Office Hours

A place for Brandon to chat with students and for students to chat with each other

Two dimensional convolutional neural network case study

There's a fresh Cottonwood case study out — a two dimensional convolutional neural network that learns to categorize handwritten digits from the MNIST data set. It will be part of the 2D ConvNet course, 322, now under construction. It builds off the code and theory we walked through in Course 321...

READ MORE

Collection of Cottonwood examples released as Course 209

My favorite documentation is a functioning implementation. To this end I've created a growing collection of Cottonwood implementations to use as a starting point for your own projects and investigations. I plan to grow it over time and make it an increasingly rich arsenal of working examples...

READ MORE

Section 7 of Course 321 is out

The one dimensional convolutional neural networks course is now complete, minus one section. In section 7 of course 321 we demonstrate advanced model development techniques, including adding layers and data augmentation. All that remains is to build the model into a lightweight tool that does...

READ MORE

Course 137. Signal Processing Techniques open for enrollment

I just realized that Course 137 wasn't allowing enrollment. I just changed that. The data munging course (131) was getting a little cluttered, and I wanted to add even more to the library of signal processing content, so I broke it out into a separate course. It, like all the 100-series...

READ MORE

Course 321 Section 6 release: Model creation

In this latest section, we get down to the business of creating a heartbeat classification model and begin to tune it and build it out. With this section, Course 321 on one dimensional convolutional neural networks has become mature enough to graduate to a full-fledged course. This is the fun...

READ MORE

Course 321 Section 5 release

Section 5 of the 1D convolutional neural networks course is out! This is the grittiest bit of the project: getting the electrocardiography data, learning what it means, and choosing what parts of it we're going to use. It's also in my opinion the most critical. This is the bedrock of the...

READ MORE

Course 321 Section 4 released

The latest section of Course 321 on one dimensional neural networks just landed. It builds off of our previous work on convolution to add ReLU, pooling, and batch normalization. Finally we have a complete convolutional neural network with all the trimmings! There's a fully worked example in there...

READ MORE

Cottonwood is outfitted for one dimensional convolutional neural networks

I’m happy to let y’all know that as of this weekend Cottonwood has a complete kit all the tools necessary to get started with one dimensional convolutional neural networks, Including convolution, rectified linear units, pooling, and batch normalization. Here’s an example showing all of the...

READ MORE

New Release: Course 137. Signal Processing Techniques

The data munging course (131) was getting a little cluttered, and I wanted to add even more to the library of signal processing content, so I broke it out into a separate course. It, like all the 100-series courses, is free. It's a collection of things I use or refer to regularly. I hope you'll...

READ MORE

Testing your code brings peace of mind

Testing a machine learning framework is challenging. When an ML algorithm runs, it operates on data to produce a result. It usually incorporates an element of randomness. Data is often unknown unfamiliar. It’s hard to automatically determine whether everything is working as it...

READ MORE