This talk explores the intersection of Deep Learning, Generative AI, and Federated Learning in advancing solutions for Multimedia Processing, Human-Computer Interaction (HCI), and Medical Healthcare Applications. The session aims to present a set of intelligent frameworks that address real-world challenges through scalable, privacy-aware, and high-performance learning systems. Beginning with biomedical signal analysis, the talk introduces LAACNN, a hybrid architecture combining convolutional, recurrent, and attention mechanisms for classifying neuromuscular diseases using raw EMG signals. This approach enhances early diagnostic potential through non-invasive, data-driven insights. In the realm of advanced multimedia, speech enhancement is tackled through two novel Generative AI models. CAR-UNet employs ConvNeXT blocks and gated attention to jointly enhance magnitude and phase information, ensuring clarity under extreme noise conditions. N2N2N, a clean-data-independent approach using a modified cGAN, refines noisy speech where clean references are unavailable, making it suitable for realistic deployment. For Human-Computer Interaction, the session features CAPRes50-GAN, a real-time, GAN-based classifier for word-level sign language recognition. It leverages multi-head attention alongside ResNet-based architectures to efficiently interpret gestures and support inclusive communication systems. The discussion continues with BreastDCGAN, an end-to-end framework for breast cancer detection using mammography and ultrasound. It combines generative segmentation and attention-based classification to assist in accurate and early medical analysis. Concluding the talk, pFLOCT is introduced—a Personalized Federated Learning framework for optical imaging that accommodates heterogeneous client data while preserving privacy, advancing collaborative learning in medical domains. Speaker(s): Dr. Celia Shahnaz, Dept. of EEE, BUET. Virtual: https://events.vtools.ieee.org/m/485235
