Parameter setting apparatus, non-transitory medium storing computer program, and parameter setting method

A computer-readable, non-transitory medium storing a computer program for setting a control parameter for a wireless communication network in response to a state variable of said wireless communication network, wherein

• said computer program causes a computer to execute a process which, through reinforcement learning, learns which of a plurality of optimizations, each for optimizing said control parameter, is to be executed to optimize said control parameter, and
• said optimizations form a group of optimizations that affect the same state variable used to determine a value function of reinforcement learning or a group of optimizations whose control parameters overlap each other, and wherein
• said computer program causes said computer to execute:
• a process in which a first agent program allocated on a first group of optimizations learns, through reinforcement learning, a first value function for determining which of said optimizations in said first group is to be executed;
• a process in which said first agent program, using said first value function and based on the state of said wireless communication network, determines which of said optimizations in said first group is to be executed as a first optimization, and activates said first optimization;
• a process in which a second agent program allocated on a second group of optimizations learns, through reinforcement learning, a second value function for determining which of said optimizations in said second group is to be executed;
• a process in which said second agent program, using said second value function and based on the state of said wireless communication network, determines which of said optimizations in said second group is to be executed as a second optimization, and activates said second optimization;
• a process in which a third agent program learns, through reinforcement learning, a third value function for determining whether, among said optimizations for optimizing the control parameters of said wireless communication network, an optimization that affects said first and second value functions is to be executed or not;
• a process in which said third agent program, using said third value function and based on the state of said wireless communication network, determines whether said optimization that affects said first and second value functions is to be executed or not and, depending on a result of said determination, activates a third optimization;
• and
• a process in which, when said third agent program activates said third optimization, the activation of said first and second optimization is stopped.

A parameter setting method for setting a control parameter for a wireless communication network in response to a state variable of said wireless communication network, wherein

• said parameter setting method organizes a plurality of optimizations, each for optimizing said control parameter, into a group of optimizations that affect the same state variable used to determine a value function of reinforcement learning that is performed to learn which of said plurality of optimizations is to be executed to optimize said control parameter, or into a group of optimizations whose control parameters overlap each other, and
• said parameter setting method causes a computer to execute:
• a process in which a first agent program allocated on a first group of optimizations learns, through reinforcement learning, a first value function for determining which of said optimizations in said first group is to be executed;
• a process in which said first agent program, using said first value function and based on the state of said wireless communication network, determines which of said optimizations in said first group is to be executed as a first optimization, and activates said first optimization;
• a process in which a second agent program allocated on a second group of optimizations learns, through reinforcement learning, a second value function for determining which of said optimizations in said second group is to be executed;
• a process in which said second agent program, using said second value function and based on the state of said wireless communication network, determines which of said optimizations in said second group is to be executed as a second optimization, and activates said second optimization;
• a process in which a third agent program learns, through reinforcement learning, a third value function for determining whether, among said optimizations for optimizing the control parameters of said wireless communication network, an optimization that affects said first and second value functions is to be executed or not;
• a process in which said third agent program, using said third value function and based on the state of said wireless communication network, determines whether said optimization that affects said first and second value functions is to be executed or not and, depending on a result of said determination, activates a third optimization;
• and
• a process in which, when said third agent program activates said third optimization, the activation of said first and second optimization is stopped.