Sciences (CASIA). I am now an Associate Professor in CASIA, and my research interests focus on the following parts:

(1) The biology-inspired spiking neural networks (SNNs): Following after the conceptual machines, shallow or deep neural networks, SNNs are considered as the third generation of neural networks. The hints and inspirations from biological systems will at least answer these questions: 1) what are the intrinsic reasons that biological networks could form cognitive functions from scratch; 2) what are the basic mathematically learning principles in this procedure; 3) how to construct powerful brain-inspired computational algorithms and hardware chips based on these rules. Until now, for the pattern recognition task, we have built SNNs based on pure biologically plausible plasticity rules, and have reached 98.64% accuracy on the standard MNIST dataset (AAAI2018, IJCAI2018). Next, we will expand them to other temporal tasks and construct task-specific neuron chips.

(2) The micron-scale biological structure reconstruction in the whole rat brain: With the provided 16,216 biological rat-brain slices from cooperators (Luo, HUST), I made a deep neural network model for the biological tissue recognition with high accuracy (94%). Then the data are loaded into Houdini (a 3D image processing software) for noise filtering and image reconstruction with the cooperation of my colleague Xinhe Zhang. The whole rat-brain atlas is multi-scales: on the microscale, detailed structures of somas, dendrites, axons are clear (but it doesn't contain the direction of synapses for the limitation of Golgi-type imaging); on the mesoscale and macroscale, the neuron group connections and their motif connection types are clear in and between different brain regions. This will be an important basis for the next-step research on brain connectome reconstruction, simulation, and functional analysis. We further built an integrative deep neural network (CNN, RNN and Tree architecture) for the 12-class neuron type classification with morphology, including ganglion, granule, medium spiny, parachromaffin, Purkinje, and pyramidal cells. The accuracy on these neurons reached 91%.

(3) Brain-inspired software or hardware strengthened cognitive robots: Nowadays, the robotic developments of hardware are much faster than that of software. With the help of the robot operating system (ROS), we built the basic information processing layer for the management of hardware. At the same time, the structural efforts (e.g., rat brain atlas) and functional efforts are combined together for a more powerful biologically plausible robot brain. This cognitive architecture will handle tens of cognitive functions for the purpose of artificial general intelligence (AGI), including the sensation with visual pathway inspired SNN, the memory with hippocampus inspired recurrent SNN, and the decision making with basal ganglia inspired reward SNN.

(4) Community contributions: I have been working as reviewers of different journals and conferences, for example, International Journal of Computer Vision, Neural Networks, Journal of Cognitive Computation, Journal of Cognitive Systems Research, SCIENCE CHINA Information Sciences, Neurocomputing, Frontiers in Neurorobotics, International Conference on Brain Informatics, International Conference on Agents, and so on.


Mail: tielin.zhang@ia.ac.cn

Institute: Institute of Automaiton, Chinese Academy of Sciences

Department: Research Center for Brain-inspired Intelligence

Address: Zhongguancun East Road 95, Haidian District, Beijing, China, 100190


(In Chinese):



【实习生招聘(Calling for interns)】有意者请联系tielin.zhang@ia.ac.cn.

Copyright Research Center for Brain-inspired Intelligence