Summary of Unknown Domain Inconsistency Minimization For Domain Generalization, by Seungjae Shin et al.
Unknown Domain Inconsistency Minimization for Domain Generalization
by Seungjae Shin, HeeSun Bae, Byeonghu Na, Yoon-Yeong Kim, Il-Chul Moon
First submitted to arxiv on: 12 Mar 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The proposed Unknown Domain Inconsistency Minimization (UDIM) objective is designed to enhance domain generalization by reducing the loss landscape inconsistency between the source and unknown domains. This approach is rooted in both parameter and data perturbed regions, aiming to achieve generalization grounded on flat minima for the unknown domains. Theoretically, UDIM establishes an upper bound for the true objective of the domain generalization task. Empirically, UDIM consistently outperforms Sharpness-Aware Minimization (SAM) variants across multiple benchmark datasets, particularly in scenarios with more restrictive domain information. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper introduces a new approach to improve domain generalization by minimizing the loss landscape inconsistency between the source and unknown domains. The proposed method, called Unknown Domain Inconsistency Minimization (UDIM), is designed to enhance transferability of models learned from one domain to unobserved domains. UDIM reduces overfitting to specific domains by aligning the loss landscapes in different domains. This approach outperforms previous methods, such as SAM, and shows promising results in unseen domains. |
Keywords
* Artificial intelligence * Domain generalization * Generalization * Overfitting * Sam * Transferability
