About this document
Gradient-Based Multi-Task Learning Optimization by zhekai0625 is a document available to read on EtoBox.
The document presents a novel gradient-based multi-objective learning algorithm called Exact Pareto Optimal (EPO) Search, designed to find preference-specific Pareto optimal solutions in multi-task learning (MTL) scenarios. It addresses the limitations of existing methods by enabling the exploration of trade-offs between conflicting tasks while ensuring robustness to initialization and scalability for large-scale deep networks. Experimental results demonstrate that EPO Search outperforms current state-of-th
- Author
- zhekai0625
- Language
- EN