Harnessing the flexibility of neural networks to predict dynamic theoretical parameters underlying human choice behavior

Reinforcement learning (RL) models are used extensively to study human behavior. These rely on normative models of behavior and stress interpretability over predictive capabilities. More recently, neural network models have emerged as a descriptive modeling paradigm that is capable of high predictive power yet with limited interpretability. Here, we seek to augment the expressiveness of theoretical RL models with the high flexibility and predictive power of neural networks. We introduce a novel framework, which we term theoretical-RNN (t-RNN), whereby a recurrent neural network is trained to predict trial-by-trial behavior and to infer theoretical RL parameters using artificial data of RL agents performing a two-armed bandit task. In three studies, we then examined the use of our approach to dynamically predict unseen behavior along with time-varying theoretical RL parameters. We first validate our approach using synthetic data with known RL parameters. Next, as a proof-of-concept, we applied our framework to two independent datasets of humans performing the same task. In the first dataset, we describe differences in theoretical RL parameters dynamic among clinical psychiatric vs. healthy controls. In the second dataset, we show that the exploration strategies of humans varied dynamically in response to task phase and difficulty. For all analyses, we found better performance in the prediction of actions for t-RNN compared to the stationary maximum-likelihood RL method. We discuss the use of neural networks to facilitate the estimation of latent RL parameters underlying choice behavior. Currently, neural network models fitted directly to behavioral human data are thought to dramatically outperform theoretical computational models in terms of predictive accuracy. However, these networks do not provide a clear theoretical interpretation of the mechanisms underlying the observed behavior. Generating plausible theoretical explanations for observed human data is a major goal in computational neuroscience. Here, we provide a proof-of-concept for a novel method where a recurrent neural network (RNN) is trained on artificial data generated from a known theoretical model to predict both trial-by-trial actions and theoretical parameters. We then freeze the RNN weights and use it to predict both actions and theoretical parameters of empirical data. We first validate our approach using synthetic data where the theoretical parameters are known. We then show, using two empirical datasets, that our approach allows dynamic estimation of latent parameters while providing better action predictions compared to theoretical models fitted with a maximum-likelihood approach. This proof-of-concept suggests that neural networks can be trained to predict meaningful time-varying theoretical parameters.