Skip to main content
— Presentation at ICML 2016 Workshop —
July 11, 2016
Machine Learning plays important role to tackle the problem that the number of experts is decreasing due to falling birth rate and the aging population which causes low efficient service quality of social infrastructure operated by non-experts. The 33rd International Conference on Machine Learning (ICML), which is one of the top conferences in machine learning research field, was held in New York City, USA during 19-24 June 2016. More than 3000 participants, almost twice more than last year, discussed latest machine learning research activities. It made me feel machine learning is getting a lot more attention nowadays over the world.
Fig. 1 Proposed method
In the ICML Workshop "Reliable Machine Learning in the Wild", we presented our research about reinforcement learning for operation management of social infrastructure, the title was "Evaluation of Multi-armed Bandit for Automatic Operation Management".
To operate social infrastructure, we need to make suitable decision considering current situation. In case of ATM (Automatic Teller Machine), loading enough amounts of cash to satisfy customers' demand of withdraws and deposit is necessary. On the other hand, loading cash costs logistics cost and others. Operators have to decide amounts of cash inside ATMs considering the balance of satisfying customers' demand and reducing cost. To support this decision making, we proposed and evaluated a method to utilize Multi-armed Bandit which is a part of reinforcement learning with current status. As a result, figure 1 shows our proposed method works well and depends on how to design reward signal which cannot be observed qualitatively like failing to satisfy customers' demand.
We plan to apply reinforcement learning technologies for operation management of social infrastructure and self-learning service robots.
(By AKIYAMA Takayuki)