The influence of Markov decision process structure on the possible strategic use of working memory and episodic memory

Learning to use past evidence in a sophisticated world model

Humans and other animals are able to discover underlying statistical structure in their environments and exploit it to achieve efficient and effective performance. However, such structure is often difficult to learn and use because it is obscure, involving long-range temporal dependencies. Here, we analysed behavioural data from an extended experiment with rats, showing that the subjects learned the underlying statistical structure, albeit suffering at times from immediate inferential imperfections as to their current state within it. We accounted for their behaviour using a Hidden Markov Model, in which recent observations are integrated with evidence from the past. We found that over the course of training, subjects came to track their progress through the task more accurately, a change that our model largely attributed to improved integration of past evidence. This learning reflected the structure of the task, decreasing reliance on recent observations, which were potentially misleading.

Humans and other animals possess the remarkable ability to find and exploit patterns and structures in their experience of a complex and varied world. However, such structures are often temporally extended and latent or hidden, being only partially correlated with immediate observations of the world. This makes it essential to integrate current and historical information, and creates a challenging statistical and computational problem. Here, we examine the behaviour of rats facing a version of this challenge posed by a brain-stimulation reward task. We find that subjects learned the general structure of the task, but struggled when immediate observations were misleading. We captured this behaviour with a model in which subjects integrated evidence from recent observations together with evidence from the past. The subjects’ performance improved markedly over successive sessions, allowing them to overcome misleading observations. According to our model, this was made possible by more effective usage of past evidence to better determine the true state of the world.

Neural circuits for learning context-dependent associations of stimuli

The use of reinforcement learning combined with neural networks provides a powerful framework for solving certain tasks in engineering and cognitive science. Previous research shows that neural networks have the power to automatically extract features and learn hierarchical decision rules. In this work, we investigate reinforcement learning methods for performing a context-dependent association task using two kinds of neural network models (using continuous firing rate neurons), as well as a neural circuit gating model. The task allows examination of the ability of different models to extract hierarchical decision rules and generalize beyond the examples presented to the models in the training phase. We find that the simple neural circuit gating model, trained using response-based regulation of Hebbian associations, performs almost at the same level as a reinforcement learning algorithm combined with neural networks trained with more sophisticated back-propagation of error methods. A potential explanation is that hierarchical reasoning is the key to performance and the specific learning method is less important.

Deep belief networks learn context dependent behavior

With the goal of understanding behavioral mechanisms of generalization, we analyzed the ability of neural networks to generalize across context. We modeled a behavioral task where the correct responses to a set of specific sensory stimuli varied systematically across different contexts. The correct response depended on the stimulus (A,B,C,D) and context quadrant (1,2,3,4). The possible 16 stimulus-context combinations were associated with one of two responses (X,Y), one of which was correct for half of the combinations. The correct responses varied symmetrically across contexts. This allowed responses to previously unseen stimuli (probe stimuli) to be generalized from stimuli that had been presented previously. By testing the simulation on two or more stimuli that the network had never seen in a particular context, we could test whether the correct response on the novel stimuli could be generated based on knowledge of the correct responses in other contexts. We tested this generalization capability with a Deep Belief Network (DBN), Multi-Layer Perceptron (MLP) network, and the combination of a DBN with a linear perceptron (LP). Overall, the combination of the DBN and LP had the highest success rate for generalization.

Cellular dynamical mechanisms for encoding the time and place of events along spatiotemporal trajectories in episodic memory

Understanding the mechanisms of episodic memory requires linking behavioral data and lesion effects to data on the dynamics of cellular membrane potentials and population interactions within brain regions. Linking behavior to specific membrane channels and neurochemicals has implications for therapeutic applications. Lesions of the hippocampus, entorhinal cortex and subcortical nuclei impair episodic memory function in humans and animals, and unit recording data from these regions in behaving animals indicate episodic memory processes. Intracellular recording in these regions demonstrates specific cellular properties including resonance, membrane potential oscillations and bistable persistent spiking that could underlie the encoding and retrieval of episodic trajectories. A model presented here shows how intrinsic dynamical properties of neurons could mediate the encoding of episodic memories as complex spatiotemporal trajectories. The dynamics of neurons allow encoding and retrieval of unique episodic trajectories in multiple continuous dimensions including temporal intervals, personal location, the spatial coordinates and sensory features of perceived objects and generated actions, and associations between these elements. The model also addresses how cellular dynamics could underlie unit firing data suggesting mechanisms for coding continuous dimensions of space, time, sensation and action.

A phase code for memory could arise from circuit mechanisms in entorhinal cortex

Neurophysiological data reveals intrinsic cellular properties that suggest how entorhinal cortical neurons could code memory by the phase of their firing. Potential cellular mechanisms for this phase coding in models of entorhinal function are reviewed. This mechanism for phase coding provides a substrate for modeling the responses of entorhinal grid cells, as well as the replay of neural spiking activity during waking and sleep. Efforts to implement these abstract models in more detailed biophysical compartmental simulations raise specific issues that could be addressed in larger scale population models incorporating mechanisms of inhibition.

A model of episodic memory: mental time travel along encoded trajectories using grid cells

The definition of episodic memory includes the concept of mental time travel: the ability to re-experience a previously experienced trajectory through continuous dimensions of space and time, and to recall specific events or stimuli along this trajectory. Lesions of the hippocampus and entorhinal cortex impair human episodic memory function and impair rat performance in tasks that could be solved by retrieval of trajectories. Recent physiological data suggests a novel model for encoding and retrieval of trajectories, and for associating specific stimuli with specific positions along the trajectory. During encoding in the model, external input drives the activity of head direction cells. Entorhinal grid cells integrate the head direction input to update an internal representation of location, and drive hippocampal place cells. Trajectories are encoded by Hebbian modification of excitatory synaptic connections between hippocampal place cells and head direction cells driven by external action. Associations are also formed between hippocampal cells and sensory stimuli. During retrieval, a sensory input cue activates hippocampal cells that drive head direction activity via previously modified synapses. Persistent spiking of head direction cells maintains the direction and speed of the action, updating the activity of entorhinal grid cells that thereby further update place cell activity. Additional cells, termed arc length cells, provide coding of trajectory segments based on the one-dimensional arc length from the context of prior actions or states, overcoming ambiguity where the overlap of trajectory segments causes multiple head directions to be associated with one place. These mechanisms allow retrieval of complex, self-crossing trajectories as continuous curves through space and time.