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

Poster Session C: Friday, August 15, 2:00 – 5:00 pm, de Brug & E‑Hall

Uncovering the Structure of Trial-to-Trial Variability in Perceptual Decision-Making Using Disentangled Recurrent Neural Networks

Isabelle Hoxha¹, Anne Urai¹; ¹Leiden University

Presenter: Anne Urai

Perceptual decision-making shows considerable trial-to-trial variability, which can be captured by latent variable models of fluctuating decision strategies. Recently, disentangled Recurrent Neural Networks (DisRNNs) have achieved data-driven discovery of such trial-to-trial latent decision strategies in multi-armed bandit tasks. In this work, we investigate the applicability of DisRNNs for uncovering trial-to-trial structure of perceptual decision-making data. We fit DisRNNs on simulated Diffusion Decision Model (DDM) data, where the starting point or drift parameters depend on past choices. We show that the trace of the starting point and drift can be recovered in the latent variables, and that the shape of the trial-to-trial dependency of these parameters can be interpreted from the update rules learned by the network. This sets the stage for data-driven discovery of the sources of across-trial variability in real perceptual data.

Topic Area: Reward, Value & Social Decision Making

Extended Abstract: Full Text PDF