Biomedical subjects

Kenji Doya

Publications and source records attributed to Kenji Doya.

17 recordsLinked to original sources

PubMed2006

Nitric oxide regulates input specificity of long-term depression and context dependence of cerebellar learning.

Recent studies have shown that multiple internal models are acquired in the cerebellum and that these can be switched under a given context of behavior. It has been proposed that long-term depression (LTD) of parallel fiber (PF)-Purkinje cell (PC) synapses forms the cellular basis of cerebellar learning, and that the presynaptically synthesized messenger nitric oxide (NO) is a crucial "gatekeeper" for LTD. Because NO diffuses freely to neighboring synapses, this volume learning is not input-specific and brings into question the biological significance of LTD as the basic mechanism for efficient supervised learning. To better characterize the role of NO in cerebellar learning, we simulated the sequence of electrophysiological and biochemical events in PF-PC LTD by combining established simulation models of the electrophysiology, calcium dynamics, and signaling pathways of the PC. The results demonstrate that the local NO concentration is critical for induction of LTD and for its input specificity. Pre- and postsynaptic coincident firing is not sufficient for a PF-PC synapse to undergo LTD, and LTD is induced only when a sufficient amount of NO is provided by activation of the surrounding PFs. On the other hand, above-adequate levels of activity in nearby PFs cause accumulation of NO, which also allows LTD in neighboring synapses that were not directly stimulated, ruining input specificity. These findings lead us to propose the hypothesis that NO represents the relevance of a given context and enables context-dependent selection of internal models to be updated. We also predict sparse PF activity in vivo because, otherwise, input specificity would be lost.

Action Potentials↗

PubMed2006

Brain mechanism of reward prediction under predictable and unpredictable environmental dynamics.

In learning goal-directed behaviors, an agent has to consider not only the reward given at each state but also the consequences of dynamic state transitions associated with action selection. To understand brain mechanisms for action learning under predictable and unpredictable environmental dynamics, we measured brain activities by functional magnetic resonance imaging (fMRI) during a Markov decision task with predictable and unpredictable state transitions. Whereas the striatum and orbitofrontal cortex (OFC) were significantly activated both under predictable and unpredictable state transition rules, the dorsolateral prefrontal cortex (DLPFC) was more strongly activated under predictable than under unpredictable state transition rules. We then modelled subjects' choice behaviours using a reinforcement learning model and a Bayesian estimation framework and found that the subjects took larger temporal discount factors under predictable state transition rules. Model-based analysis of fMRI data revealed different engagement of striatum in reward prediction under different state transition dynamics. The ventral striatum was involved in reward prediction under both unpredictable and predictable state transition rules, although the dorsal striatum was dominantly involved in reward prediction under predictable rules. These results suggest different learning systems in the cortico-striatum loops depending on the dynamics of the environment: the OFC-ventral striatum loop is involved in action learning based on the present state, while the DLPFC-dorsal striatum loop is involved in action learning based on predictable future states.

Brain↗

PubMed2006

The computational neurobiology of learning and reward.

Following the suggestion that midbrain dopaminergic neurons encode a signal, known as a 'reward prediction error', used by artificial intelligence algorithms for learning to choose advantageous actions, the study of the neural substrates for reward-based learning has been strongly influenced by computational theories. In recent work, such theories have been increasingly integrated into experimental design and analysis. Such hybrid approaches have offered detailed new insights into the function of a number of brain areas, especially the cortex and basal ganglia. In part this is because these approaches enable the study of neural correlates of subjective factors (such as a participant's beliefs about the reward to be received for performing some action) that the computational theories purport to quantify.

Animals↗

PubMed2005

Anterior and superior lateral occipito-temporal cortex responsible for target motion prediction during overt and covert visual pursuit.

