[4Rin1] Interactive 2

[4Rin1-21] Predictive Optimization based on Zero-Shot Domain Adaptation

〇Tomoya Sakai¹, Naoto Ohsaka¹ (1.NEC Corporation)

Keywords:Zero-shot domain adaptation, Predictive optimization

Making predictions in a new domain without any training sample is an important task in practice and known as zero-shot domain adaptation (ZSDA). ZSDA has various applications such as sales prediction for new products. So far, prediction in a new domain has gotten much attention, but in this paper, we investigate another potential of ZSDA. That is, instead of predicting responses in a new domain, we find a description of a new domain given prediction. This aspect of ZSDA can be regarded as predictive optimization, which allows us to find, e.g., a promising design of new products. We propose a new simple framework for predictive optimization and demonstrate its effectiveness through numerical experiments.