Takashi Kuremoto, Takaomi Hirata, Masanao Obayashi, Shingo Mabu and Kunikazu Kobayashi (April 3rd 2019). Training Deep Neural Networks with Reinforcement Learning for Time Series Forecasting, Time Series Analysis - Data, Methods, and Applications, Chun-Kit Ngan, IntechOpen, DOI: 10.5772/intechopen.85457. 

New Zealand

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 &quot;&quot;Proximal Policy Optimization&quot;&quot; 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.


AAMAS 2020

Deep reinforcement learning for large-scale epidemic control

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). 

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.


technical paper

Reward Redistribution Mechanism in Multi-Agent Reinforcement Learning

Option discovery and skill acquisition frameworks are integral to the functioning of a hierarchically organized Reinforcement learning agent. However, such techniques often yield a large number of options or skills, which can be represented succinctly by filtering out any redundant information. Such a reduction can decrease the required computation while also improving the performance on a target task. To compress an array of option policies, we attempt to find a policy basis that accurately captures the set of all options. In this work, we propose Option Encoder, an auto-encoder based framework with intelligently constrained weights, that helps discover a collection of basis policies. The policy basis can be used as a proxy for the original set of skills in a suitable hierarchically organized framework. We demonstrate the efficacy of our method on a collection of grid-worlds and Deepmind-lab by evaluating the obtained policy basis on downstream tasks.


Option Encoder : A framework for Discovering a Policy Basis in Reinforcement Learning

Electric power sharing among households based on the bidding method is studied as a future service. In order to verify the feasibility of such a service, a new multi-agent simulation model has been designed. We validated this model through some evaluations. For example, it is confirmed that the market price on this service stably changes according to the supply-demand balance between both sold and purchased bid volumes. In addition to that, the results of the household profit and contract rate of this service showed that the design for bid strategies works as intended in most cases.


Design and Evaluations of Multi-Agent Simulation Model for Electric Power Sharing among Households

We present a reinforcement learning-based approach for generating a timetable for trains traversing through a railway station.
We demonstrate how the reinforcement learning approach is used to learn heuristics, which in turn are used to generate a train movement plan.
We compare the quality of the plan generated using ‘rules used in practice’ and the plan generated by the ‘heuristics learned using the RL approach’, under different congestion scenarios
"


Reinforcement Learning for Train Movement Planning at Railway Stations

We study proportionality in approval-based multi- winner elections with a variable number of winners, where both the size and identity of the winning committee are informed by voters’ opinions. While proportionality has been studied in multi- winner elections with a fixed number of winners, it has not been considered in the variable number of winners setting. The measure of proportionality we consider is average satisfaction (AS), which intuitively measures the number of agreements on aver- age between sufficiently large and cohesive groups of voters and the output of the voting rule. First, we show an upper bound on the AS that any deterministic rule can provide, and that straightforward adaptations of deterministic rules from the fixed number of winners setting do not achieve better than a 1/2 approximation to AS even for large numbers of candidates. We then prove that a natural randomized rule achieves a 29/32 approximation to AS.


Proportionality in Approval-Based Elections With a Variable Number of Winners

Condorcet extensions have long held a prominent place in social choice theory. A Condorcet extension will return the Condorcet winner as the unique winner whenever such an alternative exists. However, the definition of a Condorcet extension does not take into account possible manipulation by the voters. A profile where all agents vote truthfully may have a Condorcet winner, but this alternative may not end up in the set of winners if agents are acting strategically. Focusing on the class of tournament solutions, we show that many natural social choice functions in this class, such as the well-known Copeland and Slater rules, cannot guarantee the preservation of Condorcet winners when agents behave strategically. Our main result in this respect is an impossibility theorem that establishes that no tournament solution satisfying a very weak decisiveness requirement can provide such a guarantee. On the bright side, we identify several indecisive but otherwise attractive tournament solutions that do guarantee the preservation of Condorcet winners under strategic manipulation.


Preserving Condorcet Winners under Strategic Manipulation

We present MsATL: the first tool for deciding the satisfiability of Alternating-time Temporal Logic (ATL) with imperfect information. MsATL combines SAT Modulo Monotonic Theories solvers with existing ATL model checkers: MCMAS and STV. The tool can deal with various semantics of ATL, including perfect and imperfect information, and can handle additional practical requirements. MsATL can be applied for synthesis of games that conform to a given specification, with the synthesised game often being minimal.


MsATL: a Tool for SAT-Based ATL Satisfiability Checking

Mechanism Design for School Choice with Soft Diversity Constraints


In this paper, we study the two-sided matching problem with soft diversity constraints in which each student belongs to one type and each school imposes soft targets on each type. We first identify limitations of type-specific quotas in a previous model and introduce a new general model that takes different levels of importance of types into account. Then we propose a new algorithm that yields a non-wasteful and fair outcome with respect to different levels of importance.


Multiple Levels of Importance in Matching with Distributional Constraints

Algorithms and machine learning models, including the decisions made by these models, are becoming ubiquitous in our daily life, including hiring. We make no value judgment regarding this development; rather, we simply acknowledge that it is quickly becoming reality that automation plays a role in hiring. Increasingly, these technologies are used in all of the small decisions that make up the modern hiring pipeline: from which resumes get selected for a first screen to who gets an on site interview. Thus, these algorithms and models may potentially amplify bias and (un)fairness issues for many historically marginalized groups. While there is a rapidly expanding literature on algorithmic decision making and fairness, there has been limited work on fairness specifically for online, multi-stakeholder decision making processes such as those found in hiring. We outline broad challenges including formulating definitions for fair treatment and fair outcomes in hiring, and incorporating these definitions into the algorithms and processes that constitute the modern hiring pipeline. We see the AAMAS community as uniquely positioned to address these challenges.

Downloads

Next from AAMAS 2020

Reward Redistribution Mechanism in Multi-Agent Reinforcement Learning

Similar lecture

Agents are Dead Long Live Agents !

.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

Reward Redistribution Mechanism in Multi-Agent Reinforcement Learning

Similar lecture

Agents are Dead Long Live Agents !

Downloads