Fit Without Fear - An Over-Fitting Perspective on Modern Deep and Shallow Learning
Offered By: MITCBMM via YouTube
Course Description
Overview
Syllabus
Intro
Supervised ML
Interpolation and Overfitting
Modern ML
Fit without Fear
Overfitting perspective
Kernel machines
Interpolation in kernels
Interpolated classifiers work
what is going on?
Performance of kernels
Kernel methods for big data
The limits of smooth kernels
Eigenpro: practical implementation
Comparison with state-of-the-art
Improving speech intelligibility
Stochastic Gradient Descent
The Power of Interpolation
Optimality of mini-batch size 1
Minibatch size?
Real data example
Learning kernels for parallel computation?
Theory vs practice
Model complexity of interpolation?
How to test model complexity?
Testing model complexity for kernels
Levels of noise
Theoretical analyses fall short
Simplicial interpolation A-fit
Nearly Bayes optimal
Parting Thoughts I
Taught by
MITCBMM
Related Courses
機器學習技法 (Machine Learning Techniques)National Taiwan University via Coursera Utilisez des modèles supervisés non linéaires
CentraleSupélec via OpenClassrooms Statistical Machine Learning
Eberhard Karls University of Tübingen via YouTube Interplay of Linear Algebra, Machine Learning, and HPC - JuliaCon 2021 Keynote
The Julia Programming Language via YouTube Interpolation and Learning With Scale Dependent Kernels
MITCBMM via YouTube