📝 Deep learning OCR system for handwritten text recognition using CTC loss & CNN-BiLSTM architecture. Features include image preprocessing with adaptive frame splitting, time-distributed convolutional layers for feature extraction, bidirectional LSTM for sequence modeling, batch normalization for training stability. LER & SER performance metrics

A deep learning framework for HIV-1 drug resistance prediction using context-aware fusion of sequence and mutation features, with hybrid rule–model evaluation and robustness analysis. The model demonstrates adaptive feature reliance and strong cross-drug generalization.

Linear-time sequence modeling that replaces attention's O(n²d) complexity with O(nd) summation-based aggregation. Demonstrates constraint-driven emergence: how functional representations can develop from optimization pressure and architectural constraints alone, without explicit pairwise interactions.

Chunked parallel associative scan in Triton: varlen packing, bidirectional, torch.compile. Drop-in primitive for gated linear RNNs.

Language Identification using Markov Chains, Hidden Markov Models, and Neural HMMs. Explores probabilistic and deep learning methods for sequence modeling, with ablation studies on context, hidden states, and optimization.