Very very interesting! I think this is a typical Class III problem of emergent synthesis.
I strongly wish that you could conduct experiments with human subjects as future work.
Dear Mr. Mizuyama, thank you for your very interesting presentation which relates to a quite real industrial context such as the steel making sector. The verification of the numerical results with direct involvement and evaluation o fof human activities is going to be very relevant and adding interest to the study. Could you suggest some other industrial context or situation appropriate for your reinforcement learning approach?
I believe the proposed reinforcement approach will be useful in various contexts. For instance, many production control decisions, such as task dispatching on the shop floor, are made online by human decision makers, and those are all possible targets of the approach. However, I see a higher potential value in the service domain. This is because the operational decisions in service providing contest tend not only to be more dynamic but also to be related more to the hidden preference of human (customers, for example). What information is helpful to properly estimate on the fly the hidden preference is usually difficult to formally describe but (I hope) this “tacit knowledge” may be (at least partly) revealed by the proposed approach or a like.
Very very interesting! I think this is a typical Class III problem of emergent synthesis.
I strongly wish that you could conduct experiments with human subjects as future work.
Thank you very much for your feedback!
Yes, and, among a lot to be done, I am especially interested in testing our working hypothesis:
“a cognitive framework suitable for a human controller also makes it easy for the reinforcement learning agent to learn an effective policy”
through human-subject experiments using the same game model.
Dear Mr. Mizuyama, thank you for your very interesting presentation which relates to a quite real industrial context such as the steel making sector. The verification of the numerical results with direct involvement and evaluation o fof human activities is going to be very relevant and adding interest to the study. Could you suggest some other industrial context or situation appropriate for your reinforcement learning approach?
Thank you very much Prof. Teti for your feedback!
I believe the proposed reinforcement approach will be useful in various contexts. For instance, many production control decisions, such as task dispatching on the shop floor, are made online by human decision makers, and those are all possible targets of the approach. However, I see a higher potential value in the service domain. This is because the operational decisions in service providing contest tend not only to be more dynamic but also to be related more to the hidden preference of human (customers, for example). What information is helpful to properly estimate on the fly the hidden preference is usually difficult to formally describe but (I hope) this “tacit knowledge” may be (at least partly) revealed by the proposed approach or a like.