IPAB Workshop - 08/05/2025

Speaker: Miaowei Wang

Title: DecoupledGaussian: Object-Scene Decoupling for Physics-Based Interaction

Abstract: We present DecoupledGaussian, a novel system that decouples static objects from their contacted surfaces captured in-the-wild videos, a key prerequisite for realistic Newtonian-based physical simulations. Unlike prior methods focused on synthetic data or elastic jittering along the contact surface, which prevent objects from fully detaching or moving independently, DecoupledGaussian allows for significant positional changes without being constrained by the initial contacted surface. Recognizing the limitations of current 2D inpainting tools for restoring 3D locations, our approach uses joint Poisson fields to repair and expand the Gaussians of both objects and contacted scenes after separation. This is complemented by a multi-carve strategy to refine the object’s geometry. Our system enables realistic simulations of decoupling motions, collisions, and fractures driven by user-specified impulses, supporting complex interactions within and across multiple scenes. We validate DecoupledGaussian through a comprehensive user study and quantitative benchmarks. This system enhances digital interaction with objects and scenes in real-world environments, benefiting industries such as VR, robotics, and autonomous driving.

Speaker: Bashayer Marghalani

Title: Physics-Informed Neural Networks in Medical Diagnosis

Abstract: For decades, accurate medical diagnoses have been and still are essential in saving lives, yet conventional diagnostic methods often rely on extensive labelled data and expert interpretation. These would lead to challenges in scalability and efficiency. Integrating Physics-Informed Neural Networks (PINNs) into medical diagnostics offers a promising approach by embedding physical laws into deep learning models. The advantage of that is addressing key limitations of traditional data-driven techniques. While conventional deep learning depends heavily on large datasets and may lack interpretability, PINNs incorporate governing physics equations into the training process. These incorporations ensure that predictions remain physiologically and biochemically consistent and will lead to accurate decisions regarding the underlying medical problem. Furthermore, many non-invasive medical imaging, such as Magnetic Resonance Imaging (MRI), Magnetic Resonance Spectroscopy (MRS), and Raman Spectroscopy (RS), are used these days for acquiring a patient’s body scan information quickly and effectively.

This research explores the application of PINNs in spectroscopic or image datasets like MRI, MRS, and RS for diagnosing brain pathologies, leveraging the benefit of the non-invasive nature of these data collection techniques and using metabolic insights to improve early disease detection. Conditions such as glioma, ischemia, and Alzheimer’s disease often exhibit subtle biochemical changes that are difficult to capture using conventional machine-learning models. This study aims to enhance diagnostic precision, classify several brain diseases, and provide clinically interpretable predictions by integrating physics-informed constraints with neural networks. The proposed methodology involves dataset acquisition, preprocessing, PINN model development, training, validation, and performance evaluation using established medical imaging and/or spectroscopy datasets.

The expected result is a robust and interpretable AI-driven diagnostic methodology that exceeds existing accuracy, efficiency, and reliability. This work will advance automated, scalable, and clinically applicable diagnostic systems by bridging the gap between physics-based algorithms and machine learning and ultimately supporting early detection and improved patient outcomes in neurological disorders.