GCTAF — Time-Series Forecasting & Flare Risk Classification
Attention-based forecasting of magnetic field trajectories combined with supervised contrastive learning for solar flare classification.
Problem
Solar flare prediction suffers from class imbalance, irregular temporal gaps, and missing values across SHARP magnetic features.
Approach
Two-stage learning: contrastive pretraining with missing-aware triplets and focal loss, then linear probing + fine-tuning.
Architecture
8 SHARP features → attention encoder → 60-step delta forecasts. Parallel contrastive head → flare risk classifier.
Results
Beat persistence baseline on forecasting; TSS 0.75 on flare classification.
Lessons learned
Representation learning with missing-aware triplets dominates direct supervised training under heavy class imbalance.
Next
Fed-BLEND — Federated Conformal Prediction for VLMs
Novel federated conformal prediction method that mitigates hallucinations in federated fine-tuned vision-language models via abstention.
Skin Lesion Classification via Dual-Weighted Learning
Imbalanced medical image classification using MobileViT v2 with dual-weighted learning — inverse square-root sampling + ENS loss — and Grad-CAM heatmaps for clinical interpretability.