Applied Machine Learning in Early-Stage Start-Ups
Applying machine learning in an early-stage start-up is particularly challenging. Unlike an established business where optimization objectives are well-defined, a "Day-One" company may change the target metrics rapidly to tackle the challenge of growth, which leaves training data small and noisy. Instead of waiting for big data to accumulate, we leverage multiple machine learning techniques, including Ensemble Learning, Deep Learning, and Causal Inference, to jointly improve search quality and help drive growth.
In this talk, we will investigate factors needed to be considered when we are designing a search ranking model in an early-stage start-up environment. We will walk through examples of designing machine learning algorithms for different types of applications, and introduce how an optimization objective can be selected, mathematically defined, and optimized with a machine learning framework. Part of the presentation is based on Airbnb Experiences, the second start-up of Airbnb focusing on recommending what to do when people travel.
Liang Wu is a machine learning data scientist working in the Search and Relevance team at Airbnb, focusing on core search ranking and recommendation of local tours and activities for Airbnb's second startup - Experiences. He previously worked on product search and web search at Etsy and Microsoft Research. He serves as a program committee member for major AI conferences such as AAAI, SIGIR, KDD and WSDM. Liang received his PhD from Arizona State University. During his academic career, he published over 30 papers, authored 2 book chapters, achieved 3rd place of KDD Cup, and received 3 patent awards that have been cited by Alibaba, Baidu, Microsoft, Tencent, The Fourth Paradigm, etc. His thesis research concentrated on building robust machine learning models with noisy data and inaccurate labels.