Hello. My name is Patryk, and I am a computational cognitive neuroscientist at The Johns Hopkins University. My PhD is in Neuroscience, from the Centers for Neuroscience and for the Neural Basis of Cognition at the University of Pittsburgh (CNUP and CNBC).

I study how neural systems learn to coordinate and control themselves.

To carry out my research, I use computational modeling (e.g., neural networks and Reinforcement Learning agent simulations) and test predictions from my models using behavioral techniques (e.g., reaction time and eye movement tracking) and brain neuroimaging techniques (e.g., fMRI).

On this site you can find information about me and my research. You can also download my CV in PDF form by clicking here.

What brain circuits are involved in learning about actions from reward? Many neuroscientists have proposed that the basal ganglia (a set of brain regions below the cortex) implement something like Reinforcement Learning, which is a computational model of learning. But we don't yet know if just some parts of the basal ganglia are involved in this kind of learning but not others (i.e., ventral versus dorsal parts). Some parts of the basal ganglia seem to be involved in learning from reinforcement but others seem to be involved in learning and maintaining habits. To gain insight into how this system contributes to learning, I designed a study to find out whether reinforcing different kind of actions used the same part of the basal ganglia or whether it used different parts. (We also tested to see if the parts that were used made sense given the connections we already know about.) My study showed that different parts of the basal ganglia are involved in reinforcing different actions (see the figure). Specifically, reward-prediction errors predicted by Reinforcement Learning were found in the putamen, dorsal caudate, and anterior caudate when right-hand movements, eye movements, or covert attention shifts were reinforced using money, respectively. For example, when participants in my study performed a hand movement and received more reward than they expected, the area with the red cubes was strongly activated. (The study is described in more detail in Laurent & Reichle, submitted.)

The results suggest that distinct parts of the striatum are involved in computing and/or storing reward-prediction errors for specific kinds of actions, including cognitive (non-motor) actions like covert attention shifts.

Here is some software I've developed during my research. You are free to use any of this source code in your work as long as you acknowledge me. I hope to get around to selecting a more specific open source license soon.

• fMRI

SVM Weight Extractor

Handy script that extracts weights from the SVM generated by MATLAB's bioinformatics toolbox so you can look at the value of your decision variables.

ASL Eye Tracker Driver

This Java code allows you to get gaze information from ASL eye trackers via a serial port connection for gaze-contingent paradigms. It interprets the binary data from the controller.

EPI Noise Echo Cancellation

If you need to hear your participant in real time during your fMRI experiment, this ChucK script allows you to cancel out the scanner noise during EPI data acquisition.

• Computational Modeling

Adaptive Reading Agent Simulator

(Runs in a GUI) Train a Reinforcement Learning agent to control its saccadic eye movements as it reads words of various lengths and word identification difficulties. Outputs data in MATLAB matrix format for easy analysis.

Recurrent McCulloch-Pitts Network

Train and test a Levy-style recurrent neural network of spiking McCulloch-Pitts neurons.

NNQL

A project to develop a simple but powerful neural network simulation language inspired by SQL: NNQL aims to (1) minimize programming time, (2) maximize clarity of model descriptions, and thereby (3) improve communication.