Machine Learning Berlin

Improving Fashion Discovery

by Pedro Silva

07 Aug 2017

Title: Improving fashion discovery with multimodal representation learning.

Abstract: This talk will cover a multimodal learning to rank model that combines traditional ecommerce search engine features with visual semantic features extracted from a product’s image. By using multimodal representations and effectively capturing fine-grained clothing style from images, we’ll show that ranking quality is significantly improved.The talk will detail the methods used and lessons learned from training these models.

Bio: Pedro is a co-founder at Cerebel, a Berlin-based startup developing visual and natural language search technology for fashion.

Meetup Link:

brand brand2 meetup