Class activities will be recorded here.
This page is the one key spot to look to see what
we have already done, what is planned for the current week, and what
is planned for the remainder of the semester. Since the lectures are
almost entirely built around Jupyter Notebooks you will see links to
these notebooks. These can be easily opened with and one extra mouse
click if you setup your browser to associate `ipynb` files with
Google Colab. Here is a short video where I demonstrate doing this in
Chrome.
My setup example video
The approach is similar in other browsers. The punchliine of the
video is you need a Google ID that you setup once to access
Google
Colaboratory and from then on links to ipynb files will just work.
-
Week 1 : August 22 - August 28
-
-
Tuesday
Introduction
Brief course introduction and a quick review of Python in the context of Jupyter
Thursday
Python Basics
One language dominates modern ML - Python. This lecture is a whirl wind review
-
Week 2 : August 29 - September 4
-
-
Tuesday
Python Basics Continued
Continue Introduction to Python
Thursday
Introduction to Numpy
Numpy is the backbone of much ML programming
-
Week 3 : September 5 - September 11
-
-
Tuesday
Numpy Continued
Continue workking through aspects and useful elements of Numpy
Thursday
Labeled Data
Loading common datasets and understanding the relationship between features and labels
-
Week 4 : September 12 - September 18
-
-
Tuesday
Vectors and Dot Products
Practical motivation from images to motivate basics of understanding vectors/points
Thursday
MatPlotLib
Introduction to visualizing data through plots and histograms using MatPlotLib
-
Week 5 : September 19 - September 25
-
-
Tuesday
Lines, Planes and Hyperplanes
Linear decision boundaries expressed as lines, planes and hyperplanes
Thursday
The Preceptron
One of the first and easiest to understand learning algorithms for linearly separable two class problems
-
Week 6 : September 26 - October 2
-
-
Tuesday
Nearest Neighbor Classifiers
Nearest Neighbor Classifer basics and a start at understanding generalization
Thursday
More Nearest Neighbors
Continued explanation of Nearest Neighbor Classifiers with regression example
-
Week 7 : October 3 - October 9
-
-
Tuesday
High Dimensional kNN and PCA
Showing how Principal Component Analysis may be Coupled with kNN classifiers
Thursday
Bridge from NN to Regression
Using a Nearest Neighbor Strategy to suggest how regression is NOT classification
-
Week 8 : October 10 - October 16
-
-
Tuesday
Introduction to Linear Regression
Fitting a parametric model to sample data to capture structure and make predictions
Thursday
Derivatives, Least-Squares and Outliers
Using linear regression to motivate cover basics of least-squares optimization and outlier detection
-
Week 9 : October 17 - October 23
-
-
Tuesday
Multivariate Linear Regression
Predicting a value from a vector using a linear function with learned parameters
Thursday
Linear Regression with Gradient Descent
Multivariate linear regression is an excellent opportunity to develop deeper insights into gradient descent
-
Week 10 : October 24 - October 30
-
-
Tuesday
Overfitting and Regularization
In the context of polynomial regression illustrate risk of overfitting and also how regularization can help
Thursday
Cross-validation and Measuring Performance
Cross validatio and stratification including visualizations then ways to measure accuracy
-
Week 11 : October 31 - November 6
-
-
Tuesday
More on Performance Measurement
A more careful assessment of what it means to say an algorithm made a mistake
Thursday
Introduction to Neural Networks
Starting simply and building up the essential with a multi-layer Perceptron
-
Week 12 : November 7 - November 13
-
-
Tuesday
Neural Networks Using Keras
Modern ML APIs such as Keras are fantastically useful but also require a degree of commitment to learn API idioms
Thursday
One Hot Encoding and Softmax on MNIST
Using a classic example problem with characters to motivate one hot encoding and softmax activations
-
Week 13 : November 14 - November 20
-
-
Tuesday
Dense Networks and Image Recognition
Begin to explore more complex dense networks on classic tasks starting with character recognition
Thursday
Convolutional Neural Networks
Using the CIFAR10 dataset to motivate and illustrate convolutional neural networks
-
Week 14 : November 21 - November 27
-
-
Week 15 : November 28 - December 4
-
-
Tuesday
Decision Trees
Classifers that work like a game of twenty questions
Thursday
Ensemble methods and Support Vectors
Combining the opinions of many classifiers - each in its own way flawed - can tap into wisdom of the crowd
-
Week 16 : December 5 - December 11
-
-
Tuesday
Course Wrapup
Student led question and answer about topics covered this semester
Thursday
Current Research on Embodied Agents
A brief overview of our work building embodied agents able to communicate using sight and speech