In smooth-pursuit eye movements (SPEM) with gain close to one, SPEM should be controlled mainly by prediction of target motion because retinal slip is nearly zero. We investigated the neural mechanisms of visual-target prediction by the three fMRI experiments. (1) Overt pursuit task: subjects pursued a sinusoidally moving target which blinked (blink condition) or did not blink (continuous condition). (2) Covert pursuit task: subjects covertly pursued the same target with eyes gazed at fixation point. (3) Attend-to-stationary target task: subjects brought attention on a stationary target with eyes gazed at fixation point. In the overt pursuit task, the SPEM gain and the delay in the blink condition were not very different from the continuous condition, indicating good prediction of the blinking target motion. Activities in the dorsolateral prefrontal, precentral, medial superior frontal, intraparietal, and lateral occipito-temporal cortexes increased in the blink-continuous subtraction. The V1 activity decreased for this contrast. In the covert pursuit task, only the anterior/superior LOTC activity remained in the blink-continuous subtraction. In the attend-to-stationary target task, the blink-continuous subtraction elicited no activation. Consequently, the a/sLOTC activity is responsible for target prediction rather than motor commands for eye movements or just target blinking such as visual saliency.

Adult↗

PubMed2005

Representation of action-specific reward values in the striatum.

The estimation of the reward an action will yield is critical in decision-making. To elucidate the role of the basal ganglia in this process, we recorded striatal neurons of monkeys who chose between left and right handle turns, based on the estimated reward probabilities of the actions. During a delay period before the choices, the activity of more than one-third of striatal projection neurons was selective to the values of one of the two actions. Fewer neurons were tuned to relative values or action choice. These results suggest representation of action values in the striatum, which can guide action selection in the basal ganglia circuit.

Action Potentials↗

PubMed2005

Robust reinforcement learning.

This letter proposes a new reinforcement learning (RL) paradigm that explicitly takes into account input disturbance as well as modeling errors. The use of environmental models in RL is quite popular for both offline learning using simulations and for online action planning. However, the difference between the model and the real environment can lead to unpredictable, and often unwanted, results. Based on the theory of H(infinity) control, we consider a differential game in which a "disturbing" agent tries to make the worst possible disturbance while a "control" agent tries to make the best control input. The problem is formulated as finding a min-max solution of a value function that takes into account the amount of the reward and the norm of the disturbance. We derive online learning algorithms for estimating the value function and for calculating the worst disturbance and the best control in reference to the value function. We tested the paradigm, which we call robust reinforcement learning (RRL), on the control task of an inverted pendulum. In the linear domain, the policy and the value function learned by online algorithms coincided with those derived analytically by the linear H(infinity) control theory. For a fully nonlinear swing-up task, RRL achieved robust performance with changes in the pendulum weight and friction, while a standard reinforcement learning algorithm could not deal with these changes. We also applied RRL to the cart-pole swing-up task, and a robust swing-up policy was acquired.

Learning↗

PubMed2004

Prediction of immediate and future rewards differentially recruits cortico-basal ganglia loops.

Evaluation of both immediate and future outcomes of one's actions is a critical requirement for intelligent behavior. Using functional magnetic resonance imaging (fMRI), we investigated brain mechanisms for reward prediction at different time scales in a Markov decision task. When human subjects learned actions on the basis of immediate rewards, significant activity was seen in the lateral orbitofrontal cortex and the striatum. When subjects learned to act in order to obtain large future rewards while incurring small immediate losses, the dorsolateral prefrontal cortex, inferior parietal cortex, dorsal raphe nucleus and cerebellum were also activated. Computational model-based regression analysis using the predicted future rewards and prediction errors estimated from subjects' performance data revealed graded maps of time scale within the insula and the striatum: ventroanterior regions were involved in predicting immediate rewards and dorsoposterior regions were involved in predicting future rewards. These results suggest differential involvement of the cortico-basal ganglia loops in reward prediction at different time scales.

Adult↗

PubMed2004

Chaos may enhance information transmission in the inferior olive.

