Mohammed Kaif Pasha

Shahnawaz Alam, Mohammed Mudassir Uddin, Mohammed Kaif Pasha

A key challenge in agricultural AI is deploying disease detection systems in remote fields with limited access to laboratories or high-performance computing (HPC) resources. While deep learning (DL) models, specifically deep convolutional networks, achieve high accuracy in identifying plant pathologies from leaf imagery, their memory footprints and computational demands limit edge deployment on devices constrained by battery life, processing power, and connectivity, such as Raspberry Pi. Few-shot learning (FSL) paradigms offer a compelling solution to the data scarcity problem inherent in agricultural applications, where obtaining labeled samples for novel disease variants proves both costly and time-sensitive. This work introduces a framework combining pruning with meta-learning for agricultural disease classification, addressing the tension between generalization capability and deployment feasibility. The proposed approach combines a novel Disease-Aware Channel Importance Scoring (DACIS) mechanism with a three-stage Prune-then-Meta-Learn-then-Prune (PMP) pipeline. Experiments on PlantVillage and PlantDoc datasets demonstrate that the proposed approach reduces model size by 78\% while maintaining 92.3\% of the original accuracy. The compressed model achieves 7 frames per second (FPS) on a Raspberry Pi 4, enabling practical real-time field diagnosis for smallholder farmers.

Mohammed Mudassir Uddin, Shahnawaz Alam, Mohammed Kaif Pasha

Mechanistic interpretability seeks to reverse-engineer neural network computations into human-understandable algorithms, yet extracting sparse computational circuits from billion-parameter language models remains challenging due to exponential search complexity and pervasive polysemanticity. The proposed Hierarchical Attribution Graph Decomposition (HAGD) framework reduces circuit discovery complexity from O(2^n) exhaustive enumeration to O(n^2 log n) through multi-resolution abstraction hierarchies and differentiable circuit search. The methodology integrates cross-layer transcoders for monosemantic feature extraction, graph neural network meta-learning for topology prediction, and causal intervention protocols for validation. Empirical evaluation spans GPT-2 variants, Llama-7B through Llama-70B, and Pythia suite models across algorithmic tasks and natural language benchmarks. On modular arithmetic tasks, the framework achieves up to 91% behavioral preservation ($\pm$2.3\% across runs) while maintaining interpretable subgraph sizes. Cross-architecture transfer experiments suggest that discovered circuits exhibit moderate structural similarity (averaging 67%) across model families, indicating potential shared computational patterns. These results provide preliminary foundations for interpretability at larger model scales while identifying significant limitations in current attribution methodologies that require future advances.