Springe direkt zu Inhalt

Zejin Lu

Zijin Li

Neural Dynamics of Visual Cognition

PhD Candidate

Address
Habelschwerter Allee 45
Room JK 24/221c
14195 Berlin

I joined the group as a PhD student in September 2021, and I am jointly supervised by Radoslaw Martin Cichy (Freie Universität Berlin) and Tim Kietzmann (Donders). Before that, I obtained my master degree in engineering from Hunan University (China) in 2021, and my master project specialized in computer vision in the context of vehicle engineering at Hunan University, China. And I completed my bachelor degree in engineering from Southwest Jiaotong University in China.

In my leisure time, I am fond of reading about history, exercising, playing board games and watching football games.

For more information, you can visit my personal website here.

General research interests

My research in the lab focuses on conditions of map formation in non-convolutional neural network models, and reinforcement learning in virtual environments. I am interested in understanding which factors shape brain representations, as well as the performance of bio-inspired deep learning models.

Research projects

  • Project 1: Our human vision may take positional regularities to recognize the objects: we better and more quickly recognize objects in their typical than in their atypical position. However, the current most popualr computation models of visino - convolutional neural networks (CNNs) - are not able to reproduce this effect due to the non-biological-plausible convolutional process. During the convolutional process, equivariant responses are generated from every spatial location in the view by weight sharing. This is biologically unrealistic. Instead, neurons in the visual system should be locally connected rather than convolutional. In my first project I compare the correspondence between locally-connected neural networks (a novel brain-inspired neural network present in my co-superviosr’ lab), CNN and human behavior.

Project 2: Deep reinforcement learning for interactive perception. In this project, I will built a robot in virtual reality to conduct classification taks by grasping and rotating various objects.

Student supervision & opportunities

I am not actively  looking for interns or thesis students. However, if you are interested in my research projects, please feel free to contact me.