Collecting image data for training machine learning models can take precious time and lots of Google image searches.

I built a tool that generates labeled data for you. Choose your categories and download.

Get labeled image data for your machine learning models

LibertyJS Presentation: Ask What AI Can Do For You

Hi friend, I hope you enjoy the presentation!

Slides

Here is a link to the slides.

Code Demos

In the talk I mention a number of code demos demonstrating neural nets in the browser. Links to each can be found here:

  • line — train a simple neural network on a line
  • moons — train a neural network to predict non-linear data (in this case, two parabolas)
  • webcam — train an image classifier with a webcam using transfer learning

Libraries

  • Tensorflow.js is the Machine Learning library we looked at during the presentation.
  • make_moons is a port from scikit's make_moons function. It generates two parabolas of x,y points.
  • ml-classifier is a library written in React for generating image classification models in your browser with drag n drop.
  • tfjs-vis is a library for visualizing your models right in your app.

Educational Resources

I've got a few articles you can read on machine learning specifically in Javascript:

Get labeled image data for your machine learning models

Kevin Scott

Collecting image data for training machine learning models can take precious time and lots of Google image searches.

I built a tool that generates labeled data for you. Choose your categories and download.

Some things I glossed over in the talk that you can learn more about:

Information on Tensorflow.js

The Tensorflow.js docs contain a wealth of information about the library.

Another useful resource is the official Tensorflow.js examples repo, with code demonstrating a number of popular machine learning techniques with live demos.

The Tensorflow playground predates Tensorflow.js. It lets you build a neural net in your browser and see the training in real time.

Courses

There are a number of great courses for learning Artificial Intelligence.

Fast.ai puts out a series of video lectures aimed at hackers. The courses are done in Python.

Andrew Ng's Coursera course is widely recognized as one of the best introductions to AI and machine learning.

References from the presentation

Here's the list of references from slides in the presentation, along with a number of cool links demonstrating what neural nets can do.

A description of linear and nonlinear data by John Sullivan.

3blue1brown's video series on machine learning

Information on Convolutions

A great website demonstrating different convolutions

An ObservableHQ doc for visualizing the different layers of mobilenet

Visualizing how CNNs learn

A list of cool things with neural networks

ML5 - friendly machine learning for the web

Article on Pose Estimation with Tensorflow.js

A wonderful series of experiments celebrating AI weirdness

A great article on differentiable parameters, in fact this whole website is amazing

GANs and generating faces

Facebook's work with audio style transfer


Finally, if you enjoyed the talk, or have questions, feel free to shoot me a message on Slack or DM me on Twitter! I'd love to hear what you think.

Get labeled image data for your machine learning models

Collecting image data for training machine learning models can take precious time and lots of Google image searches.

I built a tool that generates labeled data for you. Choose your categories and download.