Scaling Video Ad Classification Across Millions of Classes with GenAI

Explore a 24-minute conference talk on scaling video ad classification using GenAI across millions of classes. Discover how Vivvix, an ad-intel company, leverages real-time insights from diverse creatives using ML and GenAI. Learn about their journey from categorizing video ads into 30,000 product classes to expanding to six million. Understand the challenges faced with the initial transformer-based machine learning model and how they addressed these issues using OSS LLMs. Gain insights into their use of optimized LLama2(Vllm) for categorizing creatives and conducting similarity searches across labels. Compare the performance of their baseline machine learning model with the integrated LLM approach, which achieved a 15% uplift in accuracy. Explore their innovative solution combining both approaches, where the LLM model acts as a pre-processing step for subsequent machine learning analysis. Presented by Dong-Hwi Kim, Senior Machine Learning Engineer at Vivvix, and Puneet Jain, Senior Specialist Solutions Architect at Databricks.