BME Research Seminar: Functionals, Neural Nets, and Beyond: Navigating Semi-Supervised Graph Learning and Implicit Neural Representations

BME Research Seminar: Functionals, Neural Nets, and Beyond Navigating Semi-Supervised Graph Learning and  Implicit Neural Representations

Speaker: Dr Angelica Aviles-Rivero, Senior Research Associate, Department of Applied Mathematics and Theoretical Physics, University of Cambridge

Time: 11:00 am - 12:00 noon (HK Time)

Mode: Hybrid

Venue: Room Y410, The Hong Kong Polytechnic University

Zoom: https://bit.ly/45DQ3NN (Meeting ID: 969 6861 9624; Passcode: 455490)

Overview of the Lecture

In this talk, we delve into two pivotal subjects. The first topic revolves around the development of hybrid graph models tailored to the complexities of multi-modal data. We present a novel semi-supervised hypergraph learning framework, specifically designed for diagnostic purposes. Our approach adopts a hybrid perspective, where we introduce a new methodology centered on a dual embedding strategy and a semi-explicit flow. To illustrate the efficacy of our proposed model, we employ it within the realm of Alzheimer's disease diagnosis, demonstrating its capacity to uncover latent relationships within intricate multi-modal data. 
 
Transitioning seamlessly to the second subject, we delve into implicit neural representations. We introduce an innovative function designed to harness the strengths of Strong Spatial and Frequency attributes, marking a departure from conventional methods. Remarkably, our novel technique showcases exceptional enhancements in performance across a diverse array of downstream tasks, notably encompassing CT reconstruction and denoising applications. Through rigorous experimentation, we elucidate the advantages enabled by our approach.
 
Concluding our discourse, we shed light on emerging trends within Diffusion models, reflecting on their evolving role and potential. By amalgamating insights from hybrid graph models, implicit neural representations, and diffusion modelling, this talk serves as a comprehensive exploration of transformative techniques and trends in the field.

About the Speaker

Dr Angelica Aviles-Rivero is a Senior Research Associate at the Department of Applied Mathematics and Theoretical Physics (DAMTP), University of Cambridge. Her work focuses at the confluence of computational mathematics, computer vision, and machine learning, where she addresses complex real-world problems on a large scale. Her expertise lies in developing large-scale mathematical and machine learning models with minimal or even no supervision. This has led to her being sought after for consultations by various centers and companies. 

Dr Aviles-Rivero has large experience in organising  scientific  including BMVC2022 & BMVC 2023 (co-organiser), MIUA22 (co-organiser),  MICCAI Tutorials, ACCV Tutorials, IGARSS Tutorials,  and GeoMedIA Workshop. She serves as a current SIAM SIAG/IS Officer.