New Zealand

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. 

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. 

Reinforcement learning under partial observability poses two types of conceptually orthogonal learning challenges; learning a useful representation
of the observable past, and selecting actions based on that representation. While the latter is common to both fully and partially observable control
problems, the former is idiomatic to the partially observable case. Traditional reinforcement learning relies exclusively on the reward signal to train the
representation model, but rewards constitute a weak learning signal in partially observable settings. To help train richer representations, we discuss
and propose an auxiliary self-supervised learning task designed to help the primary reinforcement learning task. We further propose two training schemes
for training with auxiliary tasks, one inspired by the idea of promoting complementary representations. Empirical evaluations show that agents which
exploit auxiliary tasks and complementary representations learn better policies, and converge faster, compared to agents which use common reactive and
recurrent representations trained on the rewards alone.

AAMAS 2020

Learning Complementary Representations of the Past using Auxiliary Tasks in Partially Observable Reinforcement Learning

history representations

partially observable reinforcement learning

auxiliary tasks

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). 

technical paper

We consider solving a cooperative multi-robot object manipulation task using reinforcement learning (RL). We propose two distributed multi-agent RL approaches: distributed approximate RL (DA-RL), where each agent applies Q-learning with individual reward functions; and game-theoretic RL (GT-RL), where the agents update their Q-values based on the Nash equilibrium of a bimatrix Q-value game. We validate the proposed approaches in the setting of cooperative object manipulation with two simulated robot arms. Although we focus on a small system of two agents in this paper, both DA-RL and GT-RL apply to general multi-agent systems, and are expected to scale well to large systems.

Distributed Reinforcement Learning for Cooperative Multi-Robot Object Manipulation

Auction mechanism generally requires a trusted-third party as the market mediator to coordinate bidding and resource allocation via collecting private data from the agents, which may arouse severe privacy concerns and high computation overheads. To address such issues, we propose a novel privacy-aware double auction framework (namely PANDA) by designing an efficient cryptographic protocol to privately execute double auction for divisible resources among all the agents. To ensure privacy and truthfulness, PANDA delicately co-designs VCG auction and cryptographic protocol, which is equivalent to a mediator for sealed-bid auction of divisible resources. 


PANDA: Privacy-Aware Double Auction for Divisible Resources without a Mediator

Autonomous driving promises to transform road transport. Multi-vehicle and multi-lane scenarios, however, present unique challenges due to constrained navigation and unpredictable vehicle interactions. Learning-based methods—such as deep reinforcement learning—are emerging as a promising approach to automatically design intelligent driving policies that can cope with these challenges. Yet, the process of safely learning multi-vehicle driving behaviours is hard: while collisions—and their near-avoidance—are essential to the learning process, directly executing immature policies on autonomous vehicles raises considerable safety concerns. In this article, we present a safe and efficient framework that enables the learning of driving policies for autonomous vehicles operating in a shared workspace, where the absence of collisions cannot be guaranteed. Key to our learning procedure is a sim2real approach that uses real-world online policy adaptation in a mixed reality setup, where other vehicles and static obstacles exist in the virtual domain. This allows us to perform safe learning by simulating (and learning from) collisions between the learning agent(s) and other objects in virtual reality. Our results demonstrate that, after only a few runs in mixed reality, collisions are significantly reduced.


Mixed Reality Reinforcement Learning For Autonomous Multi-Lane Driving

An Agent-Based Model for Trajectory Modelling in Shared Spaces: A Combination of Expert Based and Deep Learning Approaches

Online social networks lack support for the collaborative management of access control. This is crucial for content that may involve multiple users such as photos, as this lack of support causes conflicts that lead to privacy violations. Previous research proposed collaborative mechanisms to support users in these cases, but most of these attempts fail to satisfy some desirable requirements, such as explainability, role-agnosticism, adaptability, and being utility- and value-driven at the same time. In this paper, we outline an agent architecture that has been designed to meet all these requirements.


