Abstract

Zoom link: https://cl-cam-ac-uk.zoom.us/j/93420534307?pwd=dTg2OXRidXZNdTdhSmhscGpJck40dz09

——————————————————————

Speaker: Lucie Charlotte Magister

Title: GCExplainer: Concept-based Explanations for Graph Neural Networks

Abstract: While graph neural networks (GNNs) have been shown to perform well on graph-based data from a variety of fields, they suffer from a lack of transparency and accountability, which hinders trust and consequently the deployment of such models in high-stake and safety-critical scenarios. Even though recent research has investigated methods for explaining GNNs, these methods are limited to single-instance explanations, also known as local explanations. In this talk, I will present GCExplainer, an unsupervised approach for post-hoc discovery and extraction of global concept-based explanations for GNNs.

——————————————————————

Speaker: Zahra Tarkhani

Title: Effective Attack Investigation Across Heterogeneous Isolation Boundaries

Abstract: Recent improvements in hardware isolation support have led to numerous new forms of application compartmentalisation, including intra-address space and TEE /enclave-based mechanisms. These systems aim to reduce the software attack surface by dividing it into isolated components that communicate using well-defined channels. However, many vulnerabilities inside and across such heterogenous isolation boundaries cannot easily be detected or debugged. This talk will propose an extensible framework for investigating attacks in compartmentalised applications across various isolation boundaries such as processes, intra-process sandboxes, and enclaves. Hence developers could analyze, debug, and audit a wide range of non-trivial threats (e.g., memory or synchronization vulnerabilities) across compartments to ensure safe interactions, data/resource sharing, or compartment migrations.

——————————————————————

Speaker: Tong Xia

Title: Uncertain-aware Machine Learning for Biosignal-based Healthcare Applications

Abstract: A biosignal is a signal in human beings that can be continually measured like respiratory sound, heart activity (ECG), brain waves (EEG), etc. Based on that, a number of machine learning models have been developed for disease detection and health status monitoring. Although those models yield promising accuracy, because of the lack of interpretation and the usual over-confident predictions, it is non-trivial to gain people’s trust for their real deployment in the healthcare field. Predictive uncertainty provides the opportunity to inspect a model’s confidence toward its output, thus enabling the less trustworthy predictions to be further checked by medical professions. In this talk, I will reflect on how to incorporate uncertainty estimation to improve the decision-making of ML models for healthcare.

——————————————————————