What is a line of “best” fit? A framework for building predictive models

Zoom link for remote participants

Abstract: In this talk, we will introduce a three-step process that we will use to understand the inner workings of linear regression, to demystify common software implementations. This process involves choosing a model, choosing a loss function, and minimizing average loss to fit our chosen model to our data. We will explore a few related models – the constant model, the simple linear regression model, and the multiple linear regression model – and understand the role of the squared loss function. We will also explore the use of Python and numerical optimization for determining the “best” model and for making predictions on unseen data.

Bio: Suraj Rampure is a Lecturer in the Halıcıoğlu Data Science Institute at UC San Diego. He teaches undergraduate courses in programming, statistical inference, and the theory and practice of machine learning. He also coordinates the data science major’s senior capstone program. Prior to arriving at UC San Diego, he completed BS and MS degrees in Electrical Engineering and Computer Sciences at UC Berkeley, where he taught for ten semesters as a teaching assistant and instructor of record in computer science and data science. At Berkeley, he received the EECS Outstanding Graduate Student Instructor award in 2018 and the campus-wide Extraordinary Teaching in Extraordinary Times award in 2021.