ANC Workshop - Matthias Hennig and Oisin Mac Aodha

Speaker: Matthias H. Hennig

Title:  Pruning, volatility and stability in neural circuits

Abstract:

Neuronal networks undergo significant pruning during development, and are constantly re-modelled even in the mature brain. Yet despite this ongoing plasticity, functionally the networks remain stable. Here I will two modelling studies that show how stability is maintained in presence of volatility. First, using a neural network model we found a biologically plausible pruning rule that optimises network architecture, mirroring developmental pruning. Second, we found that in spiking networks a small group of inhibitory neurons are responsible for maintaining stable activity, allowing plasticity in the remaining neurons without affecting function significantly.

This is joint work with Carolin Scholl, Pia Siegele and Michael Rule

Speaker: Oisin Mac Aodha

Title: Benchmarking Representation Learning on Natural World Image Collections

Abstract:

Recent progress in self-supervised learning has resulted in models that are capable of extracting rich representations from image collections without requiring any explicit label supervision. To date however, the vast majority of these approaches have restricted themselves to training on standard benchmark datasets such as ImageNet. I will present some preliminary work to show that fine-grained visual categorization problems, such as plant and animal species classification, provide an informative testbed for self-supervised learning. In order to facilitate progress in this area, I will discuss two new natural world visual classification datasets that we have developed in collaboration with domain experts. These datasets allow us to explore questions related to large-scale representation and transfer learning in the context of challenging visual concepts. I will present some initial experiments comparing feature extractors trained with and without supervision on these tasks, and shed some light on the strengths and weaknesses of different learned features across a diverse set of tasks.

Bio: Oisin Mac Aodha is a Lecturer in Machine Learning in the School of Informatics at the University of Edinburgh. His current research interests are in the areas of computer vision and machine learning, with a specific emphasis on human-in-the-loop methods such as machine teaching.