A neural network model for the orbitofrontal cortex and task space acquisition during reinforcement learning

Representational spaces in orbitofrontal and ventromedial prefrontal cortex: task states, values, and beyond

The orbitofrontal cortex (OFC) and ventromedial-prefrontal cortex (vmPFC) play a key role in decision-making and encode task states in addition to expected value. We review evidence suggesting a connection between value and state representations and argue that OFC / vmPFC integrate stimulus, context, and outcome information. Comparable encoding principles emerge in late layers of deep reinforcement learning (RL) models, where single nodes exhibit similar forms of mixed-selectivity, which enables flexible readout of relevant variables by downstream neurons. Based on these lines of evidence, we suggest that outcome-maximization leads to complex representational spaces that are insufficiently characterized by linear value signals that have been the focus of most prior research on the topic. Major outstanding questions concern the role of OFC/ vmPFC in learning across tasks, in encoding of task-irrelevant aspects, and the role of hippocampus-PFC interactions.

Orbitofrontal cortex contributes to the comparison of values underlying economic choices

Economic choices between goods entail the computation and comparison of subjective values. Previous studies examined neuronal activity in the orbitofrontal cortex (OFC) of monkeys choosing between different types of juices. Three groups of neurons were identified: offer value cells encoding the value of individual offers, chosen juice cells encoding the identity of the chosen juice, and chosen value cells encoding the value of the chosen offer. The encoded variables capture both the input (offer value) and the output (chosen juice, chosen value) of the decision process, suggesting that values are compared within OFC. Recent work demonstrates that choices are causally linked to the activity of offer value cells. Conversely, the hypothesis that OFC contributes to value comparison has not been confirmed. Here we show that weak electrical stimulation of OFC specifically disrupts value comparison without altering offer values. This result implies that neuronal populations in OFC participate in value comparison.

Neuronal origins of reduced accuracy and biases in economic choices under sequential offers

Economic choices are characterized by a variety of biases. Understanding their origins is a long-term goal for neuroeconomics, but progress on this front has been limited. Here, we examined choice biases observed when two goods are offered sequentially. In the experiments, rhesus monkeys chose between different juices offered simultaneously or in sequence. Choices under sequential offers were less accurate (higher variability). They were also biased in favor of the second offer (order bias) and in favor of the preferred juice (preference bias). Analysis of neuronal activity recorded in the orbitofrontal cortex revealed that these phenomena emerged at different computational stages. Lower choice accuracy reflected weaker offer value signals (valuation stage), the order bias emerged during value comparison (decision stage), and the preference bias emerged late in the trial (post-comparison). By neuronal measures, each phenomenon reduced the value obtained on average in each trial and was thus costly to the monkey.

From Biological Synapses to “Intelligent” Robots

This selective review explores biologically inspired learning as a model for intelligent robot control and sensing technology on the basis of specific examples. Hebbian synaptic learning is discussed as a functionally relevant model for machine learning and intelligence, as explained on the basis of examples from the highly plastic biological neural networks of invertebrates and vertebrates. Its potential for adaptive learning and control without supervision, the generation of functional complexity, and control architectures based on self-organization is brought forward. Learning without prior knowledge based on excitatory and inhibitory neural mechanisms accounts for the process through which survival-relevant or task-relevant representations are either reinforced or suppressed. The basic mechanisms of unsupervised biological learning drive synaptic plasticity and adaptation for behavioral success in living brains with different levels of complexity. The insights collected here point toward the Hebbian model as a choice solution for “intelligent” robotics and sensor systems.

A Nonlinear Hidden Layer Enables Actor-Critic Agents to Learn Multiple Paired Association Navigation

Navigation to multiple cued reward locations has been increasingly used to study rodent learning. Though deep reinforcement learning agents have been shown to be able to learn the task, they are not biologically plausible. Biologically plausible classic actor-critic agents have been shown to learn to navigate to single reward locations, but which biologically plausible agents are able to learn multiple cue-reward location tasks has remained unclear. In this computational study, we show versions of classic agents that learn to navigate to a single reward location, and adapt to reward location displacement, but are not able to learn multiple paired association navigation. The limitation is overcome by an agent in which place cell and cue information are first processed by a feedforward nonlinear hidden layer with synapses to the actor and critic subject to temporal difference error-modulated plasticity. Faster learning is obtained when the feedforward layer is replaced by a recurrent reservoir network.

