Using generative adversarial networks to develop a realistic human behavior simulator

Research output: Chapter in Book / Report / Conference proceedingConference contributionAcademicpeer-review

Abstract

Simulation environments have proven to be very useful as testbeds for reinforcement learning (RL) algorithms. For settings where an actual human user is involved, these simulation environments allow one to test out the suitability of new RL approaches without having to include real users at first. It obviously does require the simulator to have a certain degree of realism, however, realistic simulators for the behavior of humans in the health domain are rarely seen. To generate realistic behavior, the simulator could be driven by data from real users, but this might lead to privacy issues. In this paper, we propose to use Generative Adversarial Networks (GANs) for generating realistic simulation environments. In this first step, we use an existing simulator that simulates daily activities of users and the GANs are used to generate realistic sensory data that accompanies such activities. After training, the original (potentially privacy sensitive) data can be thrown away and the simulator can simply be driven by the GAN models. Results show that a model trained on real data shows similar performance on the data artificially generated by the GAN.

Original languageEnglish
Title of host publicationPRIMA 2018 Principles and Practice of Multi-Agent Systems
Subtitle of host publication21st International Conference, Tokyo, Japan, October 29-November 2, 2018, Proceedings
EditorsNir Oren, Yuko Sakurai, Itsuki Noda, Tran Cao Son, Tim Miller, Bastin Tony Savarimuthu
PublisherSpringer/Verlag
Pages476-483
Number of pages8
ISBN (Electronic)9783030030988
ISBN (Print)9783030030971
DOIs
Publication statusPublished - 2018
Event21st International Conference on Principles and Practice of Multi-Agent Systems, PRIMA 2018 - Tokyo, Japan
Duration: 29 Oct 20182 Nov 2018

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume11224 LNAI
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference21st International Conference on Principles and Practice of Multi-Agent Systems, PRIMA 2018
CountryJapan
CityTokyo
Period29/10/182/11/18

Fingerprint

Human Behavior
Simulator
Simulators
Simulation Environment
Reinforcement learning
Reinforcement Learning
Privacy
Generative Models
Testbeds
Testbed
Learning algorithms
Network Model
Learning Algorithm
Health
Human

Keywords

  • Simulation
  • Generative adversarial networks
  • Reinforcement learning
  • Deep learning
  • e-Health

Cite this

el Hassouni, A., Hoogendoorn, M., & Muhonen, V. (2018). Using generative adversarial networks to develop a realistic human behavior simulator. In N. Oren, Y. Sakurai, I. Noda, T. Cao Son, T. Miller, & B. T. Savarimuthu (Eds.), PRIMA 2018 Principles and Practice of Multi-Agent Systems: 21st International Conference, Tokyo, Japan, October 29-November 2, 2018, Proceedings (pp. 476-483). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11224 LNAI). Springer/Verlag. https://doi.org/10.1007/978-3-030-03098-8_32
el Hassouni, Ali ; Hoogendoorn, Mark ; Muhonen, Vesa. / Using generative adversarial networks to develop a realistic human behavior simulator. PRIMA 2018 Principles and Practice of Multi-Agent Systems: 21st International Conference, Tokyo, Japan, October 29-November 2, 2018, Proceedings. editor / Nir Oren ; Yuko Sakurai ; Itsuki Noda ; Tran Cao Son ; Tim Miller ; Bastin Tony Savarimuthu. Springer/Verlag, 2018. pp. 476-483 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
@inproceedings{86ca692812814b9ca1e8bf10675e2a4c,
title = "Using generative adversarial networks to develop a realistic human behavior simulator",
abstract = "Simulation environments have proven to be very useful as testbeds for reinforcement learning (RL) algorithms. For settings where an actual human user is involved, these simulation environments allow one to test out the suitability of new RL approaches without having to include real users at first. It obviously does require the simulator to have a certain degree of realism, however, realistic simulators for the behavior of humans in the health domain are rarely seen. To generate realistic behavior, the simulator could be driven by data from real users, but this might lead to privacy issues. In this paper, we propose to use Generative Adversarial Networks (GANs) for generating realistic simulation environments. In this first step, we use an existing simulator that simulates daily activities of users and the GANs are used to generate realistic sensory data that accompanies such activities. After training, the original (potentially privacy sensitive) data can be thrown away and the simulator can simply be driven by the GAN models. Results show that a model trained on real data shows similar performance on the data artificially generated by the GAN.",
keywords = "Simulation, Generative adversarial networks, Reinforcement learning, Deep learning, e-Health",
author = "{el Hassouni}, Ali and Mark Hoogendoorn and Vesa Muhonen",
year = "2018",
doi = "10.1007/978-3-030-03098-8_32",
language = "English",
isbn = "9783030030971",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
publisher = "Springer/Verlag",
pages = "476--483",
editor = "Nir Oren and Yuko Sakurai and Itsuki Noda and {Cao Son}, Tran and Tim Miller and Savarimuthu, {Bastin Tony}",
booktitle = "PRIMA 2018 Principles and Practice of Multi-Agent Systems",

