ANC Workshop - Ian Simpson, Milad Jabbari

Tuesday, 23rd May 2023

Use of knowledge graphs in disease research - Ian Simpson

Abstract: Networks have been used extensively to model biological and biomedical datasets for many years. Typically these have been mono-partite with modest if any feature details encoded, their analysis has been dominated by unsupervised clustering followed by post-hoc assessment of cluster features. I will introduce some emerging applications of more complex network models including “knowledge” graphs for the biomedical domain. These are well suited to modelling multi-modal data from cohort based longitudinal clinal genomics studies. I will illustrate these topics using examples from our ongoing work in Breast Cancer, Autism, and Parkinson’s disease.

A Specific Deep Learning Model for Real-Time Hand Gesture Control in Amputee people - Milad Jabbari

Abstract: Amputees face numerous challenges and limitations in accomplishing daily tasks due to the loss of an upper limb. Although advanced and powered prostheses are available commercially, only a small percentage of individuals with upper limb differences are able to benefit from them. Despite significant academic attention to understanding user needs over the past two decades, the abandonment rate of such devices remains unchanged at 44%. There are several reasons for this user rejection, including the limited functionality and high cognitive effort required by most commercially available prostheses, which utilize an on-off control strategy. In contrast, utilizing surface electrocardiogram (sEMG) signals from stump muscles as input for Machine Learning (ML) and Deep Learning (DL) models in upper-limb hand prosthetics offers a reliable control strategy. This approach enables the restoration of a greater range of gestures compared to existing commercial systems. Consequently, it can be confidently stated that the mapping of raw EMG signals for triggering prosthetic hands is a critical aspect of a pattern-recognition-based system. A desirable solution for real-time EMG-based hand gesture control is a model that blends the strengths of DL models with the simplicity of basic ML models. By harnessing the advantages of both approaches, this model can attain optimal performance while remaining user-friendly. The Temporal Convolutional Network (TCN) is a CNN variant that employs multiple layers of one-dimensional dilated causal convolutions. This architecture is particularly suitable for handling time-series data and sequence modeling tasks.

Event type: Workshop

Date: Tuesday, 23rd May 2023

Time: 11:00

Location: G.03

Speaker(s): Ian Simpson, Milad Jabbari