Economic Choices under Simultaneous or Sequential Offers Rely on the Same Neural Circuit

A series of studies in which monkeys chose between two juices offered in variable amounts identified in the orbitofrontal cortex (OFC) different groups of neurons encoding the value of individual options (offer value), the binary choice outcome (chosen juice), and the chosen value. These variables capture both the input and the output of the choice process, suggesting that the cell groups identified in OFC constitute the building blocks of a decision circuit. Several lines of evidence support this hypothesis. However, in previous experiments offers were presented simultaneously, raising the question of whether current notions generalize to when goods are presented or are examined in sequence. Recently, Ballesta and Padoa-Schioppa (2019) examined OFC activity under sequential offers. An analysis of neuronal responses across time windows revealed that a small number of cell groups encoded specific sequences of variables. These sequences appeared analogous to the variables identified under simultaneous offers, but the correspondence remained tentative. Thus, in the present study, we examined the relation between cell groups found under sequential versus simultaneous offers. We recorded from the OFC while monkeys chose between different juices. Trials with simultaneous and sequential offers were randomly interleaved in each session. We classified cells in each choice modality, and we examined the relation between the two classifications. We found a strong correspondence; in other words, the cell groups measured under simultaneous offers and under sequential offers were one and the same. This result indicates that economic choices under simultaneous or sequential offers rely on the same neural circuit.SIGNIFICANCE STATEMENT Research in the past 20 years has shed light on the neuronal underpinnings of economic choices. A large number of results indicates that decisions between goods are formed in a neural circuit within the orbitofrontal cortex. In most previous studies, subjects chose between two goods offered simultaneously. Yet, in daily situations, goods available for choice are often presented or examined in sequence. Here we recorded neuronal activity in the primate orbitofrontal cortex alternating trials under simultaneous and under sequential offers. Our analyses demonstrate that the same neural circuit supports choices in the two modalities. Hence, current notions on the neuronal mechanisms underlying economic decisions generalize to choices under sequential offers.

Category learning in a recurrent neural network with reinforcement learning

It is known that humans and animals can learn and utilize category information quickly and efficiently to adapt to changing environments, and several brain areas are involved in learning and encoding category information. However, it is unclear that how the brain system learns and forms categorical representations from the view of neural circuits. In order to investigate this issue from the network level, we combine a recurrent neural network with reinforcement learning to construct a deep reinforcement learning model to demonstrate how the category is learned and represented in the network. The model consists of a policy network and a value network. The policy network is responsible for updating the policy to choose actions, while the value network is responsible for evaluating the action to predict rewards. The agent learns dynamically through the information interaction between the policy network and the value network. This model was trained to learn six stimulus-stimulus associative chains in a sequential paired-association task that was learned by the monkey. The simulated results demonstrated that our model was able to learn the stimulus-stimulus associative chains, and successfully reproduced the similar behavior of the monkey performing the same task. Two types of neurons were found in this model: one type primarily encoded identity information about individual stimuli; the other type mainly encoded category information of associated stimuli in one chain. The two types of activity-patterns were also observed in the primate prefrontal cortex after the monkey learned the same task. Furthermore, the ability of these two types of neurons to encode stimulus or category information was enhanced during this model was learning the task. Our results suggest that the neurons in the recurrent neural network have the ability to form categorical representations through deep reinforcement learning during learning stimulus-stimulus associations. It might provide a new approach for understanding neuronal mechanisms underlying how the prefrontal cortex learns and encodes category information.

A recurrent neural network framework for flexible and adaptive decision making based on sequence learning

The brain makes flexible and adaptive responses in a complicated and ever-changing environment for an organism's survival. To achieve this, the brain needs to understand the contingencies between its sensory inputs, actions, and rewards. This is analogous to the statistical inference that has been extensively studied in the natural language processing field, where recent developments of recurrent neural networks have found many successes. We wonder whether these neural networks, the gated recurrent unit (GRU) networks in particular, reflect how the brain solves the contingency problem. Therefore, we build a GRU network framework inspired by the statistical learning approach of NLP and test it with four exemplar behavior tasks previously used in empirical studies. The network models are trained to predict future events based on past events, both comprising sensory, action, and reward events. We show the networks can successfully reproduce animal and human behavior. The networks generalize the training, perform Bayesian inference in novel conditions, and adapt their choices when event contingencies vary. Importantly, units in the network encode task variables and exhibit activity patterns that match previous neurophysiology findings. Our results suggest that the neural network approach based on statistical sequence learning may reflect the brain's computational principle underlying flexible and adaptive behaviors and serve as a useful approach to understand the brain.

