The Joint Accelerator Conferences Website (JACoW) is an international collaboration that publishes the proceedings of accelerator conferences held around the world.
TY - CONF AU - Shin, Shin, S.W. AU - Chai, J.-S. AU - Ghergherehchi, M. ED - Koscielniak, Shane ED - Satogata, Todd ED - Schaa, Volker RW ED - Thomson, Jana TI - Using Deep Reinforcement Learning for Designing Sub-Relativistic Electron Linac J2 - Proc. of IPAC2018, Vancouver, BC, Canada, April 29-May 4, 2018 C1 - Vancouver, BC, Canada T2 - International Particle Accelerator Conference T3 - 9 LA - english AB - Generally, when designing an accelerator device, the design is based on the experience and knowledge of the designer. Most of the design process proceeds by chang-ing the parameters and looking at the trends and then determining the optimal values. This process is time-consuming and tedious. In order to efficiently perform this tedious design process, a method using an optimization algorithm is used. Recently, many people started to get interested in the algorithm used in AlphaGo, which became famous when it won the professional Go player developed by google The algorithm used in AlphaGo is an algorithm called reinforcement learning that learns how to get optimal reward in various states by moving around a solution space that the agent has not told beforehand. In this paper, we will discuss about designing an particle accelerator by applying Deep Q-network algorithm which is one kind of deep learning reinforcement learning. PB - JACoW Publishing CP - Geneva, Switzerland SP - 4720 EP - 4722 DA - 2018/06 PY - 2018 SN - 978-3-95450-184-7 DO - 10.18429/JACoW-IPAC2018-THPML032 UR - http://jacow.org/ipac2018/papers/thpml032.pdf ER -