Despite unique well characterized neuronal properties, such as extensive electrical coupling and low firing rates, the role of the inferior olive (IO), which is the source of the climbing fiber inputs to cerebellar Purkinje cells, is still controversial. We propose that the IO stochastically recodes the high-frequency information carried by its synaptic inputs into stochastic, low-rate spikes in its climbing fiber output. Computer simulations of realistic IO networks showed that moderate electrical coupling produced chaotic firing, which maximized the input-output mutual information. This "chaotic resonance" may allow rich error signals to reach individual Purkinje cells, even at low firing rates, allowing efficient cerebellar learning.

Animals↗

PubMed2004

A neural correlate of reward-based behavioral learning in caudate nucleus: a functional magnetic resonance imaging study of a stochastic decision task.

Humans can acquire appropriate behaviors that maximize rewards on a trial-and-error basis. Recent electrophysiological and imaging studies have demonstrated that neural activity in the midbrain and ventral striatum encodes the error of reward prediction. However, it is yet to be examined whether the striatum is the main locus of reward-based behavioral learning. To address this, we conducted functional magnetic resonance imaging (fMRI) of a stochastic decision task involving monetary rewards, in which subjects had to learn behaviors involving different task difficulties that were controlled by probability. We performed a correlation analysis of fMRI data by using the explanatory variables derived from subject behaviors. We found that activity in the caudate nucleus was correlated with short-term reward and, furthermore, paralleled the magnitude of a subject's behavioral change during learning. In addition, we confirmed that this parallelism between learning and activity in the caudate nucleus is robustly maintained even when we vary task difficulty by controlling the probability. These findings suggest that the caudate nucleus is one of the main loci for reward-based behavioral learning.

Adult↗

PubMed2004

Cerebellar aminergic neuromodulation: towards a functional understanding.

Although a number of neuromodulators influence the cerebellar circuitry, their functions remain largely unknown. By reviewing and combining results from data-driven and theory-driven studies, we attempt to provide an integrated systems view of cerebellar neuromodulation. First, we review the short- and long-term effects of neuromodulators on the cerebellar circuitry. Second, we review recent theories of the cerebellum and show that a number of modulatory signals are needed for powerful cerebellar learning and control. Finally, we attempt to match each theoretically derived modulatory signal with a specific neuromodulator. In particular, we propose that serotonin controls the 'responsibility' of each cerebellar unit (or microcomplex) in cerebellar learning and control; norepinephrine gates unsupervised learning in the cerebellar cortex; dopamine enhances goal-oriented cerebellar learning; and, finally, acetylcholine controls the speed of supervised learning in Purkinje cells.

Animals↗

PubMed2004

Reinforcement learning with via-point representation.

In this paper, we propose a new learning framework for motor control. This framework consists of two components: reinforcement learning and via-point representation. In the field of motor control, conventional reinforcement learning has been used to acquire control sequences such as cart-pole or stand-up robot control. Recently, researchers have become interested in hierarchical architecture, such as multiple levels, and multiple temporal and spatial scales. Our new framework contains two levels of hierarchical architecture. The higher level is implemented using via-point representation, which corresponds to macro-actions or multiple time scales. The lower level is implemented using a trajectory generator that produces primitive actions. Our framework can modify the ongoing movement by means of temporally localized via-points and trajectory generation. Successful results are obtained in computer simulation of the cart-pole swing up task.

Computer Simulation↗

PubMed2004

Hierarchical Bayesian estimation for MEG inverse problem.

