报告题目：What is the Neural Mechanism of Few-shot Learning? A Simulation from Functions of Ventral Visual Stream

报告时间：2022年07月06日，星期三，上午10:00—11:00

腾讯会议 ID：103-393-346

报告人：金德泉副教授，广西大学数学与信息科学学院

报告简介：Few-shot learning is an intrinsic human cognitive capacity, but few brain-like learning models are intrinsically suitable for it, despite their great success in machine intelligence. To address this issue, we propose a new neural model to simulate the neural mechanism of few-shot learning from the function of the ventral visual stream in object recognition. The proposed model is biologically plausible, parameter-free, suitable for few-shot learning, and without a black-box issue. With some biologically inspired learning algorithms, the proposed model achieves good results and outperforms some state-of-the-art few-shot and zero-shot learning algorithms in the extensive experiments on real-world image datasets, which experimentally verifies the few-shot learning capacity of the proposed model and implies a novel and efficient learning framework for machine intelligence.

报告人简介：金德泉，广西大学数学与信息科学学院副教授，硕士生导师。在2005年、2008年和2011年于西安交通大学获得信息与计算科学学士学位，应用数学硕士学位和数学博士学位，主要从事认知学习、小样本学习和数据分析等相关领域的数学理论基础研究，在非线性系统稳定性分析、动力神经场方程动力学性质研究、基于信息熵的度量学习和介观尺度类脑模型及其学习算法等方面有较为深入的研究，发表学术论文10余篇，其中5篇发表在神经网络领域三大期刊Neural Computation, Neural Networks和IEEE Transactions on Neural Networks and Learning Systems上。主持完成国家自然科学基金项目3项，主持在研广西自然科学基金1项，完成1项。