Robust Artificial Intelligence for Noisy and Private Data
Lydia Y. Chen
Artificial Intelligence (AI) and Machine Learning (ML) are transforming how today’s society, science, and business operate unprecedentedly, example including self-driving car, disease diagnosis, and AI stock trading. Huge data volume and variety greatly power up AI, and at the same time introduce new challenges arising from data poisoning and privacy concerns. In this talk, I will first discuss limitations of training ML models when encountering noisy stream data with privacy constraints, e.g., classification accuracy and utility loss. I will present a novel learning framework that combines human and artificial intelligence in distilling the detrimental impact of noisy data. Specific examples on classifying images with dirty labels will be given. I will conclude this talk by the challenges and potential solutions of generalizing and applying a plethora of trained models on applications, particularly for edge devices.
29.11.2019 | 12:00 - 13:15