Faheem Shaukat

Deep learning has transformed medical image analysis, achieving high accuracy in classification, segmentation, and diagnosis. However, the substantial computational and energy demands of advanced models, such as large convolutional neural networks (CNNs) and multimodal vision-language systems, hinder their deployment on resource-constrained edge devices like smartphones, especially in mobile health. This challenge is critical in low-resource, bandwidth-limited settings with limited high-performance computing access, exacerbating healthcare disparities. This paper introduces MobileNet-Lite-Health (MLH), a lightweight CNN framework optimized via quantization, pruning, and knowledge distillation for sustainable medical imaging inference.We assess MLH on MedMNIST (PathMNIST subset) and HAM10000 datasets, representing benchmark and clinical contexts, respectively. Key contributions include: (1) a carbon-aware evaluation pipeline integrating CodeCarbon for CO2 tracking and real-time battery profiling on devices like the Google Pixel 6; (2) comparisons with ResNet-18 and MobileNetV2, showing 74% energy reduction and 76% CO2 decrease per inference with a 2% accuracy drop; and (3) scaling simulations projecting approximately 1.1 g annual CO2 savings per clinic for 50,000 screenings. MLH serves as a green alternative to energy-intensive models like GPT-4V and MedPaLM, unsuitable for edge use. Experimental results confirm MLH’s effectiveness, with 84.7% accuracy on PathMNIST and 80.1% on HAM10000, plus low latency (65 ms) and memory use (95 MB). This work pioneers carbon-aware edge AI, promoting sustainable, accessible healthcare in resource-scarce environments by balancing efficiency, utility, and environmental impact.

Movie genre classification is essential for organizing cinematic content, enhancing recommendation systems, and supporting market analysis. Unimodal approaches relying solely on plot summaries often fail to capture experiential insights from user reviews, limiting their ability to model genre-specific nuances. Multimodality, which integrates diverse data sources such as textual narratives and user-generated feedback, addresses this by leveraging complementary information for richer representations. This paper introduces the Multimodal Graph Attention Network (MGAN), a novel framework that integrates Graph Neural Networks (GNNs) with a late fusion attention mechanism to model relational structures in plot summaries and user reviews. Utilizing DeBERTa (Decoding-enhanced BERT with disentangled attention) for plot embeddings, leveraging its strength in capturing deep contextual nuances in narrative texts, and MiniLM (Miniature Language Model) for review embeddings, benefiting from its computational efficiency and effectiveness in processing short, sentiment-rich content, MGAN leverages the Trailers12K and Large Movie Review Dataset (LMRD) to achieve a micro-average precision (μAP) of 87.50%, surpassing state-of-the-art models like DeBERTa (80.07%) and the Genre Attention Model (83.63%). MGAN notably achieves a Hamming Loss of 0.069, indicating high prediction fidelity in multi-label classification tasks. MGAN offers significant improvements in both performance and interpretability, supported by SHAP (SHapley Additive exPlanations) explainable AI. Its design strikes a balance between relational learning and computational efficiency, making it a robust solution for scalable, text-based multimedia genre classification in real-world applications.

Automatic movie genre detection is vital for improving content recommendations, user experiences, and organization. Multi-label generation detection assigns multiple labels to a movie, recognizing a movie’s diverse themes. Although there are many existing methods for generating multiple genre labels from movies, they do not provide comprehensive analysis and visual depiction. This work introduces GenVis, a visualization system that provides a better understanding of multi-label genres extracted from movie trailers. The system initially uses text and visual features to classify trailers, assigning multiple genre labels and probabilities. Next, GenVis provides four visualization views: a video view for trailer observation, an overall genre view for getting insights into genre distribution, a genre timeline view for temporal genre evolution, and finally, a genre flow summary for more focused genre analysis. The system allows users to pause the frames, sort the results, and process multiple videos. The multi-label classification is rigorously evaluated using MSE, cross-entropy loss, precision, recall, and F1-score metrics, achieving high accuracy and demonstrating strong genre correlations with notable precision in effectively classifying and distinguishing movie genres. Additionally, a user evaluation for visualization evaluation demonstrated the effectiveness and intuitive usability of GenVis with a high overall rating of 4.25 out of 5.