Towards a Value-driven Explainable Agent for Collective Privacy

The booming sharing economy has a common challenge in motivating agents, by truthful actions, to follow the economic design of the market maker. For the well known BitTorrent network, there has been an array of theoretical research in the study of the truthfulness of the proportional response protocol formulated to model its tit-for-tat strategy. Some works confirm the truthfulness for deviations against the false name attack, weight cheating, and edge deleting. Others reveal it not truthful against Sybil attack where an agent may split its resource among its different copies. Though truthfulness cannot be guaranteed, the incentives are shown to generate a limited gain for agents pursuing such an attack in several special networks, such as trees, cliques, and cycles. In this work, we establish the first such results for general networks, proving a \theta(1) incentive ratio for any agent playing such an attack.


Limiting the Deviation Incentives in Resource Sharing Networks

Adversarial attacks have exposed a significant security vulnerability in state-of-the-art machine learning models. Among these models include deep reinforcement learning agents. The existing methods for attacking reinforcement learning agents assume the adversary either has access to the target agent's learned parameters or the environment that the agent interacts with. In this work, we propose a new class of threat models, called snooping threat models, that are unique to reinforcement learning. In these snooping threat models, the adversary does not have the ability to interact with the target agent's environment, and can only eavesdrop on the action and reward signals being exchanged between agent and environment. We show that adversaries operating in these highly constrained threat models can still launch devastating attacks against the target agent by training proxy models on related tasks and leveraging the transferability of adversarial examples.


Snooping Attacks on Deep Reinforcement Learning

An interaction protocol specifies a decentralized multiagent system operationally by specifying constraints on messages exchanged by its member agents.  Engineering with protocols requires support for a notion of refinement, whereby a protocol may be substituted without loss of correctness by one that refines it.  We identify two desiderata for refinement. One, generality: refinement should not restrict enactments by limiting protocols or infrastructures under consideration.  Two, preservation: to facilitate modular verification, refinement should preserve liveness and safety.

  We contribute a novel formal notion of protocol refinement based on enactments. We demonstrate generality by tackling the declarative framework of information protocols.  We demonstrate preservation by formally establishing that our notion of refinement is safety and liveness preserving. We show the practical benefits of refinement by implementing a checker. We demonstrate that it is less time-intensive to check refinement (and thereby gain safety and liveness) than to recheck safety and liveness of a composition.


Refinement for Multiagent Protocols

We propose a formalism to model and reason about multi-agent systems. We allow agents to interact and communicate in different modes so that they can pursue joint tasks; agents may dynamically synchronize, exchange data, adapt their behaviour, and reconfigure their communication interfaces. The formalism defines a local behaviour based on shared variables and a global one based on message passing. We extend LTL to be able to reason explicitly about the intentions of the different agents and their interaction protocols. We also study the complexity of satisfiability and model-checking of this extension.


Reconfigurable Interaction for MAS Modelling

In this paper we present a novel approach to optimise tactical and strategic decision making in football (soccer). We model the game of football as a multi-stage game which is made up from a Bayesian game to model the pre-match decisions and a stochastic game to model the in-match state transitions and decisions. Using this formulation, we propose a method to predict the probability of game outcomes and the payoffs of team actions. Building upon this, we develop algorithms to optimise team formation and in-game tactics with different objectives. Empirical evaluation of our approach on real-world datasets from 760 matches shows that by using optimised tactics from our Bayesian and stochastic games, we increase a team chances of winning by 16.1% and 3.4% respectively. 


Downloads

Next from AAMAS 2020

Distributed Reinforcement Learning for Cooperative Multi-Robot Object Manipulation

Similar lecture

Optimal Control in Partially Observable Complex Social Systems

.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

Distributed Reinforcement Learning for Cooperative Multi-Robot Object Manipulation

Similar lecture

Optimal Control in Partially Observable Complex Social Systems

Downloads