ADAlign tackles graph domain adaptation by adaptively aligning discrepancies via Neural Spectral Discrepancy (NSD). Uses neural characteristic functions and minimax sampling without heuristics. Outperforms SOTA on 10 datasets with efficiency gains.
Key Points
- 1.Captures attribute-structure interplay
- 2.Learnable frequency emphasis
- 3.Robust to varying shifts
Impact Analysis
Simplifies GDA for diverse scenarios, boosting transfer learning reliability. Reduces manual tuning for real-world graphs.
Technical Details
NSD encodes high-order dependencies in spectral domain. Joint alignment of multi-faceted shifts.