Herlock Rahimi

Publications

• 2025

• Convergence of Agnostic Federated Averaging

  Herlock (SeyedAbolfazl) Rahimi, Dionysis Kalogerias

  CAMSAP 2025 Student Best Paper Contest (Top 5) FL · Theory Rate: O(1/√T)
  arXiv DOI

  We introduced availability in context of Federated Learning that changes FL paradigm. Proved Convergence in this new paradigm.

• Weighted Stochastic Differential Equation to Implement Wasserstein-Fisher-Rao Gradient Flow

  Herlock Rahimi

  Preprint Information Geometry · Optimal Transport Wasserstein Fisher Rao Geometry
  arXiv DOI

  When Optimal Transport Meets Information Geometry to implement Faster Sampling. A review Paper.

• FedAVOT: Exact Distribution Alignment in Federated Learning via Masked Optimal Transport

  Herlock (SeyedAbolfazl) Rahimi, Dionysis Kalogerias

  Preprint ICASSP (submitted) FL · Optimal Transport Rate: O(1/√T)
  arXiv DOI

  Proposes transport-based aggregation to align availability and importance distributions via masked OT and Sinkhorn scaling; proves convergence under nonsmooth convex settings and strong empirical gains in low-availability regimes.

• Uncertainty-Aware Generative Oversampling Using an Entropy-Guided Conditional Variational Autoencoder

  Amirhossein Zare, Amirhessam Zare, Parmida Sadat Pezeshki, Herlock (SeyedAbolfazl) Rahimi, Ali Ebrahimi, Ignacio Vázquez-García, Leo Anthony Celi

  Preprint ML · Biomedical CVAEs · Class Imbalance
  arXiv DOI

  Introduces entropy-guided weighting and sampling for CVAE oversampling (LEO-CVAE), boosting performance on ADNI and TCGA lung cancer datasets versus SMOTE-style and generative baselines.