scholarly journals Imitation as a model-free process in human reinforcement learning

2019 ◽  
Author(s):  
Anis Najar ◽  
Emmanuelle Bonnet ◽  
Bahador Bahrami ◽  
Stefano Palminteri
Keyword(s):  
Reinforcement Learning ◽  
Model Comparison ◽  
Value Function ◽  
Action Selection ◽  
Learning Task ◽  
Social Signals ◽  
Model Free ◽  
Free Process ◽  
New Variant ◽  
New Perspective

While there is not doubt that social signals affect human reinforcement learning, there is still no consensus about their exact computational implementation. To address this issue, we compared three hypotheses about the algorithmic implementation of imitation in human reinforcement learning. A first hypothesis, decision biasing, postulates that imitation consists in transiently biasing the learner’s action selection without affecting her value function. According to the second hypothesis, model-based imitation, the learner infers the demonstrator’s value function through inverse reinforcement learning and uses it for action selection. Finally, according to the third hypothesis, value shaping, demonstrator’s actions directly affect the learner’s value function. We tested these three psychologically plausible hypotheses in two separate experiments (N = 24 and N = 44) featuring a new variant of a social reinforcement learning task, where we manipulated the quantity and the quality of the demonstrator’s choices. We show through model comparison that value shaping is favored, which provides a new perspective on how imitation is integrated into human reinforcement learning.

scholarly journals The actions of others act as a pseudo-reward to drive imitation in the context of social reinforcement learning

PLoS Biology ◽  
2020 ◽  
Vol 18 (12) ◽  
pp. e3001028
Author(s):  
Anis Najar ◽  
Emmanuelle Bonnet ◽  
Bahador Bahrami ◽  
Stefano Palminteri
Keyword(s):  
Reinforcement Learning ◽  
Model Comparison ◽  
Value Function ◽  
Model Simulation ◽  
Action Selection ◽  
Independent Experiment ◽  
Social Reinforcement ◽  
Social Signals ◽  
Hypothesis Model ◽  
New Variant

While there is no doubt that social signals affect human reinforcement learning, there is still no consensus about how this process is computationally implemented. To address this issue, we compared three psychologically plausible hypotheses about the algorithmic implementation of imitation in reinforcement learning. The first hypothesis, decision biasing (DB), postulates that imitation consists in transiently biasing the learner’s action selection without affecting their value function. According to the second hypothesis, model-based imitation (MB), the learner infers the demonstrator’s value function through inverse reinforcement learning and uses it to bias action selection. Finally, according to the third hypothesis, value shaping (VS), the demonstrator’s actions directly affect the learner’s value function. We tested these three hypotheses in 2 experiments (N = 24 and N = 44) featuring a new variant of a social reinforcement learning task. We show through model comparison and model simulation that VS provides the best explanation of learner’s behavior. Results replicated in a third independent experiment featuring a larger cohort and a different design (N = 302). In our experiments, we also manipulated the quality of the demonstrators’ choices and found that learners were able to adapt their imitation rate, so that only skilled demonstrators were imitated. We proposed and tested an efficient meta-learning process to account for this effect, where imitation is regulated by the agreement between the learner and the demonstrator. In sum, our findings provide new insights and perspectives on the computational mechanisms underlying adaptive imitation in human reinforcement learning.

Model-Based and Model-Free Social Cognition

2019 ◽  
Author(s):  
Leor M Hackel ◽  
Jeffrey Jordan Berg ◽  
Björn Lindström ◽  
David Amodio
Keyword(s):  
Reinforcement Learning ◽  
Social Cognition ◽  
Learning Strategies ◽  
Memory Systems ◽  
Learning Task ◽  
Financial Advisors ◽  
Model Based ◽  
Model Free ◽  
Systems Model ◽  
Task Assessment

Do habits play a role in our social impressions? To investigate the contribution of habits to the formation of social attitudes, we examined the roles of model-free and model-based reinforcement learning in social interactions—computations linked in past work to habit and planning, respectively. Participants in this study learned about novel individuals in a sequential reinforcement learning paradigm, choosing financial advisors who led them to high- or low-paying stocks. Results indicated that participants relied on both model-based and model-free learning, such that each independently predicted choice during the learning task and self-reported liking in a post-task assessment. Specifically, participants liked advisors who could provide large future rewards as well as advisors who had provided them with large rewards in the past. Moreover, participants varied in their use of model-based and model-free learning strategies, and this individual difference influenced the way in which learning related to self-reported attitudes: among participants who relied more on model-free learning, model-free social learning related more to post-task attitudes. We discuss implications for attitudes, trait impressions, and social behavior, as well as the role of habits in a memory systems model of social cognition.

