Jun Xu

Shipeng Fu, Zhen Li, Jun Xu et al.

Knowledge distillation is a standard teacher-student learning framework to train a light-weight student network under the guidance of a well-trained large teacher network. As an effective teaching strategy, interactive teaching has been widely employed at school to motivate students, in which teachers not only provide knowledge but also give constructive feedback to students upon their responses, to improve their learning performance. In this work, we propose an InterActive Knowledge Distillation (IAKD) scheme to leverage the interactive teaching strategy for efficient knowledge distillation. In the distillation process, the interaction between teacher and student networks is implemented by a swapping-in operation: randomly replacing the blocks in the student network with the corresponding blocks in the teacher network. In the way, we directly involve the teacher's powerful feature transformation ability to largely boost the student's performance. Experiments with typical settings of teacher-student networks demonstrate that the student networks trained by our IAKD achieve better performance than those trained by conventional knowledge distillation methods on diverse image classification datasets.

Adele Lu Jia, Xiaoxue Shen, Siqi Shen et al.

Crowdsourced video systems like YouTube and Twitch.tv have been a major internet phenomenon and are nowadays entertaining over a billion users. In addition to video sharing and viewing, over the years they have developed new features to boost the community engagement and some managed to attract users to donate, to the community as well as to other users. User donation directly reflects and influences user engagement in the community, and has a great impact on the success of such systems. Nevertheless, user donations in crowdsourced video systems remain trade secrets for most companies and to date are still unexplored. In this work, we attempt to fill this gap, and we obtain and provide a publicly available dataset on user donations in one crowdsourced video system named BiliBili. Based on information on nearly 40 thousand donators, we examine the dynamics of user donations and their social relationships, we quantitively reveal the factors that potentially impact user donation, and we adopt machine-learned classifiers and network representation learning models to timely and accurately predict the destinations of the majority and the individual donations.