The paper proposes a theory-inspired loss that minimizes real-synthetic prediction discrepancy and local robustness, and reports state-of-the-art few-shot accuracy, but the main bound is not a valid guarantee for the trained classifier as stated.
IS SYNTHETIC DATA FROM GENERATIVE MODELS READY FOR IMAGE RECOGNITION ? In The Eleventh International Conference on Learning Representations, 2023
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Provably Improving Generalization of Few-Shot Models with Synthetic Data
The paper proposes a theory-inspired loss that minimizes real-synthetic prediction discrepancy and local robustness, and reports state-of-the-art few-shot accuracy, but the main bound is not a valid guarantee for the trained classifier as stated.