We propose a biologically constrained, large-scale neural network model that solves the Wisconsin Card Sorting Test (WCST). The WCST has been widely used in clinical and research settings to study cognitive flexibility and executive function. The model shows a good quantitative match with human responses across a number of WCST scoring indices, while consisting of neural networks that functionally and anatomically map to brain areas and structures implicated in the task, such as the prefrontal cortex and the cortico-basal ganglia-thalamus-cortical loop. We argue that the model provides a mechanistic account of WCST solving, and demonstrate its robustness by examining its performance across a range of biologically motivated parameter values.

According to mental model theory, spatial reasoning is basedon the construction and variation of mental modelsrepresenting spatial arrangements. Several effects in humanspatial reasoning are known to support this theory, forexample the ordering effect. Yet, reasoning effects have beenobserved for which the cognitive mechanisms are not entirelyexplained. To investigate how these effects can be attributedto neural computation, we modeled spatial reasoning in theNeural Engineering Framework.We selected three experiments to simulate tasks in a cognitivemodel based on an internal display. In our model,performance declines with an increase of objects which isexplained by the neural drift over time. We replicated effectsfrom the studies which we have found to be due to continuouspremise integration. By modeling and simulating spatialreasoning tasks, we showed that effects reported inpsychological studies can be explained by the emergentproperties of neural computation.

Biological brains learn much more quickly than standard deep neural network reinforcement learning algorithms. One reason for this is that the deep neural networks need to learn a representation that is appropriate for the task at hand, whilst biological systems already possess an appropriate representation. Here, we bypass this problem by imposing on the neural network a representation based on what is observed in biology, such as grid cells. This study explores the impact of using a biologically-inspired grid-cell representation vs. a one-hot representation, on the speed at which a Temporal Difference-based Actor-Critic network learns to solve a simple 2D grid-world reinforcement learning task. The results suggest that the use of grid cells does promote faster learning. Furthermore, the grid cells implemented here have the potential for accurately representing unbounded continuous space. Thus, their promising performance on this discrete task acts as a first step in exploring their utility for reinforcement learning in continuous space.

Social environments often impose tradeoffs between pursuing personal goals and maintaining a favorable reputation. We studied how individuals navigate these tradeoffs using Reinforcement Learning (RL), paying particular attention to the role of social value orientation (SVO). We had human participants play an interated Trust Game against various software opponents and analyzed the behaviors. We then incorporated RL into two cognitive models, trained these RL agents against the same software opponents, and performed similar analyses. Our results show that the RL agents reproduce many interesting features in the human data, such as the dynamics of convergence during learning and the tendency to defect once reciprocation becomes impossible. We also endowed some of our agents with SVO by incorporating terms for altruism and inequality aversion into their reward functions. These prosocial agents differed from proself agents in ways that resembled the differences between prosocial and proself participants. This suggests that RL is a useful framework for understanding how people use feedback to make social decisions.

We have previously shown that a biologically realistic spikingneuron implementation of an action selection/executionsystem (constrained by the neurological connectivity of thecortex, basal ganglia, and thalamus) is capable of performingcomplex tasks, such as the Tower of Hanoi, n-Back, andsemantic memory search. However, because the neuralimplementation approximates a strict rule-based structure of aproduction system, such models have involved hand-tweakingof multiple parameters to get the desired behaviour. Here, weshow that a simple, local, online learning rule can be used tolearn these parameters, resulting in neural models of cognitivebehaviours that are more reliable and easier to construct thanwith prior methods.

The cerebellum is classically described in terms of its role inmotor control. Recent evidence suggests that the cerebellumsupports a wide variety of functions, including timing-relatedcognitive tasks and perceptual prediction. Correspondingly,deciphering cerebellar function may be important to advanceour understanding of cognitive processes. In this paper, webuild a model of eyeblink conditioning, an extensively studiedlow-level function of the cerebellum. Building such a modelis of particular interest, since, as of now, it remains unclearhow exactly the cerebellum manages to learn and reproducethe precise timings observed in eyeblink conditioning that arepotentially exploited by cognitive processes as well. We em-ploy recent advances in large-scale neural network modelingto build a biologically plausible spiking neural network basedon the cerebellar microcircuitry. We compare our simulationresults to neurophysiological data and demonstrate how therecurrent Granule-Golgi subnetwork could generate the dynam-ics representations required for triggering motor trajectoriesin the Purkinje cell layer. Our model is capable of reproduc-ing key properties of eyeblink conditioning, while generatingneurophysiological data that could be experimentally verified.

Existing models of the basal ganglia assume the existence of separate channels of neuron populations for representing each available action. This type of localist mapping limits models to small, discrete action spaces, since additional actions require additional channels, costing neural resources and imposing new connective tracts. In contrast, evidence suggests that the basal ganglia plays a role in the selection of both discrete action units, and continuously-valued action kinematics. In this work, we model the basal ganglia with distributed action representations, using high-dimensional vectors. This method lends itself to representing both discrete and continuous action spaces. Vectors that represent actions are weighted by a scalar value (their salience to the current task), and bundled together to form a single input vector. This paper provides an overview of the encoding method and network structure, as well as a demonstration of the model solving an action selection task using spiking neurons.