Attention and working memory: a dynamical model of neuronal activity in the prefrontal cortex Gustavo Deco 1 and Edmund T. Rolls 2 1 Institucio  Catalana de Recerca i Estudis Avanc Ëats (ICREA), Universitat Pompeu Fabra, Department of Technology, Computational Neuroscience, Passeig de Circumval.lacio Â, 08003 Barcelona, Spain 2 University of Oxford, Department of Experimental Psychology, South Parks Road, Oxford OX1 3UD, UK Keywords: attractor network, autoassociation network, dopamine, executive function, prefrontal cortex, short-term memory, task switching Abstract Cognitive behaviour requires complex context-dependent mapping between sensory stimuli and actions. The same stimulus can lead to different behaviours depending on the situation, or the same behaviour may be elicited by different cueing stimuli. Neurons in the primate prefrontal cortex show task-speci®c ®ring activity during working memory delay periods. These neurons provide a neural substrate for mapping stimulus and response in a ¯exible, context- or rule-dependent, fashion. We describe here an integrate-and-®re networkmodeltoexplainandinvestigatethedifferenttypesofworking-memory-relatedneuronalactivityobserved.Themodelcontains differentpopulations(orpools)ofneurons(asfoundneurophysiologically)inattractornetworkswhichrespondinthedelayperiodtothe stimulus object, the stimulus position (`sensory pools'), to combinations of the stimulus sensory properties (e.g. the object identity or object location) and the response (`intermediate pools'), and to the response required (left or right) (`premotor pools'). The pools are arranged hierarchically, are linked by associative synaptic connections, and have global inhibition through inhibitory interneurons to implement competition. It is shown that a biasing attentional input to de®ne the current rule applied to the intermediate pools enables the system to select the correct response in what is a biased competition model of attention. The integrate-and-®re model not only produces realistic spiking dynamicals very similar to the neuronal data but also shows how dopamine could weaken and shorten the persistentneuronalactivityinthedelayperiod;andallowsustopredictmoreresponseerrorswhendopamineiselevatedbecausethere is less different activity in the different pools of competing neurons, resulting in more con¯ict. Introduction There is much evidence that the prefrontal cortex (PFC) is involved in at least some types of working memory and related processes (Jacobsen, 1935; Goldman-Rakic, 1995; Goldman-Rakic, 1996; Fuster, 2000). Working memory refers to a system for maintaining and manipulating information in mind, held during a short period, usually of seconds (Baddeley, 1986). Neuronal recording studies indicate continuing ®ring of prefrontal neurons during the delay period of working memory tasks (Fuster & Alexander, 1971; Kubota & Niki, 1971; Funahashi et al., 1989; Funahashi et al., 1993), and imaging studies have con®rmed activation of the PFC (Leung et al., 2002; Ungerleider et al., 1998; Adcock et al., 2000; Zarahn et al., 2000). Prefrontal lesions in humans (Milner, 1963; Goldman-Rakic et al., 1987)andmonkeys(Butters&Pandya,1969;Levy&Goldman-Rakic, 1999) produce severe de®cits in tasks requiring short-term memory processing. The memory-related persistent prefrontal neuronal activity during the delay period of short-term memory tasks could be maintained by assuming recurrent collateral excitatory loops (Hebb, 1949; Goldman- Rakic, 1995) which can be formally modelled and analysed by autoassociation networks. These networks store a set of memory patterns in the recurrent synaptic connections between the excitatory neurons (pyramidal cells) in the network, and when triggered with any one of the memory cue patterns, maintain that pattern of neuronal ®ring even when the cue is removed in a stable `attractor' state (see Amit, 1995; Rolls & Treves, 1998; Rolls & Deco, 2002). The neurophysiological investigations of the functions of the PFC in working memory have been extended recently by analysing neuronal activity when the monkey performs two different working memory tasks using the same stimuli and responses (Asaad et al., 1998; Hoshi et al., 1998; White & Wise, 1999; Asaad et al., 2000). The aim of the present work is to model, and therefore help to understand, the underlying mechanisms which implement the working-memory- related activity observed in PFC neurons in the context-dependent stimulus±response (associative) and delayed spatial response tasks investigated by Asaad etal. (2000) and Asaad etal. (1998). The model builds on the integrate-and-®re attractor network treatment of Brunel & Wang (2001) (which was discussed in the context of a simple object working memory with no arbitrary stimulus to response mapping, reversal, or rule change from associative to delayed spatial response). The new model introduced in this paper greatly extends this approach by introducing a hierarchically organised set of different attractor networks each with a different population of neurons (corresponding to the different types of neuron recorded during the performance of these tasks), and by introducing biasing inputs to the intermediate layer European Journal of Neuroscience, Vol. 18, pp. 2374±2390, 2003 ß Federation of European Neuroscience Societies doi:10.1046/j.1460-9568.2003.02956.x Correspondence: Professor Edmund T Rolls, as above. E-mail: Edmund.Rolls@psy.ox.ac.uk Web: www.cn5.ox.ac.uk Received 28 January 2003, revised 25 July 2003, accepted 29 July 2003