A Research on Defects and Improvements to Multi-Armed Slots

Abstract

In recent years, various large language models have sprung up, which are constantly being trained and updated. Most of the time, these models can provide answers more accurately. However, in some specific cases, the answers generated by training on a large corpus of non-emotion text may not be correct. This paper proposes a feasible method to reduce the sum of squares as much as possible so that the fitting function is closer to the true value. By adding the concept of emotion to deep learning, combined with the fusion model of multi-layer linear regression and neural network learning, it was found that adding emotion is more helpful to the language model through multiple trainings, so that the "arm" pulled by the AI each time gets a larger expectation than the original, which significantly improves the reliability and provides a new idea and direction for the problem in this direction.