New Zealand

Kunikazu Kobayashi, Koji Nakano, Takashi Kuremoto and Masanao Obayashi (February 1st 2010). Objective-based Reinforcement Learning System for Cooperative Behavior Acquisition, Application of Machine Learning, Yagang Zhang, IntechOpen, DOI: 10.5772/8615. 

Yoshiyuki Tajima and Takehisa Onisawa (February 1st 2010). Model-based Reinforcement Learning with Model Error and Its Application, Application of Machine Learning, Yagang Zhang, IntechOpen, DOI: 10.5772/8607. 

We report a previously unidentified issue with model-free, value-based approaches to multiobjective reinforcement learning in the context of environments with stochastic state transitions. An example multiobjective Markov Decision Process (MOMDP) is used to demonstrate that under such conditions these approaches may be unable to discover the policy which maximises the Scalarised Expected Return, and in fact may converge to a Pareto-dominated solution. We discuss several alternative methods which may be
more suitable for maximising SER in MOMDPs with stochastic transitions.

AAMAS 2020

A Demonstration of Issues with Value-Based Multiobjective Reinforcement Learning Under Stochastic State Transitions

stochastic mdps

multiobjective mdps

multiobjective reinforcement learning

The purpose of the AAAI conference is to promote research in artificial intelligence (AI) and scientific exchange among AI researchers, practitioners, scientists, and engineers in affiliated disciplines. 

AAAI 2021

technical paper

AAMAS is the leading scientific conference for research in autonomous agents and multi-agent systems. The AAMAS conference series was initiated in 2002 as the merging of three respected scientific meetings: the International Conference on Multi-Agent Systems (ICMAS), the International Workshop on Agent Theories, Architectures, and Languages (ATAL), and the International Conference on Autonomous Agents (AA). The aim of the joint conference is to provide a single, high-profile, internationally-respected archival forum for scientific research in the theory and practice of autonomous agents and multi-agent systems.

Browse keynotes, discussions, panels and over 300 presentations.


AAMAS is the leading scientific conference for research in autonomous agents and multi-agent systems. The AAMAS conference series was initiated in 2002 as the merging of three respected scientific meetings: the International Conference on Multi-Agent Systems (ICMAS), the International Workshop on Agent Theories, Architectures, and Languages (ATAL), and the International Conference on Autonomous Agents (AA). 

Autonomous agents (AA) will increasingly be interacting with us in our daily lives. While we want the benefits attached to AAs, it is essential that their behavior is aligned with our values and norms. Hence, an AA will need to estimate the values and norms of the humans it interacts with, which is not a straightforward task when solely observing an agent's behavior. This paper analyses to what extent an AA is able to estimate the values and norms of a simulated human agent (SHA) based on its actions in the ultimatum game. We present two methods to reduce ambiguity in profiling the SHAs: one based on search space exploration and another based on counterfactual analysis. We found that both methods are able to increase the confidence in estimating human values and norms, but differ in their applicability, the latter being more efficient when the number of interactions with the agent is to be minimized. These insights are useful to improve the alignment of AAs with human values and norms.

Improving Confidence in the Estimation of Values and Norms

A digital twin is a digital representation of a physical asset reproducing its data model, its behaviour and its communication with other physical assets. Although most of their concrete applications can be found mainly in the industrial context, healthcare represents another relevant area where digital twins can have a disruptive impact. The main research question tackled by this paper is about the integration of digital twins with agents and Multi-Agent Systems (MAS) technologies in healthcare, discussing what the benefits are and how such an integration could be realised.


On the Integration of Agents and Digital Twins in Healthcare

In many multi-agent interactions in the real world, agents receive payoffs over multiple distinct criteria; i.e. the payoffs are multi-objective in nature. However, the same multi-objective payoff vector may lead to different utilities for each participant. Therefore, it is essential for an agent to learn about the behaviour of other agents in the system. In this work, we present the first study of the effects of such opponent modelling on multi-objective multi-agent interactions with non-linear utilities. Specifically, we consider multi-objective normal form games with non-linear utility functions under the scalarised expected returns optimisation criterion. We contribute a novel actor-critic formulation to allow reinforcement learning of mixed strategies in this setting, along with an extension that incorporates opponent policy reconstruction using conditional action frequencies. Empirical results in five different MONFGs demonstrate that opponent modelling can drastically alter the learning dynamics in this setting. When equilibria are present opponent modelling can confer significant benefits on agents that implement it. However, when there are no Nash equilibria, opponent modelling can have adverse effects on utility, and has a neutral effect at best (after extensive hyper-parameter optimisation).


Opponent Modelling using Policy Reconstruction for Multi-Objective Normal Form Games

We compare synthetic population-based travel demand modeling with the state of the art
travel demand models used by metropolitan planning offices in the United States. Our 
comparison of the models for three US cities shows that synthetic population-based
models match the state of the art models closely for the temporal trip distributions and
the spatial distribution of destinations. The advantages of the synthetic population-based 
method are that it provides greater spatial resolution, can be generalized to any region,
and can be used for studying correlations with demographics and activity types, which are
useful for modeling the effects of policy changes.