Source current estimation from MEG measurement is an ill-posed problem that requires prior assumptions about brain activity and an efficient estimation algorithm. In this article, we propose a new hierarchical Bayesian method introducing a hierarchical prior that can effectively incorporate both structural and functional MRI data. In our method, the variance of the source current at each source location is considered an unknown parameter and estimated from the observed MEG data and prior information by using the Variational Bayesian method. The fMRI information can be imposed as prior information on the variance distribution rather than the variance itself so that it gives a soft constraint on the variance. A spatial smoothness constraint, that the neural activity within a few millimeter radius tends to be similar due to the neural connections, can also be implemented as a hierarchical prior. The proposed method provides a unified theory to deal with the following three situations: (1) MEG with no other data, (2) MEG with structural MRI data on cortical surfaces, and (3) MEG with both structural MRI and fMRI data. We investigated the performance of our method and conventional linear inverse methods under these three conditions. Simulation results indicate that our method has better accuracy and spatial resolution than the conventional linear inverse methods under all three conditions. It is also shown that accuracy of our method improves as MRI and fMRI information becomes available. Simulation results demonstrate that our method appropriately resolves the inverse problem even if fMRI data convey inaccurate information, while the Wiener filter method is seriously deteriorated by inaccurate fMRI information.

Algorithms↗

PubMed2003

A unifying computational framework for motor control and social interaction.

Recent empirical studies have implicated the use of the motor system during action observation, imitation and social interaction. In this paper, we explore the computational parallels between the processes that occur in motor control and in action observation, imitation, social interaction and theory of mind. In particular, we examine the extent to which motor commands acting on the body can be equated with communicative signals acting on other people and suggest that computational solutions for motor control may have been extended to the domain of social interaction.

Humans↗

PubMed2003

Meta-learning in reinforcement learning.

Meta-parameters in reinforcement learning should be tuned to the environmental dynamics and the animal performance. Here, we propose a biologically plausible meta-reinforcement learning algorithm for tuning these meta-parameters in a dynamic, adaptive manner. We tested our algorithm in both a simulation of a Markov decision task and in a non-linear control task. Our results show that the algorithm robustly finds appropriate meta-parameter values, and controls the meta-parameter time course, in both static and dynamic environments. We suggest that the phasic and tonic components of dopamine neuron firing can encode the signal required for meta-learning of reinforcement learning.

Algorithms↗

PubMed2003

Inter-module credit assignment in modular reinforcement learning.

Critical issues in modular or hierarchical reinforcement learning (RL) are (i) how to decompose a task into sub-tasks, (ii) how to achieve independence of learning of sub-tasks, and (iii) how to assure optimality of the composite policy for the entire task. The second and last requirements are often under trade-off. We propose a method for propagating the reward for the entire task achievement between modules. This is done in the form of a 'modular reward', which is calculated from the temporal difference of the module gating signal and the value of the succeeding module. We implement modular reward for a multiple model-based reinforcement learning (MMRL) architecture and show its effectiveness in simulations of a pursuit task with hidden states and a continuous-time non-linear control task.

Learning↗

PubMed2002

Multiple model-based reinforcement learning.

We propose a modular reinforcement learning architecture for nonlinear, nonstationary control tasks, which we call multiple model-based reinforcement learning (MMRL). The basic idea is to decompose a complex task into multiple domains in space and time based on the predictability of the environmental dynamics. The system is composed of multiple modules, each of which consists of a state prediction model and a reinforcement learning controller. The "responsibility signal," which is given by the softmax function of the prediction errors, is used to weight the outputs of multiple modules, as well as to gate the learning of the prediction models and the reinforcement learning controllers. We formulate MMRL for both discrete-time, finite-state case and continuous-time, continuous-state case. The performance of MMRL was demonstrated for discrete case in a nonstationary hunting task in a grid world and for continuous case in a nonlinear, nonstationary control task of swinging up a pendulum with variable physical parameters.

Humans↗

PubMedDate not supplied

Metalearning and neuromodulation.

This paper presents a computational theory on the roles of the ascending neuromodulatory systems from the viewpoint that they mediate the global signals that regulate the distributed learning mechanisms in the brain. Based on the review of experimental data and theoretical models, it is proposed that dopamine signals the error in reward prediction, serotonin controls the time scale of reward prediction, noradrenaline controls the randomness in action selection, and acetylcholine controls the speed of memory update. The possible interactions between those neuromodulators and the environment are predicted on the basis of computational theory of metalearning.

Algorithms↗