Post by Kanchon Kanti Podder

I'm excited to share that our latest paper is now available online! "DILoc: Replay-Based Domain-Incremental Learning for Lifelong Multimodal WiFi–Magnetic Indoor Localization" Indoor localization systems often degrade over time as environments evolve due to layout changes, new WiFi access points, and shifting signal characteristics. In this work, we introduce DILoc, a continual learning framework that enables indoor localization models to adapt continuously without retraining from scratch. Our approach combines: - 📍 Domain-incremental learning for evolving indoor environments - 🧠 A novel ReSCUR replay memory mechanism for effective knowledge retention - 📡 Multimodal fusion of WiFi RSSI and magnetic field measurements - 🔄 A balanced replay strategy that mitigates catastrophic forgetting while learning from new environments Our experiments demonstrate that DILoc achieves 53.9% of predictions within 1 meter. This work sits at the intersection of: - Continual Learning - Edge AI - Wireless & Mobile Computing - IoT Systems - Indoor Localization - Multimodal Machine Learning As I continue my research, I'm particularly interested in developing efficient, adaptive, and trustworthy AI systems that can learn continuously in dynamic real-world environments without catastrophic forgetting. I hope this work contributes to advancing lifelong learning for next-generation intelligent IoT and robotics systems. I'm sincerely grateful to my co-author (Pritom Dutta ) for their outstanding collaboration and to my mentors(Jian Zhang , Shiwen Mao , Xiangyu Wang ) for their invaluable guidance, support, and encouragement throughout this research. I'd be happy to discuss potential collaborations or exchange ideas with researchers and practitioners working in continual learning, wireless sensing, robotics, and AI systems. Find the full paper : https://lnkd.in/eRFYCZAf #ContinualLearning #ArtificialIntelligence #IoT #IndoorLocalization #WirelessSensing #EdgeAI #Robotics