Featured Discovery

Home > Featured Discovery > Breakthrough: EIEE Research Team Published Article on Sub-journal of Nature

Breakthrough: EIEE Research Team Published Article on Sub-journal of Nature

June 17, 2020      Author: Du Xiaodong

Recently, a research team from the School of Electronic Information and Electrical Engineering, SJTU, published a research paper about high-dimensional visual sequences titled "Complex sequential understanding through the awareness of spatial and temporal concepts" on Nature Machine Intelligence, a sub-journal of Nature. Shanghai Jiao Tong University is the only affiliation of this paper and it is also SJTU's first paper that got published on this journal.

By imitating human cognitive mechanism, this paper proposes a semi-coupled structure (SCS) for understanding high-dimensional sequences. SCS can develop the awareness of the spatial and temporal concepts and decouple them to understand sequential information. Such ability means that machine learning model can autonomously develop spatial and temporal concepts, which is intelligence at a higher level. Furthermore, the separation of spatial and temporal information makes "concept editing" possible.

References

[1] Kitamura, T. etal. Entorhinal cortical ocean cells encode specific contexts and drive context-specific fear memory. Neuron 87, 1317-1331 (2015).

[2] Simonyan, K. & Zisserman, A. Two-stream convolutional networks for action recognition in videos. In Proceedings of the 27th International Conference on Neural Information Processing Systems 568-576 (ACM, 2014).

[3] https://motchallenge.net/tracker/3184&chl=5

[4] Feichtenhofer, C., Fan, H., Malik, J., & He, K. (2019). Slowfast networks for video recognition. In Proceedings of the IEEE International Conference on Computer Vision (pp. 6202-6211).

[5] Tang, J., Xia, J., Mu, X., Pang, B., & Lu, C. (2020). Asynchronous Interaction Aggregation for Action Detection. arXiv preprint arXiv:2004.07485.

 

Translated  by Zhou Rong     Reviewed by Wang Bingyu