Â鶹ŮÀÉ

Juan, Ashan and Suvosree present all of MLab's accepted papers at the 2026 IEEE International Conference on Consumer Electronics (ICCE) held in Dubai, United Arab Emirates. The contributed research spans across multiple domains such as Video Coding standards, Cybersecurity, Applied AI and Smart Cities.

Juan's work shed light on the impact of split point selection in split inference paradigm where a neural network is literally cut to utilize available compute near the sensing device (edge) and send features over to the task device (server) in bandwidth constrained environments. He argued that the point at which the network is split matters and presents trade-offs in terms of compressibility and compute offload. Split at early and mid layers offer greater compute offload for the edge device whereas deeper layers tend to improve compressibility.Ìý

Suvosree introduced a novel adversarial algorithm designed to steer popular classifiers such as YOLO toward misclassification. Her work aims to make existing vision systems more robust towards imperceptible deviations in images. Our undergraduate researcher Christina proposed a speech and gesture-driven video editor named QuickCut that edits videos based on vocal commands such as "rolling", "cut", "zoom", etc. Thus, democratizing the complex task of video editing which is usually reserved for media professionals.

Read the full papers here: