The decision-making process and how the brain predicts reward outcomes as part of this process is a popular topic of study in neuroscience. In fact, the study of how this system functions and the brain structures involved could provide insight into how addiction arises due to malfunctions in this system.

At the Canadian Centre for Behavioural Neuroscience, Master of Science candidate Clifford Donovan aims to solve this puzzle using a fascinating method of data analysis. With a passion for computers and a background in psychology and neuroscience, Clifford is utilizing artificial neural networks and deep learning to process huge data sets of neural recordings. Deep learning is what allows organisms, or computer programs, to categorize things based on their experiences and make predictions based on that learning. For instance, understanding the visual differences between a cat and a dog by being exposed to multiple examples of each, and being able to correctly categorize new examples of each. This simple example, however, only scratches the surface of the potential uses of deep learning and artificial neural networks.

Clifford utilizes these deep learning systems to analyze large amounts of collected data from neural recordings in rats. The recordings were taken from seven brain structures involved in the reward system of the brain to study how this system guides motor actions in a 50/50 chance reward task. By developing an artificial neural network capable of classifying and categorizing the outcomes of hundreds of trials, Clifford’s neural network is able to identify patterns in the brain activity data to predict if the animal experienced a win or a loss in the task.

The predictions from this artificial neural network provide insight into the brain structures involved in producing the reward and omission of reward experience. This line of research into the reward system’s function could help a multitude of disorders including depression, anorexia, and addiction by understanding each brain structures’ function in the system.

Clifford goes on to say that the potential for deep learning and neural networks goes beyond the scope of this project and, with enough data available, these methods could be used to find patterns in any data-set that traditional inferential statistics or regressions/models may miss.