Short course on the development of open-source machine learning packages
Tuesday, April 04, 2023
Many researchers now release experiment codes used for their papers so that others can reproduce results. However, experiment code is different from software, for which the primary purpose is to serve users broadly. In fact, developing a sound and easy-to-use machine learning package is always challenging. It involves issues ranging from algorithms, implementations and many design considerations. In this short course, based on my past experiences in developing several popular machine learning packages, we discuss issues including the start of a project, the choices of supported functionalities, the identification of research problems from user feedback, and others. At the end of the course, we check the challenges and concerns in designing future machine learning software. The discussion is biased towards small-scale software done in a research group, though most results still apply to large industry-scale projects.
Post Talk Link: Click Here
Passcode: KcDd$m5R
Speaker/s
Chih-Jen Lin is an affiliated professor at MBZUAI. He is also a distinguished professor at National Taiwan University. In the past, he developed several widely used machine learning packages. For example, in the hay day of SVM (support vector machines), his LIBSVM was a package that almost everybody used. His latest efforts are on the package LibMultiLabel for multi-class and multi-label document classification. He is an IEEE fellow, a AAAI fellow, and an ACM fellow for his contribution to machine learning algorithms and software design. More information about him can be found at http://www.csie.ntu.edu.tw/~cjlin.
Related
Polygenic Score Modeling to Investigate Genotype-Phenotype Associations
Read More
Trustworthy Machine Learning: Transparency, Collaboration, and Evaluation
Read More