Homotopy-Based Learning Algorithm

Mitchell noted that learning is a computer program acts to improve the processingperformance of a task through experience. We can study machine learning in the mappingspace. In recent years, homotopy as one of the tools to study the issue of continuity, it hasbeen widely used in many areas, such as image analysis. In this paper, we use homotopy todo research in the machine learning. Our main work is as follws:1. For the continuous transformation of the sample feature, we proposed thehomotopy-based feature learning algorithm to extract the homotopy invariance of thesample so that it can strengthen the robustness of the classifier to continuoustransformation of the sample.2. For the continuous transformation between the objective functions of learning, weproposed multi-tasking homotopy learning algorithm, combined with multi-taskinglearning framework to reduce the demand for training samples without lowering therecognition rate.3. We apply the homotopy-based feature learning algorithm and the multi-taskinghomotopy learning algorithm in face recognition to verify their validity.