Algoritmos de aprendizaje de máquina para conjuntos de datos reducidos con aprendizaje por reforzamiento y optimización

As computers evolve and their computational complexity and performance increase, many forms of machine learning have found new implementations and applications to take advantage of these computer features. Previously, areas dominated exclusively by purpose-built software or fully manual applications have found newly developed innovation from Machine Learning development. Reinforcement Learning is a way to implement some algorithms that are being explored. Working in a similar way to teaching a young child to perform a task by giving positive or negative feedback depending on their actions, this form of machine learning has great potential with some disadvantages. The results can sometimes greatly outperform computational deterministic implementations. Being an unexplored area, this investigation has the purpose of implementing and comparing multiple machine learning algorithms with a reinforcement learning optimization algorithm to find out the advantages and disadvantages of each one. Most of the research involves a reinforcement learning Intel project that would benefit greatly from the results of this investigation and will also suggest that there are certain restrictions that the algorithm implementations must follow. It is also the goal of the investigation to find which algorithm is the most optimal for the Intel project mentioned. Benchmarking and thorough testing will be done with a real system and implementation of each algorithm to compare objective data during the investigation, as well as an analysis of advantages and harmful properties of each algorithm. Results will be published in the last sections of the document with as much detail as possible without giving confidential information from Intel’s intellectual property.