A Neurocomputational Model of Nicotine Addiction Based on Reinforcement Learning Selin Metin 1 and Neslihan Serap S ¸eng¨or 1 Istanbul Technical University, Electrical and Electronics Engineering Faculty, Maslak 34469, Istanbul, Turkey selinmetin@gmail.com, sengorn@itu.edu.tr Abstract. Continuous exposure to nicotine causes behavioral choice to be modiﬁed by dopamine to become rigid, resulting in addiction. In this work, a computational model for nicotine addiction is proposed and the proposed model captures the eﬀect of continuous nicotine exposure in be- coming addict through reinforcement learning. The computational model is composed of three subsystems each corresponding to neural substrates taking part in nicotine addiction and these subsystems are realized by nonlinear dynamical systems. Even though the model is suﬃcient in ac- quiring addiction, it needs to be further developed to give a better ex- planation for the process responsible in turning a random choice into a compulsive behavior. Keywords: computational model, dynamic system, nicotine addiction, reinforcement learning. 1 Introduction The value of an experience or an action is imposed by the reward gained after- wards. An action inducing a greater reward is sensed as a better action, and thus rewarding it is repeated frequently [1]. In the case of addiction, the abusive sub- stance (nicotine, drugs, etc.) has a greater value in the brain than other forms of reward imposing actions. It is believed that some persistent modiﬁcations in the synaptic plasticity is the cause of addiction, thus we can deﬁne addiction as a disorder in the mesolimbic system which modiﬁes responses of rewarding actions. Mislead by overemphasized reward sensations addicts compulsively seek the ob- ject of their addiction. As the reward mechanism has persistently changed, addicts are usually not completely cured and relapse into drug use after treatment [2]. The two main approaches in explaining addiction are the opponent process theory and reward related learning [3,4,5]. Using reinforcement learning theory, addiction is explained as the cumulative result obtained by the administration of a drug as a positive reinforcer [5,6,7]. The opponent-process theory of motiva- tion [3] is used to explain the conditioning principles leading to pleasurable and compulsive activity. According to this model, emotions are paired and when one emotion in a pair is experienced, the other is suppressed. In [8], these two ap- proaches are considered together in deriving a computational model for nicotine addiction. K. Diamantaras, W. Duch, L.S. Iliadis (Eds.): ICANN 2010, Part II, LNCS 6353, pp. 228–233, 2010. c  Springer-Verlag Berlin Heidelberg 2010