A Continual Machine Learning Framework for Accelerating Scientific Discovery
Abstract
Machine learning models, which have found tremendous success in several commercial applications where large-scale data is available (e.g., computer vision), are beginning to play an important role in scientific disciplines such as life sciences and biomedicine. Over the past few years, several knowledge discovery frameworks have been proposed and shown great promise in the areas of drug repurposing, precision medicine, and clinical care. Despite significant advances made, most of the existing approaches adopt a static learning paradigm, wherein the approaches are unable to continually accommodate the new information from the non-stationary data distributions. As a result, the static learning paradigm becomes limiting for interdisciplinary domains that evolve at a breakneck speed. For instance, in the life-sciences domain alone, around 3,000 articles are added to the repository every single day. This rapid proliferation of scientific information necessitates the development of innovative knowledge discovery systems that are capable of continually adapting to the new information without having to retrain the complex deep learning models from scratch. To address this, we propose to shift the research direction from the currently dominant paradigm of static learning to continual learning, wherein the proposed approaches continually learn over time by accommodating the new information while retaining the previously acquired knowledge. This continual learning strategy enables the proposed knowledge discovery systems to efficiently acquire, refine, transfer knowledge over sustained periods of time and generate actionable insights for driving scientific discoveries.
Committee:
- Hongning Wang, Committee Chair, (CS/SEAS/UVA)
- Aidong Zhang, Advisor, (CS, BME/SEAS, SDS/UVA)
- Jundong Li (CS, ECE/SEAS, SDS/UVA)
- Nathan Sheffield (Public Health Sciences/SEAS/UVA)
- Jing Gao (Purdue University)