Deep Learning (AI 2.0) has rapidly become the main stream for machine learning research and development. However, serious technical barriers still effectively prevent its cost-effective deployment to mobile/edge applications. Most notably: the network complexity, power consumption, and long latency. This calls for a critical need of systematical design of high-performing and low-complexity network models. This talk proposes a viable solution using a notion of structural gradient for network structural design. Pursuant to our simulation finding, X-learning achieves high accuracy with substantially reduced complexity and, by and large, outperforms other state-of-the-art methods, e.g.it edges the 2018 LPIRC winner when applied to ImageNet classification and (b) outperforms the 2018 PIRM winner in SR Imaging applications.

活動時間：2019年12月17日 13:20-15:10

活動地點：交通大學光復校區合勤講堂