}

el Hassouni, A, Hoogendoorn, M & Muhonen, V 2018, Using generative adversarial networks to develop a realistic human behavior simulator. in N Oren, Y Sakurai, I Noda, T Cao Son, T Miller & BT Savarimuthu (eds), PRIMA 2018 Principles and Practice of Multi-Agent Systems: 21st International Conference, Tokyo, Japan, October 29-November 2, 2018, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 11224 LNAI, Springer/Verlag, pp. 476-483, 21st International Conference on Principles and Practice of Multi-Agent Systems, PRIMA 2018, Tokyo, Japan, 29/10/18. https://doi.org/10.1007/978-3-030-03098-8_32

Using generative adversarial networks to develop a realistic human behavior simulator. / el Hassouni, Ali; Hoogendoorn, Mark; Muhonen, Vesa.

PRIMA 2018 Principles and Practice of Multi-Agent Systems: 21st International Conference, Tokyo, Japan, October 29-November 2, 2018, Proceedings. ed. / Nir Oren; Yuko Sakurai; Itsuki Noda; Tran Cao Son; Tim Miller; Bastin Tony Savarimuthu. Springer/Verlag, 2018. p. 476-483 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11224 LNAI).

Research output: Chapter in Book / Report / Conference proceedingConference contributionAcademicpeer-review

TY - GEN

T1 - Using generative adversarial networks to develop a realistic human behavior simulator

AU - el Hassouni, Ali

AU - Hoogendoorn, Mark

AU - Muhonen, Vesa

PY - 2018

Y1 - 2018

N2 - Simulation environments have proven to be very useful as testbeds for reinforcement learning (RL) algorithms. For settings where an actual human user is involved, these simulation environments allow one to test out the suitability of new RL approaches without having to include real users at first. It obviously does require the simulator to have a certain degree of realism, however, realistic simulators for the behavior of humans in the health domain are rarely seen. To generate realistic behavior, the simulator could be driven by data from real users, but this might lead to privacy issues. In this paper, we propose to use Generative Adversarial Networks (GANs) for generating realistic simulation environments. In this first step, we use an existing simulator that simulates daily activities of users and the GANs are used to generate realistic sensory data that accompanies such activities. After training, the original (potentially privacy sensitive) data can be thrown away and the simulator can simply be driven by the GAN models. Results show that a model trained on real data shows similar performance on the data artificially generated by the GAN.

AB - Simulation environments have proven to be very useful as testbeds for reinforcement learning (RL) algorithms. For settings where an actual human user is involved, these simulation environments allow one to test out the suitability of new RL approaches without having to include real users at first. It obviously does require the simulator to have a certain degree of realism, however, realistic simulators for the behavior of humans in the health domain are rarely seen. To generate realistic behavior, the simulator could be driven by data from real users, but this might lead to privacy issues. In this paper, we propose to use Generative Adversarial Networks (GANs) for generating realistic simulation environments. In this first step, we use an existing simulator that simulates daily activities of users and the GANs are used to generate realistic sensory data that accompanies such activities. After training, the original (potentially privacy sensitive) data can be thrown away and the simulator can simply be driven by the GAN models. Results show that a model trained on real data shows similar performance on the data artificially generated by the GAN.

KW - Simulation

KW - Generative adversarial networks

KW - Reinforcement learning

KW - Deep learning

KW - e-Health

UR - http://www.scopus.com/inward/record.url?scp=85056465098&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85056465098&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-03098-8_32

DO - 10.1007/978-3-030-03098-8_32

M3 - Conference contribution

SN - 9783030030971

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 476

EP - 483

BT - PRIMA 2018 Principles and Practice of Multi-Agent Systems

A2 - Oren, Nir

A2 - Sakurai, Yuko

A2 - Noda, Itsuki

A2 - Cao Son, Tran

A2 - Miller, Tim

A2 - Savarimuthu, Bastin Tony

PB - Springer/Verlag

ER -

el Hassouni A, Hoogendoorn M, Muhonen V. Using generative adversarial networks to develop a realistic human behavior simulator. In Oren N, Sakurai Y, Noda I, Cao Son T, Miller T, Savarimuthu BT, editors, PRIMA 2018 Principles and Practice of Multi-Agent Systems: 21st International Conference, Tokyo, Japan, October 29-November 2, 2018, Proceedings. Springer/Verlag. 2018. p. 476-483. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). https://doi.org/10.1007/978-3-030-03098-8_32