Neural mechanisms of economic choices in mice

Economic choices entail computing and comparing subjective values. Evidence from primates indicates that this behavior relies on the orbitofrontal cortex. Conversely, previous work in rodents provided conflicting results. Here we present a mouse model of economic choice behavior, and we show that the lateral orbital (LO) area is intimately related to the decision process. In the experiments, mice chose between different juices offered in variable amounts. Choice patterns closely resembled those measured in primates. Optogenetic inactivation of LO dramatically disrupted choices by inducing erratic changes of relative value and by increasing choice variability. Neuronal recordings revealed that different groups of cells encoded the values of individual options, the binary choice outcome and the chosen value. These groups match those previously identified in primates, except that the neuronal representation in mice is spatial (in monkeys it is good-based). Our results lay the foundations for a circuit-level analysis of economic decisions.

Activity in orbitofrontal neuronal ensembles reflects inhibitory control

Stopping, or inhibition, is a form of self-control that is a core element of flexible and adaptive behavior. Its neural origins remain unclear. Some views hold that inhibition decisions reflect the aggregation of widespread and diverse pieces of information, including information arising in ostensible core reward regions (i.e., outside the canonical executive system). We recorded activity of single neurons in the orbitofrontal cortex (OFC) of macaques, a region associated with economic decisions, and whose role in inhibition is debated. Subjects performed a classic inhibition task known as the stop signal task. Ensemble decoding analyses reveal a clear firing rate pattern that distinguishes successful from failed inhibition and that begins after the stop signal and before the stop signal reaction time (SSRT). We also found a different and orthogonal ensemble pattern that distinguishes successful from failed stopping before the beginning of the trial. These signals were distinct from, and orthogonal to, value encoding, which was also observed in these neurons. The timing of the early and late signals was, respectively, consistent with the idea that neuronal activity in OFC encodes inhibition both proactively and reactively.

Simple framework for constructing functional spiking recurrent neural networks

Cortical microcircuits exhibit complex recurrent architectures that possess dynamically rich properties. The neurons that make up these microcircuits communicate mainly via discrete spikes, and it is not clear how spikes give rise to dynamics that can be used to perform computationally challenging tasks. In contrast, continuous models of rate-coding neurons can be trained to perform complex tasks. Here, we present a simple framework to construct biologically realistic spiking recurrent neural networks (RNNs) capable of learning a wide range of tasks. Our framework involves training a continuous-variable rate RNN with important biophysical constraints and transferring the learned dynamics and constraints to a spiking RNN in a one-to-one manner. The proposed framework introduces only 1 additional parameter to establish the equivalence between rate and spiking RNN models. We also study other model parameters related to the rate and spiking networks to optimize the one-to-one mapping. By establishing a close relationship between rate and spiking models, we demonstrate that spiking RNNs could be constructed to achieve similar performance as their counterpart continuous rate networks.

Simple framework for constructing functional spiking recurrent neural networks

Cortical microcircuits exhibit complex recurrent architectures that possess dynamically rich properties. The neurons that make up these microcircuits communicate mainly via discrete spikes, and it is not clear how spikes give rise to dynamics that can be used to perform computationally challenging tasks. In contrast, continuous models of rate-coding neurons can be trained to perform complex tasks. Here, we present a simple framework to construct biologically realistic spiking recurrent neural networks (RNNs) capable of learning a wide range of tasks. Our framework involves training a continuous-variable rate RNN with important biophysical constraints and transferring the learned dynamics and constraints to a spiking RNN in a one-to-one manner. The proposed framework introduces only 1 additional parameter to establish the equivalence between rate and spiking RNN models. We also study other model parameters related to the rate and spiking networks to optimize the one-to-one mapping. By establishing a close relationship between rate and spiking models, we demonstrate that spiking RNNs could be constructed to achieve similar performance as their counterpart continuous rate networks.