Improved Travel Demand Modeling with Synthetic Populations

We study the problem of two competing camps aiming to maximize the adoption of their respective opinions, by optimally investing in nodes of a social network in two phases. The final opinion of a node in phase 1 acts as its bias in phase 2, and this bias determines the effectiveness of a camp's investment on the node. Using an extension of Friedkin-Johnsen model of opinion dynamics, we formulate the camps' utility functions. We show the existence and polynomial time computability of Nash equilibrium under reasonable assumptions. Using simulations, we quantify the effects of the nodes' biases and the weightage attributed to them, as well as that of a camp deviating from its equilibrium strategy.


A Two Phase Investment Game for Competitive Opinion Dynamics in Social Networks

We define noisy combinatorial markets, a specific class of combinatorial markets and introduce a model to study them. We present a learning methodology we developed to learn competitive equilibria of noisy combinatorial markets. We summarize preliminary experimental results to evaluate the performance of our learning algorithms.


Learning Competitive Equilibria  in Noisy Combinatorial Markets

This talk describes a policy search approach for railway scheduling using the covariance matrix adaptation evolution strategy (CMA-ES). The goal is to define arrival/departure times and track allocations for all trains such that the resource and operating constraints of the railway line are satisfied, and priority-weighted train departure delay is minimised. The proposed approach is scalable in the sense that (i) the optimised policy can be applied to an arbitrarily long railway line, independent of the number of trains, tracks, and stations, and (ii) the on-line implementation is computationally light enough to be applied in real-time. Our experiments show that policies computed with CMA-ES are able to produce solutions with lower priorityweighted delay than heuristics and reinforcement learning (RL) algorithms reported in literature, on synthetic examples as well as actual railway line data from portions of the Indian Railway network.

Optimising a Real-Time Scheduler for Railway Lines using Policy Search

Robustness is an important property of software systems, and the availability of proper feedback is seen as crucial to obtain it, especially in the case of systems of distributed and interconnected components.
Multiagent Systems (MAS) are valuable for conceptualizing and implementing distributed systems, but the current design methodologies for MAS fall short in addressing robustness in a systematic way at design time.
In this paper we outline our vision of how robustness in MAS can be granted as a design property. To this end, we exploit the notion of accountability as a mechanism for building reporting frameworks and, then, we describe how robustness is gained.
We exemplify our vision on the JaCaMo agent platform.

Fragility and Robustness in Multiagent Systems

Reinforcement learning requires exploration, leading to repeated execution of sub-optimal actions. Naive exploration techniques address this problem by changing gradually from exploration to exploitation. This approach employs a wide search resulting in exhaustive exploration and low sample-efficiency. More advanced search methods explore optimistically based on an upper bound estimate of expected rewards. These methods employ deep search, aiming to reach states not previously visited. Another deep search strategy is found in action-elimination methods, which aim to discover and eliminate sub-optimal actions. Despite the effectiveness of advanced deep search strategies, some problems are better suited to naive exploration. We devise a new method, called Interval Q-Learning, that finds a balance between wide and deep search. It assigns a small probability to taking sub-optimal actions and combines both greedy and optimistic exploration. This allows for fast convergence to a near-optimal policy, and then exploration around it. We demonstrate the performance of tabular and deep Q-network versions of Interval Q-Learning, showing that it offers convergence speed-up both in problems that favor wide exploration methods and those that favor deep search strategies.


Interval Q-Learning: Balancing Deep and Wide Exploration

Epidemics of infectious diseases are an important threat to public health and global economies. 
Yet, the development of prevention strategies remains a challenging process.
For this reason, we investigate a deep reinforcement learning approach to automatically learn prevention strategies in an epidemiological model, in the context of pandemic influenza.
To this end, we construct a new epidemiological meta-population model, with 379 patches, that balances between model complexity and computational efficiency such that the use of reinforcement learning techniques becomes attainable.
First, we set up a ground truth such that we can evaluate the performance of the ""Proximal Policy Optimization"" algorithm to learn in a single district of this epidemiological model.
Next, we consider a larger scale problem, by conducting an experiment where we aim to learn a joint policy to control the districts in a community of 11 tightly coupled districts, for which no ground truth can be established. 
This experiment shows that deep reinforcement learning can be used to learn mitigation policies in complex epidemiological models with a large state space.
Moreover, through this experiment, we demonstrate that there can be an advantage to consider collaboration between districts when designing prevention strategies.


Downloads

Next from AAMAS 2020

Improving Confidence in the Estimation of Values and Norms

Similar lecture

Wasserstein Distributionally Robust Inverse Multiobjective Optimization

.css-70qvj9{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;}Downloads

Next from AAMAS 2020

Improving Confidence in the Estimation of Values and Norms

Similar lecture

Wasserstein Distributionally Robust Inverse Multiobjective Optimization

Downloads