One paper has been accepted for presentation at AAAI2025 (A*-ranked)

Our joint work HiGDA has been accepted for presentation at AAAI2025.

Hung Ba Ngo *, Doanh C. Bui *, Nhat-Tuong Do-Tran, Choi Tae Jong

(*) Dr. Hung Ngo and Doanh Bui equally contributes to this work.

In this study, we formulate the domain adaptation problem as a hierarchical graph of nodes. From this perspective, an individual image is considered a local graph of nodes, and a set of image embeddings is considered a global graph, where edges exist if two nodes belong to the same category.

Some visualizations show that our local graph successfully aggregates nodes of interest, which are discriminative for prediction:

We will provide the source code of HiGDA very soon.