The prefrontal cortex has long been thought to subserve both working memory (the holding of information online for processing) and executive functions (deciding how to manipulate working memory and perform processing). Although many computational models of working memory have been developed, the mechanistic basis of executive function remains elusive, often amounting to a homunculus. This article presents an attempt to deconstruct this homunculus through powerful learning mechanisms that allow a computational model of the prefrontal cortex to control both itself and other brain areas in a strategic, task-appropriate manner. These learning mechanisms are based on subcortical structures in the midbrain, basal ganglia, and amygdala, which together form an actor-critic architecture. The critic system learns which prefrontal representations are task relevant and trains the actor, which in turn provides a dynamic gating mechanism for controlling working memory updating. Computationally, the learning mechanism is designed to simultaneously solve the temporal and structural credit assignment problems. The model’s performance compares favorably with standard backpropagation-based temporal learning mechanisms on the challenging 1-2-AX working memory task and other benchmark working memory tasks.

The frontal cortex and basal ganglia interact via a relatively well-understood and elaborate system of interconnections. In the context of motor function, these interconnections can be understood as disinhibiting or "releasing the brakes" on frontal motor action plans --- the basal ganglia detect appropriate contexts for performing motor actions, and enable the frontal cortex to execute such actions at the appropriate time. We build on this idea in the domain of working memory through the use of computational neural network models of this circuit. In our model, the frontal cortex exhibits robust active maintenance, while the basal ganglia contribute a selective, dynamic gating function that enables frontal memory representations to be rapidly updated in a task-relevant manner. We apply the model to a novel version of the continuous performance task (CPT) that requires subroutine-like selective working memory updating, and compare and contrast our model with other existing models and th...

Long Short-Term Memory (LSTM, Hochreiter & Schmidhuber, 1997) can solve numerous tasks not solvable by previous learning algorithms for recurrent neural networks (RNNs). We identify a weakness of LSTM networks processing continual input streams that are not a priori segmented into subsequences with explicitly marked ends at which the network's internal state could be reset. Without resets, the state may grow indenitely and eventually cause the network to break down. Our remedy is a novel, adaptive \forget gate" that enables an LSTM cell to learn to reset itself at appropriate times, thus releasing internal resources. We review illustrative benchmark problems on which standard LSTM outperforms other RNN algorithms. All algorithms (including LSTM) fail to solve continual versions of these problems. LSTM with forget gates, however, easily solves them in an elegant way.

Over the decades, computational models of human cognition have advanced from programs that produce output similar to that of human problem solvers to systems that mimic both the products and processes of human performance. In this paper, we describe a model that achieves the next step in this progression: predicting individual participants' performance across multiple tasks after estimating a single individual difference parameter. We demonstrate this capability in the context of a model of working memory, where the individual difference parameter for each participant represents working memory capacity. Specifically, our model is able to make zero-parameter predictions of individual participants' performance on a second task after separately fitting performance on a preliminary task. We argue that this level of predictive ability offers an important test of the theory underlying our model.

Abstract not found

dynamic task switching

Abstract — Why is it that people cannot keep their hands still when they talk? One reason may be that gesturing actually lightens cognitive load while a person is thinking of what to say. We asked adults and children to remember a list of letters or words while explaining how they solved a math problem. Both groups remembered significantly more items when they gestured during their math explanations than when they did not gesture. Gesturing appeared to save the speakers’ cognitive resources on the explanation task, permitting the speakers to allocate more resources to the memory task. It is widely accepted that gesturing reflects a speaker’s cognitive state, but our observations suggest that, by reducing cognitive load, gesturing may also play a role in shaping that state. Gesturing occurs across ages, tasks, and cultures (Feyereisen & de Lannoy, 1991). Although in theory gesture could be nothing more than meaningless hand waving, recent research has found that gesturing conveys meaningful information (Clark, 1996; Goldin-Meadow, Mc-Neill, & Singleton, 1996; Kendon, 1980; McNeill, 1992), information that is not always found in the speech it accompanies (Goldin-Meadow, Alibali, & Church, 1993). For example, a speaker might say, “I ran all the way upstairs ” while moving her index finger upward in a spiral. It is through the speaker’s gestures, and only her gestures, that the listener knows the staircase is a spiral. Moreover, gesture is noticed. The information that gesture conveys frequently has an impact on the message listeners take from the communication (Alibali, Flevares,

This study tested the hypothesis that dorsolateral prefrontal cortex deficits contribute to both working memory and long-term memory disturbances in schizophrenia. It also examined whether such deficits were more severe for verbal than nonverbal stimuli. Functional magnetic resonance imaging was used to assess cortical activation during performance of verbal and nonverbal versions of a working memory task and both encoding and recognition tasks in 38 individuals with schizophrenia and 48 healthy controls. Performance of both working memory and long-term memory tasks revealed disturbed dorsolateral prefrontal cortex activation in schizophrenia, although medial temporal deficits were also present. Some evidence was found for more severe cognitive and functional deficits with verbal than nonverbal stimuli, although these results were mixed. A large literature on cognitive function in schizophrenia suggests that individuals with this illness display deficits in several different cognitive domains, including deficits in working memory (WM) and long-term memory (LTM). Deficits in WM have often been hypothesized to reflect a disturbance in prefrontal cortex (PFC) function (D’Esposito et al., 1998), whereas deficits in LTM

We present a neural network model that shows how the prefrontal cortex, interacting with the basal ganglia, can maintain a sequence of phonological information in activation-based working memory (i.e., the phonological loop). The primary function of this phonological loop may be to transiently encode arbitrary bindings of information necessary for tasks | the combinatorial expressive power of language enables very exible binding of essentially arbitrary pieces of information. Our model takes advantage of the closed-class nature of phonemes, which allows dierent neural representations of all possible phonemes at each sequential position to be encoded. To make this work, we suggest that the basal ganglia provide a region-speci c update signal that allocates phonemes to the appropriate sequential coding slot. To demonstrate that exible, arbitrary binding of novel sequences can be supported by this mechanism, we show that the model can generalize to novel sequences after moderate amounts of training.

This work focuses on a particular neurocomputational account of working memory function that has been used to explain a wide range of working memory phenomena in terms of interactions between the prefrontal cortex and the mesolimbic dopamine system. Using the mechanisms prescribed by this theory, we have constructed a software toolkit for creating working memory modules for use in robotic control systems. The challenges faced by embodied robots are similar to those experienced by humans in everyday living, making this domain useful for testing the utility and scalability of this computational theory of working memory. We report the results of a feasibility study, involving a robotic version of the delayed saccade task, and we discuss future plans to test our working memory model in the context of robot control and learning.