scholarly journals Variability in Action Selection Relates to Striatal Dopamine 2/3 Receptor Availability in Humans: A PET Neuroimaging Study Using Reinforcement Learning and Active Inference Models

2020 ◽  
Vol 30 (6) ◽  
pp. 3573-3589 ◽  
Author(s):  
Rick A Adams ◽  
Michael Moutoussis ◽  
Matthew M Nour ◽  
Tarik Dahoun ◽  
Declan Lewis ◽  
...  
Keyword(s):  
Decision Making ◽  
Reinforcement Learning ◽  
Model Comparison ◽  
Behavioral Model ◽  
Action Selection ◽  
Striatal Dopamine ◽  
Active Inference ◽  
Inference Models ◽  
Positron Emission ◽  
Dopamine Signaling

Abstract Choosing actions that result in advantageous outcomes is a fundamental function of nervous systems. All computational decision-making models contain a mechanism that controls the variability of (or confidence in) action selection, but its neural implementation is unclear—especially in humans. We investigated this mechanism using two influential decision-making frameworks: active inference (AI) and reinforcement learning (RL). In AI, the precision (inverse variance) of beliefs about policies controls action selection variability—similar to decision ‘noise’ parameters in RL—and is thought to be encoded by striatal dopamine signaling. We tested this hypothesis by administering a ‘go/no-go’ task to 75 healthy participants, and measuring striatal dopamine 2/3 receptor (D2/3R) availability in a subset (n = 25) using [11C]-(+)-PHNO positron emission tomography. In behavioral model comparison, RL performed best across the whole group but AI performed best in participants performing above chance levels. Limbic striatal D2/3R availability had linear relationships with AI policy precision (P = 0.029) as well as with RL irreducible decision ‘noise’ (P = 0.020), and this relationship with D2/3R availability was confirmed with a ‘decision stochasticity’ factor that aggregated across both models (P = 0.0006). These findings are consistent with occupancy of inhibitory striatal D2/3Rs decreasing the variability of action selection in humans.

scholarly journals Risk-Sensitive Reinforcement Learning Applied to Control under Constraints

2005 ◽  
Vol 24 ◽  
pp. 81-108 ◽  
Author(s):  
P. Geibel ◽  
F. Wysotzki
Keyword(s):  
Reinforcement Learning ◽  
Value Function ◽  
Learning Algorithm ◽  
Optimal Solution ◽  
Feed Tank ◽  
Model Free ◽  
Constrained Problem ◽  
Risk Sensitive ◽  
Markov Decision ◽  
The Value Function

In this paper, we consider Markov Decision Processes (MDPs) with error states. Error states are those states entering which is undesirable or dangerous. We define the risk with respect to a policy as the probability of entering such a state when the policy is pursued. We consider the problem of finding good policies whose risk is smaller than some user-specified threshold, and formalize it as a constrained MDP with two criteria. The first criterion corresponds to the value function originally given. We will show that the risk can be formulated as a second criterion function based on a cumulative return, whose definition is independent of the original value function. We present a model free, heuristic reinforcement learning algorithm that aims at finding good deterministic policies. It is based on weighting the original value function and the risk. The weight parameter is adapted in order to find a feasible solution for the constrained problem that has a good performance with respect to the value function. The algorithm was successfully applied to the control of a feed tank with stochastic inflows that lies upstream of a distillation column. This control task was originally formulated as an optimal control problem with chance constraints, and it was solved under certain assumptions on the model to obtain an optimal solution. The power of our learning algorithm is that it can be used even when some of these restrictive assumptions are relaxed.

scholarly journals Multiobjective model-free learning for robot pathfinding with environmental disturbances

2019 ◽  
Vol 16 (6) ◽  
pp. 172988141988570 ◽  
Author(s):  
Changyun Wei ◽  
Fusheng Ni
Keyword(s):  
Reinforcement Learning ◽  
Action Selection ◽  
Two Dimensions ◽  
Selection Strategy ◽  
Sequential Decision ◽  
Environmental Disturbances ◽  
Learning Framework ◽  
Model Free ◽  
Conflicting Objectives ◽  
Save Energy

