Small-scale research in local continual learning and recurrent sequence models, with reproducible experiments, matched baselines, and documented limits.
Drug-target affinity prediction · TransGNN combines Transformer and GNN for DAVIS/KIBA regression, with PyTorch training and preprocessing scripts.
A bit-native predictive machine: next-bit (0/1) prediction via a learned binary-address content-addressable memory. No tokens, no embeddings — vocabulary of 2. An honest, adversarially-verified research log.
GRU sequence model predicts flood risk 24-72h ahead over Maharashtra's Godavari, Krishna & Konkan basins. Upstream-lag spatial features + isotopic calibration (4 risk tiers). Baseline ablation with stacking classifier (KNN+SVM+RF+DT→LR). Time-based 70/15/15 splits. Open-Meteo 5yr dataset (27K+ station-days).
To associate your repository with the sequence-model topic, visit your repo's landing page and select "manage topics."