Save 37% off PRO during our Black Friday Sale! »

Själv introduktion

1852ac80648a76e2a64589c7d6ee75c3?s=47 hajimizu
April 17, 2019

Själv introduktion

Introduction of myself and my research.

1852ac80648a76e2a64589c7d6ee75c3?s=128

hajimizu

April 17, 2019
Tweet

Transcript

  1. Själv introduktion April 17th, 2019 Hajime Mizuyama hajimizu@kth.se 1

  2. Academic background Typical topics addressed in MSE Production scheduling and

    simulation, Production system design and facility layout, Process analysis and optimization, etc. Mathematical bases Applied operations research, Applied statistics and machine learning, Discrete event systems, Engineering economics, etc. Recent interests Serious games, Participatory simulation, Human computation, Game theory and mechanism design, Multi-agent systems, (Deep) reinforcement learning, etc. Software and programming languages Python/Django, R, Processing/p5.js, etc. 2 (shifted/expanded to human aspects)
  3. Humans will act differently according to Situational and procedural knowledge

    • Granularity of situational awareness • Forecasting models and decision heuristics What actually are the perceptual and action skills? How to support learning and performing the skills? Preference of one’s own and others’ • Objective function, goals, targets, directions, etc. • Strategic interaction with others How to align Incentives towards an overall objective? How to incentivize cooperation/coordination? 3
  4. Not only manufacturing systems but also Application fields of interest

    • I am currently interested not only in manufacturing systems or factories, but also in service providing systems, supply chains, etc. • In these fields, in a sense, humans and machines (i.e. computers) are collaboratively deliver values to customers. Aims and goals of research • To deepen the understanding on, and to support enhancing the performance of, such human involved collaboration systems through a combination of behavioral and mathematical approaches. • To improve and refine the combination approach/perspective. 4
  5. Research directions 5 Skills support/development Cooperation/coordination Skills analysis Incentive alignment

    Collaboration systems
  6. Example behavioral/mathematical tools 6 Skills support/development Pedagogical games Inverse RL

    Cooperation/coordination Participatory simulation Multi-agent simulation Simulation games Reinforcement learning Skills analysis Experimental economics Game theory Incentive alignment Collaboration systems
  7. Our approach: Still immature and evolving What model? • Agent-based

    model, which first models each of the humans in the system as an incentivized agent with skills, and study the system- level behavior from the interactions among the agents. Why? • The system’s overall behavior can be related to the actions and their background rationale of each participant. • It will become straightforward to relate serious game and computer simulation. Each player simply corresponds to a computer agent. 7 (a) Construct a mathematical model of the collaboration system
  8. Model outline: How the system rolls Humans can observe (some

    portion of) the system status, and will act on condition of the observation (to achieve their own goals). 8 System status: s0 Observe System status: s1 Intervene System status: s2 System status: s3 System status: s5 Communication among the agents may also be incorporated Each individual is modelled as an incentivized agent with skills
  9. Model outline: How each agent works 9 Heuristic/strategy in a

    dynamic task environment Decision, action Input variables on how current situation is captured The role can be captured as a transformation from input to output. How to define input/output spaces, timing, etc. might also be a key.
  10. Our approach: Still immature and evolving 10 (a) Construct a

    mathematical model of the collaboration system (d) Collect and analyze the play logs of game (b) (e) Collect and analyze the results of simulation (c) (f) Relate and compare (d) and (e), and draw some findings (b) Develop a serious game based on (a) (c) Develop a computer simulation model based on (a) Refine (a) according to (f) and repeat the analysis Utilize the findings in (f) (f) Relate and compare (d) and (e), and draw some findings
  11. How to combine two aspects #1 Behavioral analysis • Observe

    humans’ behavior mainly in the game world Mathematical analysis • Need to rely on computer simulation in many cases 11 Hypotheses on the skills Behavioral experiments As-expected and surprising results Computer experiments
  12. How to combine two aspects #2 Behavioral analysis • Observe

    humans’ behavior mainly in the game world Mathematical analysis • Need to rely on computer simulation in many cases 12 Behavioral experiments Formalize/general ize/quantify Patterns/rules Computer experiments
  13. Task environment classification: Tentative Single-agent setting Multi-agent setting 13 Simple

    Complex
  14. Mathematical bases for modelling Single-agent setting Multi-agent setting 14 Simple

    Complex Advanced RL models, such as DQN Simple MDP and reinforcement learning Periodical optimization Multi-agent reinforcement learning Mechanism design Distributed periodical optimization Game theory models Evolutionary game models
  15. Recent cases Single-agent setting Multi-agent setting 15 Simple Complex Market

    research (How to elicit preference truthfully?) In-house supply chain of a large steel manufacturer (ColPMan) Food supply chain (How to reduce food loss/waste?) A restaurant (Trade-off between efficiency and CS) Crane operation task in a factory shop floor A restaurant (Trade-off between efficiency and CS) Crane operation task in a factory shop floor Knowledge sharing in a sales department
  16. Factory shop floor example In a factory shop floor, different

    kinds of jobs arrive randomly, and a due date is assigned to each of them. An operator needs to handle the jobs with a crane between multiple LIFO buffers and a machine. High production rate and low due date lateness penalty should be accomplished. The performance depends heavily on the skills of the operator. 16
  17. Restaurant example #1 17 Waiting Room Dining Hall

  18. Restaurant example #2 18 • Six tables (squares) are handled

    by a single waitperson (a blue circle). • The color of each table and dish changes from white to yellow, orange, and finally red as waiting time passes.
  19. To-do list for the cases • Refine the visual design

    and operability of the games before releasing them to the public on the web. • Develop and test various data mining and machine learning approaches for game log data analysis. • Generalize the games and simulation models to multi-agent setting. • Collect information on recent advanced RL models, and incorporate them to the simulation models if useful. • Collect information on inverse RL models, and examine how they can be utilized for designing pedagogical aspects of the games. 19
  20. What I wound like to do here • Learn what

    sort of problems and issues are there in the healthcare logistics field. • Discuss whether some of the topics may be framed in a similar perspective, and addressed through a similar approach. • More specifically; – Agent-based modelling of some health care logistics problems – Serious game development and testing based on the models – Simulation study based on the models • If there are, let’s actually do this together! 20
  21. Summary • I am interested in human involved collaboration systems,

    and addressing them through a combination approach/perspective of behavioral and mathematical. • The combination approach is characterized by agent-based modelling framework, which makes it straightforward to relate humans’ game play and computer simulation. • What I have proposed here is only a plan, and I am flexible and open to your suggestions about fruitful collaboration. • I would be delighted if anyhow a first step towards lasting collaboration can be taken in the period of my stay. 21
  22. Thank you! Questions & Comments are welcome 22