On-device Deep Learning
On-device deep learning aims to enable privacy-preserving, always-on intelligence at the edge. Unfortunately, deep learning algorithms demand extensive computation and storage, limiting their adoption to resource-constrained devices. At TEC, we are researching algorithm and software support for deep learning on resource-constrained mobile and embedded platforms. Particularly, we focus on model compression and resource scheduling for the efficient execution of deep neural networks under application-driven resource constraints.
Publications
- X. He, Z. Zhou, L. Thiele, ‘‘Multi-Task Zipping via Layer-wise Neuron Sharing’’, in Proceedings of Advances in Neural Information Processing Systems (NeurIPS), 2018 external page [Code].
- S. Liu, Y. Lin, Z. Zhou, K. Nan, H. Liu, J. Du, ‘‘On-Demand Deep Model Compression for Mobile Devices: A Usage-Driven Model Selection Framework’’, in Proceedings of the ACM International Conference on Mobile Systems, Applications, and Services (MobiSys), 2018.
- Z. Qu, Z. Zhou, Y. Cheng, L. Thiele, ‘‘Adaptive Loss-Aware Quantization for Multi-bit Networks’’, in Proceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2020 external page [Code].