Robust machine learning for multimodal multitask learning
Abstract
This project aims to design and implement robust learning and data-centric optimization techniques for advancing state-of-the-art machine learning algorithms where data is geographically distributed, sensitive, and large. Robust machine learning and data-centric optimization algorithms empower models through multi-level (local, global and hybrid) training, learning, and inference with data-centric optimization for scarce data and non-standard model settings. By creating unique features (e.g., decentralized training, learning and inference, fault-tolerant against failures and attacks, data-centric optimization, robustness), this project addresses the challenges in the following areas: robust learning; learning with large-data and multimodal settings; lack of theoretical knowledge to build manual models; computation efficient learning and optimization for obtaining more accurate and robust models with applications to constraint environments (i.e., legal search, healthcare systems) and edge infrastructures.
