Hao Gong presents on Meta-Profile Network at ACM CIKM 2020

EVENTS
 
Hao Gong, Research Scientist of RIT Tokyo, presented his accepted paper at this year's ACM CIKM. The theme for ACM CIKM 2020 was data and knowledge for the next generation: sustainability, transparency and fairness. The conference, celebrating its 29th annual event, sought to identify challenging problems facing the development of future knowledge and information systems, and shaping future directions of research and initiatives. CIKM was conducted virtually this year from October 19-23. 
 
The paper, "Learning to Profile: User Meta-Profile Network for Few-Shot Learning” (https://dl.acm.org/doi/10.1145/3340531.3412722),  was written collaboratively with other members of RIT Tokyo, including Qifang Zhao, Tianyu Li, Derek Cho, and DuyKhuong Nguyen. The paper proposes an industrial meta-learning framework called “Meta-Profile Network” to solve challenging knowledge transfer and fast adaptation problems with few samples in representation learning domain. As one of the sub projects of RIT’s AIPP customer program pre-training model, the paper introduces the following aspects:
 
1) Meta-learning model: In the context of representation learning with e-commerce user behavior data, they propose a meta-learning framework called the Meta-Profile Network, which extends the ideas of matching network and relation network for knowledge transfer and fast adaptation 
2) Encoding strategy: To keep high fidelity of large-scale long-term sequential behavior data, the authors propose a time-heatmap encoding strategy that allows the model to encode data effectively
3) Deep network architecture: A multi-modal model combined with multi-task learning architecture to address the cross-domain knowledge learning and insufficient label problems.
 
The Meta-Profile Network shows significant improvement in the model performance when compared to baseline models. To maximize the value of the Rakuten ecosystem’s data, with advanced deep architecture and encoding strategies, the Meta-Profile Network has proved that it can quickly adapt learned knowledge to down-stream services. The model has better robustness and uncertainty under various extreme conditions as well. Finally, Meta-Profile Network shows strong flexibility and scalability in practice, which could promote meta-learning research and development in large-scale industrial applications.  
 
Hao Gong's research received a lot of interest and  questions about industrial implementation at the conference this year. We want to thank ACM CIKM, attendees, and sponsors.