Modular Universal Reparameterization: Deep Multi-task Learning Across Diverse Domains
This work addresses the challenge of general problem-solving in AI by enabling multi-task learning across disparate domains, though it appears incremental in extending existing multi-task learning approaches.
The paper tackles the problem of multi-task learning across diverse domains with different architectures by decomposing tasks into related subproblems and optimizing sharing via a stochastic algorithm. Results show improved performance across vision, NLP, and genomics tasks, confirming the benefits of sharing learned functionality.
As deep learning applications continue to become more diverse, an interesting question arises: Can general problem solving arise from jointly learning several such diverse tasks? To approach this question, deep multi-task learning is extended in this paper to the setting where there is no obvious overlap between task architectures. The idea is that any set of (architecture,task) pairs can be decomposed into a set of potentially related subproblems, whose sharing is optimized by an efficient stochastic algorithm. The approach is first validated in a classic synthetic multi-task learning benchmark, and then applied to sharing across disparate architectures for vision, NLP, and genomics tasks. It discovers regularities across these domains, encodes them into sharable modules, and combines these modules systematically to improve performance in the individual tasks. The results confirm that sharing learned functionality across diverse domains and architectures is indeed beneficial, thus establishing a key ingredient for general problem solving in the future.