Calum Heggan

Calum Heggan

PhD Candidate @ University Of Edinburgh & partnered with Thales UK. Interested in Few-Shot learning and Anomaly Detection for sequential signals

Talks and presentations

MetaAudio: A Few-Shot Audio Classification Benchmark

September 06, 2022

Talk, ICANN22 (International Conference of Artificial Neural Networks), Bristol (UWE), England

Abstract Currently available benchmarks for few-shot learning (machine learning with few training examples) are limited in the domains they cover, primarily focusing on image classification. This work aims to alleviate this reliance on image-based benchmarks by offering the first comprehensive, public and fully reproducible audio based alternative, covering a variety of sound domains and experimental settings. We compare the few-shot classification performance of a variety of techniques on seven audio datasets (spanning environmental sounds to human-speech). Extending this, we carry out in-depth analyses of joint training (where all datasets are used during training) and cross-dataset adaptation protocols, establishing the possibility of a generalised audio few-shot classification algorithm. Our experimentation shows gradient-based meta-learning methods such as MAML and Meta-Curvature consistently outperform both metric and baseline methods. We also demonstrate that the joint training routine helps overall generalisation for the environmental sound databases included, as well as being a somewhat-effective method of tackling the cross-dataset/domain setting.

What is Meta-Learning? An Introduction to the Setting and Algorithms

December 09, 2021

Talk, University Of Edinburgh, Department of Engineering (Institute for Digital Communications), Edinburgh, Scotland

Abstract To date, the majority of the breakthroughs seen in machine learning have been in domains or settings where there was an abundance of data, either real or simulated. In contrast, the capability for humans to quickly recognise and discriminate between types of phenomena, for example in visual or acoustic settings, remains unmatched. Meta-learning, or learning to learn, has proven to be a popular and successful framework in dealing with these kind of few-shot problems. In this presentation I will give some general context to how these frameworks are created and used as well as some details on the different algorithms that have been a staple of the field.

Improving Generalisation of Meta-learning for Sequential Data Anomaly Detection

September 21, 2021

Talk, Thales UK,

Abstract Meta-learning, or learning to learn, has proven to be a popular framework in dealing with few shot problems in recent times. Most of the work using this family of techniques focuses on the image domain, leaving many other interesting data modalities and settings largely unexplored. This project looks at expanding this scope of the available research by considering both time series data and the task of anomaly detection. In the first year of the PhD, work has largely been devoted to time series, focusing specifically on few-shot acoustic classification and its standardisation within literature.