ANC Workshop - Lazaros Mitskopoulos, Michael Camilleri
Tuesday, 17th January 2023
Using weighted NMF to characterize non-parametric copulas in neural populations - Lazaros Mitskopouls
Abstract: Information processing in the brain relies on the coordinated activity of neural ensembles. Neural dependencies can have complex shapes, heavy tails and high-order structure. Due to these features, conventional pairwise pearson correlation approaches can be misleading for characterizing neural interactions. Non-parametric copula-based methods address these challenges by capturing dependencies without imposing any assumptions on their structure. In previous work, we demonstrated that normalizing flows are a useful tool for non-parametric estimation of copula densities. However, moderate to large populations present additional challenges as the number of interactions to account for grows rapidly. In this talk, I will discuss our work on using weighted Non-negative Matrix Factorization to obtain a low dimensional description of copula densities in a neural population. As regular NMF struggles with accurately capturing overlapping tail structures, this weighted version of the method places higher importance on the tail regions of copula densities, thus yielding better results.
Of Mice and ... More Mice! Identifying the Behaviours of Group Housed Mice - Michael Camilleri
Abstract: In this talk I will go over my recent work on classifying individual mouse behaviours in the close confines of a home-cage. This is achieved by fusing video data and RFID location information. The challenges here include (a) the lack of distinguishing visual features for identifying each mouse, (b) multiple individuals with varying behaviours, (c) limited visibility (frequent occlusion) and (d) imbalanced datasets.
Event type: Workshop
Date: Tuesday, 17th January 2023
Speaker(s): Lazaros Mitskopoulos, Michael Camilleri
Chair/Host: Kia Nazarpour