EMAN: Optimization-Driven Capacity Growth through Path Emergence in Multi-Task Learning
Researchers have proposed a new framework for multi-task learning called EMAN (Emergent Modular Atomic Network). Unlike existing methods, EMAN allows networks to grow new paths only when optimization evidence appears. This is achieved through latent relative phases and monitoring multiple decision signals during training. The authors claim that EMAN can adaptively allocate shared and task-specific representation capacity, improving performance at a competitive computational c
Researchers have proposed a new framework for multi-task learning called EMAN (Emergent Modular Atomic Network). Unlike existing methods, EMAN allows networks to grow new paths only when optimization evidence appears. This is achieved through latent relative phases and monitoring multiple decision signals during training. The authors claim that EMAN can adaptively allocate shared and task-specific representation capacity, improving performance at a competitive computational cost. Experiments on various datasets, including PASCAL-Context and NYUv2, show the effectiveness of EMAN.
---
Why it matters: This matters to AI researchers because it offers a new approach to multi-task learning that can adaptively allocate resources. This could lead to improved performance in complex tasks with varying requirements.
Source: https://arxiv.org/abs/2608.16930
This article was originally published at: https://arxiv.org/abs/2608.16930