Economic Decisions through Circuit Inhibition

Economic choices between goods are thought to rely on the orbitofrontal cortex (OFC), but the decision mechanisms remain poorly understood. To shed light on this fundamental issue, we recorded from the OFC of monkeys choosing between two juices offered sequentially. An analysis of firing rates across time windows revealed the presence of different groups of neurons similar to those previously identified under simultaneous offers. This observation suggested that economic decisions in the two modalities are formed in the same neural circuit. We then examined several hypotheses on the decision mechanisms. OFC neurons encoded good identities and values in a juice-based representation (labeled lines). Contrary to previous assessments, our data argued against the idea that decisions rely on mutual inhibition at the level of offer values. In fact, we showed that previous arguments for mutual inhibition were confounded by differences in value ranges. Instead, decisions seemed to involve mechanisms of circuit inhibition, whereby each offer value indirectly inhibited neurons encoding the opposite choice outcome. Our results reconcile a variety of previous findings and provide a general account for the neuronal underpinnings of economic choices.

Decoding electroencephalographic signals for direction in brain-computer interface using echo state network and Gaussian readouts

BACKGROUND

Noninvasive brain-computer interfaces (BCI) for movement control via an electroencephalogram (EEG) have been extensively investigated. However, most previous studies decoded user intention for movement directions based on sensorimotor rhythms during motor imagery. BCI systems based on mapping imagery movement of body parts (e.g., left or right hands) to movement directions (left or right directional movement of a machine or cursor) are less intuitive and less convenient due to the complex training procedures. Thus, direct decoding methods for detecting user intention about movement directions are urgently needed.

METHODS

Here, we describe a novel direct decoding method for user intention about the movement directions using the echo state network and Gaussian readouts. Importantly parameters in the network were optimized using the genetic algorithm method to achieve better decoding performance. We tested the decoding performance of this method with four healthy subjects and an inexpensive wireless EEG system containing 14 channels and then compared the performance outcome with that of a conventional machine learning method.

RESULTS

We showed that this decoding method successfully classified eight directions of intended movement (approximately 95% of an accuracy).

CONCLUSIONS

We suggest that the echo state network and Gaussian readouts can be a useful decoding method to directly read user intention of movement directions even using an inexpensive and portable EEG system.

Models that learn how humans learn: The case of decision-making and its disorders

Popular computational models of decision-making make specific assumptions about learning processes that may cause them to underfit observed behaviours. Here we suggest an alternative method using recurrent neural networks (RNNs) to generate a flexible family of models that have sufficient capacity to represent the complex learning and decision- making strategies used by humans. In this approach, an RNN is trained to predict the next action that a subject will take in a decision-making task and, in this way, learns to imitate the processes underlying subjects' choices and their learning abilities. We demonstrate the benefits of this approach using a new dataset drawn from patients with either unipolar (n = 34) or bipolar (n = 33) depression and matched healthy controls (n = 34) making decisions on a two-armed bandit task. The results indicate that this new approach is better than baseline reinforcement-learning methods in terms of overall performance and its capacity to predict subjects' choices. We show that the model can be interpreted using off-policy simulations and thereby provides a novel clustering of subjects' learning processes-something that often eludes traditional approaches to modelling and behavioural analysis.

Optimal coding and neuronal adaptation in economic decisions

During economic decisions, offer value cells in orbitofrontal cortex (OFC) encode the values of offered goods. Furthermore, their tuning functions adapt to the range of values available in any given context. A fundamental and open question is whether range adaptation is behaviorally advantageous. Here we present a theory of optimal coding for economic decisions. We propose that the representation of offer values is optimal if it ensures maximal expected payoff. In this framework, we examine offer value cells in non-human primates. We show that their responses are quasi-linear even when optimal tuning functions are highly non-linear. Most importantly, we demonstrate that for linear tuning functions range adaptation maximizes the expected payoff. Thus value coding in OFC is functionally rigid (linear tuning) but parametrically plastic (range adaptation with optimal gain). Importantly, the benefit of range adaptation outweighs the cost of functional rigidity. While generally suboptimal, linear tuning may facilitate transitive choices.