This article addresses the robot pathfinding problem with environmental disturbances, where a solution to this problem must consider potential risks inherent in an uncertain and stochastic environment. For example, the movements of an underwater robot can be seriously disturbed by ocean currents, and thus any applied control actions to the robot cannot exactly lead to the desired locations. Reinforcement learning is a formal methodology that has been extensively studied in many sequential decision-making domains with uncertainty, but most reinforcement learning algorithms consider only a single objective encoded by a scalar reward. However, the robot pathfinding problem with environmental disturbances naturally promotes multiple conflicting objectives. Specifically, in this work, the robot has to minimise its moving distance so as to save energy, and, moreover, it has to keep away from unsafe regions as far as possible. To this end, we first propose a multiobjective model-free learning framework, and then proceed to investigate an appropriate action selection strategy by improving a baseline with respect to two dimensions. To demonstrate the effectiveness of the proposed learning framework and evaluate the performance of three action selection strategies, we also carry out an empirical study in a simulated environment.

Planning Complexity Registers as a Cost in Metacontrol

2018 ◽  
Vol 30 (10) ◽  
pp. 1391-1404 ◽  
Author(s):  
Wouter Kool ◽  
Samuel J. Gershman ◽  
Fiery A. Cushman
Keyword(s):  
Decision Making ◽  
Reinforcement Learning ◽  
Formal Analysis ◽  
Learning Task ◽  
Costs And Benefits ◽  
Model Free ◽  
Versus Model ◽  
Specific Manner ◽  
Decision Making Processes ◽  
Planning Approaches

Decision-making algorithms face a basic tradeoff between accuracy and effort (i.e., computational demands). It is widely agreed that humans can choose between multiple decision-making processes that embody different solutions to this tradeoff: Some are computationally cheap but inaccurate, whereas others are computationally expensive but accurate. Recent progress in understanding this tradeoff has been catalyzed by formalizing it in terms of model-free (i.e., habitual) versus model-based (i.e., planning) approaches to reinforcement learning. Intuitively, if two tasks offer the same rewards for accuracy but one of them is much more demanding, we might expect people to rely on habit more in the difficult task: Devoting significant computation to achieve slight marginal accuracy gains would not be “worth it.” We test and verify this prediction in a sequential reinforcement learning task. Because our paradigm is amenable to formal analysis, it contributes to the development of a computational model of how people balance the costs and benefits of different decision-making processes in a task-specific manner; in other words, how we decide when hard thinking is worth it.

scholarly journals Gaze-contingent reinforcement learning reveals incentive value of social signals in young children and adults

2017 ◽  
Vol 284 (1850) ◽  
pp. 20162747 ◽  
Author(s):  
Angélina Vernetti ◽  
Tim J. Smith ◽  
Atsushi Senju
Keyword(s):  
Reinforcement Learning ◽  
Visual Cues ◽  
Real Life ◽  
Learning Task ◽  
Social Cues ◽  
Social Signals ◽  
Social Orienting ◽  
Incentive Value ◽  
Reward Delivery ◽  
New Findings

While numerous studies have demonstrated that infants and adults preferentially orient to social stimuli, it remains unclear as to what drives such preferential orienting. It has been suggested that the learned association between social cues and subsequent reward delivery might shape such social orienting. Using a novel, spontaneous indication of reinforcement learning (with the use of a gaze contingent reward-learning task), we investigated whether children and adults' orienting towards social and non-social visual cues can be elicited by the association between participants' visual attention and a rewarding outcome. Critically, we assessed whether the engaging nature of the social cues influences the process of reinforcement learning. Both children and adults learned to orient more often to the visual cues associated with reward delivery, demonstrating that cue–reward association reinforced visual orienting. More importantly, when the reward-predictive cue was social and engaging, both children and adults learned the cue–reward association faster and more efficiently than when the reward-predictive cue was social but non-engaging. These new findings indicate that social engaging cues have a positive incentive value. This could possibly be because they usually coincide with positive outcomes in real life, which could partly drive the development of social orienting.

