Fragkiskos D. Malliaros

Atompack achieves a staggering 96x improvement in read performance for atomistic ML datasets, revolutionizing how we handle training data efficiency.

Spectral GNNs' generalization ability hinges on avoiding frequency amplification, and this paper provides a Fourier-domain lens to see it clearly.

Ditch the data drop: training-free graph imputation lets multimodal recommender systems handle missing data better than ever, boosting performance without retraining.