| Time | 12:00-13:00,Sep 5, 2023 |
| Venue | E203, Biomedicine Hall, Tsinghua University |
| Host | |
| Speaker | Yueqiao Li (李岳桥) |
| Topic | Finding Engram in DNN and Design Engram Network |
| Abstract | Inspired by biological neural connections, artificial neural networks have been proposed and have achieved significant success in recent years. However, artificial neural networks suffer from poor interpretability, demanding substantial training data that needs to be balanced distributed, and they also grapple with the problem of catastrophic forgetting. In contrast, biological memory boasts distinct advantages in these aspects. In this study, we introduce the concept of biological engrams to common artificial neural networks and further analyze them from the perspective of memory encoding. The results offer a biologically grounded understanding of the hierarchical distribution, encoded content, and classification criteria of artificial neural networks. Guided by the engram map, we enhance the classification accuracy of artificial neural networks. Finally, we design a novel neural network feedforward structure to achieve one-class continual learning, which closely emulates biological learning in the real world. |
请扫二维码填写报名问卷（报告环节提供午餐）