Reinforcement Learning in Continuous Time and Space

2000 ◽  
Vol 12 (1) ◽  
pp. 219-245 ◽  
Author(s):  
Kenji Doya
Keyword(s):  
Reinforcement Learning ◽  
Dynamic Model ◽  
Continuous Time ◽  
Value Function ◽  
Infinite Horizon ◽  
Euler Approximation ◽  
Model Free ◽  
A Value ◽  
Backward Euler ◽  
Gradient Based

This article presents a reinforcement learning framework for continuous-time dynamical systems without a priori discretization of time, state, and action. Basedonthe Hamilton-Jacobi-Bellman (HJB) equation for infinite-horizon, discounted reward problems, we derive algorithms for estimating value functions and improving policies with the use of function approximators. The process of value function estimation is formulated as the minimization of a continuous-time form of the temporal difference (TD) error. Update methods based on backward Euler approximation and exponential eligibility traces are derived, and their correspondences with the conventional residual gradient, TD (0), and TD (λ) algorithms are shown. For policy improvement, two methods—a continuous actor-critic method and a value-gradient-based greedy policy—are formulated. As a special case of the latter, a nonlinear feedback control law using the value gradient and the model of the input gain is derived. The advantage updating, a model-free algorithm derived previously, is also formulated in the HJB-based framework. The performance of the proposed algorithms is first tested in a nonlinear control task of swinging a pendulum up with limited torque. It is shown in the simulations that (1) the task is accomplished by the continuous actor-critic method in a number of trials several times fewer than by the conventional discrete actor-critic method; (2) among the continuous policy update methods, the value-gradient-based policy with a known or learned dynamic model performs several times better than the actor-critic method; and (3) a value function update using exponential eligibility traces is more efficient and stable than that based on Euler approximation. The algorithms are then tested in a higher-dimensional task: cart-pole swing-up. This task is accomplished in several hundred trials using the value-gradient-based policy with a learned dynamic model.

scholarly journals Hierarchical Reinforcement Learning with the MAXQ Value Function Decomposition

2000 ◽  
Vol 13 ◽  
pp. 227-303 ◽  
Author(s):  
T. G. Dietterich
Keyword(s):  
Reinforcement Learning ◽  
Optimal Policy ◽  
Value Function ◽  
Learning Algorithm ◽  
Value Functions ◽  
Procedural Semantics ◽  
Hierarchical Reinforcement Learning ◽  
Model Free ◽  
Function Decomposition ◽  
The Value Function

This paper presents a new approach to hierarchical reinforcement learning based on decomposing the target Markov decision process (MDP) into a hierarchy of smaller MDPs and decomposing the value function of the target MDP into an additive combination of the value functions of the smaller MDPs. The decomposition, known as the MAXQ decomposition, has both a procedural semantics---as a subroutine hierarchy---and a declarative semantics---as a representation of the value function of a hierarchical policy. MAXQ unifies and extends previous work on hierarchical reinforcement learning by Singh, Kaelbling, and Dayan and Hinton. It is based on the assumption that the programmer can identify useful subgoals and define subtasks that achieve these subgoals. By defining such subgoals, the programmer constrains the set of policies that need to be considered during reinforcement learning. The MAXQ value function decomposition can represent the value function of any policy that is consistent with the given hierarchy. The decomposition also creates opportunities to exploit state abstractions, so that individual MDPs within the hierarchy can ignore large parts of the state space. This is important for the practical application of the method. This paper defines the MAXQ hierarchy, proves formal results on its representational power, and establishes five conditions for the safe use of state abstractions. The paper presents an online model-free learning algorithm, MAXQ-Q, and proves that it converges with probability 1 to a kind of locally-optimal policy known as a recursively optimal policy, even in the presence of the five kinds of state abstraction. The paper evaluates the MAXQ representation and MAXQ-Q through a series of experiments in three domains and shows experimentally that MAXQ-Q (with state abstractions) converges to a recursively optimal policy much faster than flat Q learning. The fact that MAXQ learns a representation of the value function has an important benefit: it makes it possible to compute and execute an improved, non-hierarchical policy via a procedure similar to the policy improvement step of policy iteration. The paper demonstrates the effectiveness of this non-hierarchical execution experimentally. Finally, the paper concludes with a comparison to related work and a discussion of the design tradeoffs in hierarchical reinforcement learning.

Sign in / Sign up

Export Citation Format

Share Document