- New things make history change! The purpose of this organization is to help people to solve problems and create new things systematically through invenrelation!- Career consulting- Data analytics consulting

• Develop semi-supervised multiple regression model to evaluate users expense behaviors that helps customers to reduce at least 10% of their reimburse costs.• Help to transform Appzen machine learning tasks into microservices implemented by Seldon, Kubernetes with AWS.• Design and implement machine translator by applying transformer attention mechanism. This service helps Appzen to save several thousands (US dollars) per day.• Apply Faster-RCNN to detect graphic symbols of currency and date symbols, the transaction amount and date extraction accuracies are improved from 50% to 80% for Asian-based language, e.g., Chinese, Japanese. • Successful building models for auditor behavior on expense reports with AUC of 0.84. Components that aided in result:o Utilization of second order Markov chain to measure receipt clarity.o Application of BERT embedding to train for semantic similarity between customers’ expense types and Appzen’s expense types. Accuracy of method is improved from 65% to 95%.

• Design and implement collaborative optimizer analytic function library, the time efficiency for distributed machine learning algorithms is improved 4X. • Design distributed model training framework with light data by applying sample size theorem, the training data and time are reduced by 1-2 order of magnitude.

Performance engineer, focusing at Oracle Public Cloud (OPC) networking performance.

o Published 3 books, 50+ journals and conferences, focusing mathematical/statistical modelling and performance evaluation at wireless communications, networking and signal processing, more details can be found in https://sites.google.com/site/websychang/o Predicted IT marketing by utilizing particle swarm optimization (PSO) and the real-valued genetic algorithms (RGA) based on Least squares support vector machine (LS-SVM).o Built a mobile-based learning management system with NoSQL, PHP, R language, which improves the student average grade by 15%. o Proposed Hierarchical SVM to perform multiclass classification with parallel learning algorithm, which can reduce the traditional SVM learning complexity from O(n^3) to O(n^2). o Designed a framework to generate new ideas mathematically, namely, INVENRELATION (sold by http://www.amazon.com/INVENRELATION-Shih-Yu-Chang-book/dp/B00EPF4RRI), guided students to implement following invention methods by Android - Morphological Analysis - Equivalent Transformation - Plurioperad Invention (Chap 5 at INVENRELATION)