Thomas V. Wiecki’s research while affiliated with Brown University and other places

Probabilistic Programming allows for automatic Bayesian inference on user-defined probabilistic models. Recent advances in Markov chain Monte Carlo (MCMC) sampling allow inference on increasingly complex models. This class of MCMC, known as Hamliltonian Monte Carlo, requires gradient information which is often not readily available. PyMC3 is a new open source Probabilistic Programming framework written in Python that uses Theano to compute gradients via automatic differentiation as well as compile probabilistic programs on-the-fly to C for increased speed. Contrary to other Probabilistic Programming languages, PyMC3 allows model specification directly in Python code. The lack of a domain specific language allows for great flexibility and direct interaction with the model. This paper is a tutorial-style introduction to this software package.

Probabilistic Programming allows for automatic Bayesian inference on user-defined probabilistic models. Recent advances in Markov chain Monte Carlo (MCMC) sampling allow inference on increasingly complex models. This class of MCMC, known as Hamliltonian Monte Carlo, requires gradient information which is often not readily available. PyMC3 is a new open source Probabilistic Programming framework written in Python that uses Theano to compute gradients via automatic differentiation as well as compile probabilistic programs on-the-fly to C for increased speed. Contrary to other Probabilistic Programming languages, PyMC3 allows model specification directly in Python code. The lack of a domain specific language allows for great flexibility and direct interaction with the model. This paper is a tutorial-style introduction to this software package.

Probabilistic programming (PP) allows flexible specification of Bayesian statistical models in code. PyMC3 is a new, open-source PP framework with an intutive and readable, yet powerful, syntax that is close to the natural syntax statisticians use to describe models. It features next-generation Markov chain Monte Carlo (MCMC) sampling algorithms such as the No-U-Turn Sampler (NUTS; Hoffman, 2014), a self-tuning variant of Hamiltonian Monte Carlo (HMC; Duane, 1987). Probabilistic programming in Python confers a number of advantages including multi-platform compatibility, an expressive yet clean and readable syntax, easy integration with other scientific libraries, and extensibility via C, C++, Fortran or Cython. These features make it relatively straightforward to write and use custom statistical distributions, samplers and transformation functions, as required by Bayesian analysis.

Psychiatric research is in crisis. We highlight efforts to overcome current challenges by focusing on the emerging field of computational psychiatry, which might enable the field to move from a symptom-based description of mental illness to descriptors based on objective computational multidimensional functional variables. We survey recent efforts toward this goal and describe a set of methods that together form a toolbox to aid this research program. We identify four levels in computational psychiatry: (a) behavioral tasks that index various psychological processes, (b) computational models that identify the generative psychological processes, (c) parameter-estimation methods concerned with quantitatively fitting these models to subject behavior by focusing on hierarchical Bayesian estimation as a rich framework with many desirable properties, and (d) machine-learning clustering methods that identify clinically significant conditions and subgroups of individuals. As a proof of principle, we apply these methods to two different data sets. Finally, we highlight challenges for future research.

Depression is characterized by poor executive function, but - counterintuitively - in some studies, it has been associated with highly accurate performance on certain cognitively demanding tasks. The psychological mechanisms responsible for this paradoxical finding are unclear. To address this issue, we applied a drift diffusion model (DDM) to flanker task data from depressed and healthy adults participating in the multi-site Establishing Moderators and Biosignatures of Antidepressant Response for Clinical Care for Depression (EMBARC) study. One hundred unmedicated, depressed adults and 40 healthy controls completed a flanker task. We investigated the effect of flanker interference on accuracy and response time, and used the DDM to examine group differences in three cognitive processes: prepotent response bias (tendency to respond to the distracting flankers), response inhibition (necessary to resist prepotency), and executive control (required for execution of correct response on incongruent trials). Consistent with prior reports, depressed participants responded more slowly and accurately than controls on incongruent trials. The DDM indicated that although executive control was sluggish in depressed participants, this was more than offset by decreased prepotent response bias. Among the depressed participants, anhedonia was negatively correlated with a parameter indexing the speed of executive control (r = -0.28, p = 0.007). Executive control was delayed in depression but this was counterbalanced by reduced prepotent response bias, demonstrating how participants with executive function deficits can nevertheless perform accurately in a cognitive control task. Drawing on data from neural network simulations, we speculate that these results may reflect tonically reduced striatal dopamine in depression.

The stop-signal paradigm is frequently used to study response inhibition. In this paradigm, participants perform a two-choice response time (RT) task where the primary task is occasionally interrupted by a stop-signal that prompts participants to withhold their response. The primary goal is to estimate the latency of the unobservable stop response (stop signal reaction time or SSRT). Recently, Matzke et al. (2013) have developed a Bayesian parametric approach (BPA) that allows for the estimation of the entire distribution of SSRTs. The BPA assumes that SSRTs are ex-Gaussian distributed and uses Markov chain Monte Carlo sampling to estimate the parameters of the SSRT distribution. Here we present an efficient and user-friendly software implementation of the BPA—BEESTS—that can be applied to individual as well as hierarchical stop-signal data. BEESTS comes with an easy-to-use graphical user interface and provides users with summary statistics of the posterior distribution of the parameters as well various diagnostic tools to assess the quality of the parameter estimates. The software is open source and runs on Windows and OS X operating systems. In sum, BEESTS allows experimental and clinical psychologists to estimate entire distributions of SSRTs and hence facilitates the more rigorous analysis of stop-signal data.

Current psychiatric research is in crisis. In this review I will describe the causes of this crisis and highlight recent efforts to overcome current challenges. One particularly promising approach is the emerging field of computational psychiatry. By using methods and insights from computational cognitive neuroscience, computational psychiatry might enable us to move from a symptom-based description of mental illness to descriptors based on objective computational multidimensional functional variables. To exemplify this I will survey recent efforts towards this goal. I will then describe a set of methods that together form a toolbox of cognitive models to aid this research program. At the core of this toolbox are sequential sampling models which have been used to explain diverse cognitive neuroscience phenomena but have so far seen little adoption in psychiatric research. I will then describe how these models can be fitted to subject data and highlight how hierarchical Bayesian estimation provides a rich framework with many desirable properties and benefits compared to traditional optimization-based approaches. Finally, non-parametric Bayesian methods provide general solutions to the problem of classifying mental illness within this framework.