Synchronisms Using Reinforcement Learning as an Heuristic

Abstract

The concept of Synchronism describes circumstances that appear meaningfully related yet nevertheless lack a casual connection. Every day we make use of this concept, from forming new relationships to grooming a new set of skills. Therefore one day in the future we will be able to make use of them for a specific goal. Reinforcement Learning (RL) is a type of learning framework where the solution is presented as a Markov Decision Process (MDP) where the goal is to learn by trial and error, emulating how the human begins to learn. With this type of framework, we can solve very complex non-linear and not deterministic problems. In this novel method, we train the model using a non-deterministic RL and try to find the causal link between those circumstances by observation. This is known as the “Keen Eye” where each circumstance finds a consensus link affecting the outcome using the knowledge from each circumstance involved. We trained agents in two experiments with different domains, the first experiment tries to solve a linear non-deterministic problem, while the second is a non-linear and non-deterministic simulation. With this type of framework, the most common metric is to gather the reward value obtained from the outcome of the experiments, therefore we use this value to measure how performance behaves. Finally, with the results of experiments we discovered that compared with traditional RL, this method can train RL models faster, by obtaining better rewards earlier and makes the RL less susceptible to fall into a local minimum.