TY - JOUR T1 - Recurrent Auto-Encoding Drift Diffusion Model JF - bioRxiv DO - 10.1101/220517 SP - 220517 AU - Moens Vincent AU - Zenon Alexandre Y1 - 2018/01/01 UR - http://biorxiv.org/content/early/2018/05/13/220517.abstract N2 - The Drift Diffusion Model (DDM) is a popular model of behaviour that accounts for patterns of accuracy and reaction time data. In the Full DDM implementation, parameters are allowed to vary from trial-to-trial, making the model more powerful but also more challenging to fit to behavioural data. Current approaches yield typically poor fitting quality, are computationally expensive and usually require assuming constant threshold parameter across trials. Moreover, in most versions of the DDM, the sequence of participants’ choices is considered independent and identically distributed (i.i.d.), a condition often violated in real data.Our contribution to the field is threefold: first, we introduce Variational Bayes as a method to fit the full DDM. Second, we relax the i.i.d. assumption, and propose a data-driven algorithm based on a Recurrent Auto-Encoder (RAE-DDM), that estimates the local posterior probability of the DDM parameters at each trial based on the sequence of parameters and data preceding the current data point. Finally, we extend this algorithm to illustrate that the RAE-DDM provides an accurate modelling framework for regression analysis. An important result of the approach we propose is that inference at the trial level can be achieved efficiently for each and every parameter of the DDM, threshold included. This data-driven approach is highly generic and self-contained, in the sense that no external input (e.g. regressors or physiological measure) is necessary to fit the data. Using simulations, we show that this method outperforms i.i.d.-based approaches (either Markov Chain Monte Carlo or i.i.d.-VB) without making any assumption about the nature of the between-trial correlation of the parameters. ER -