Contributed Talk Sessions | Poster Sessions | All Posters | Search Papers

Poster Session A: Tuesday, August 12, 1:30 – 4:30 pm, de Brug & E‑Hall

Comparing different criteria for neural dimensionality estimation

Francesco Edoardo Vaccari¹, Stefano Diomedi², Edoardo Bettazzi, Patrizia Fattori; ¹University of Bologna, ²National Research Council

Presenter: Francesco Edoardo Vaccari

Despite dimensionality reduction is essential in modern Neuroscience and Principal Component Analysis (PCA) continues to serve as the standard approach, in the field it is still missing a widely accepted criterion for choosing the number of components to retain. To fill this gap, we aimed to compare the performance of different retention criteria. We designed a data simulation procedure to generate data matrices with a ground-truth latent structure. Simulation parameters were varied to compare the different retention criteria in several scenarios. Among the tested criteria, Parallel Analysis and a cross-validation scheme, specifically conceived for dimensionality reduction, resulted to be the most effective methods. Finally, by applying these criteria to real spiking activity, we show that different criteria can lead to significantly different results in the estimation of dimensionality and noise. Our study highlights the need for an explicit definition of “dimensionality” in the analysis of population spike activity and a consequent careful choice of the retention criterion to be used, as this can lead to important biases and non-comparable results between studies.

Topic Area: Methods & Computational Tools

Extended Abstract: Full Text PDF