Introduction to Numerical Regression
by Dr. Ranjeet Tate
Numerical regression – finding the quantitative relationship between variables – is at the core of machine learning. This class is intended for students who want to understand machine learning from the inside out, and by its end participants should have a solid understanding of the basic statistical techniques that underpin modern machine learning.
All classes will be taught in person in the Principia Mathematica room at the Academia.edu offices, at California and Kearny in downtown San Francisco.
The course will last eight weeks, with one class meeting per week from 6:30pm to 9:30pm. If you have to miss a class, we'll make sure you're able to catch up, either by scheduling a one-one meeting or sharing a video of the class.
Additional, students may opt to augment the course with weekly one on one tutorials to get more detailed individual feedback and instruction.
We will be covering the following topics:
- 1-dimensional linear models
- Form of data
- Error "vector"
- Target function for "best fit"
- Manual minimization in a spreadsheet
- Closed form solution for Ordinary Least Squares (OLS)
- Propagation of errors (errors in a function of stochastic variables)
- Standard Error (SE) in the parameters
- Confidence interval
- Prediction interval
- Higher dimensions, closed form solution for OLS
- Linear regression for a non-linear model
- Underfitting, overfitting, SE in the parameters
- Training Data vs Test Data
- Exponential model, power law, logistic model
- Transformations on the data
- Other "norms" on the error vector
- Numerical approaches to optimization
- Gradient descent
- Problems with gradient descent
- Second order methods: Newton-Raphson
- Importance of phenomenological models
Participants should have a solid understanding of high-school mathematics. Some differential calculus is preferable but not essential, as is some familiarity with linear algebra.
580 California Street
San Francisco, CA 94104
Tue, Sep 4th, 2018, 6:30pm –
Tue, Oct 23rd, 2018, 9:30pm.
Attendance is free for students who subscribe to an EDeeU Education Director of Studies.