The First Cry of Atom

Had a Talk at TensorFlow.js Workshop

Last week, I attend the first ever TensorFlow.js workshop held by Machine Learning Tokyo (a.k.a MLT) community as a speaker with Rei. Rei was a co-author of Deep Learning in the Browser that is a book about the technology of the basic of deep learning running with JavaScript or TypeScript. In this workshop, we spoke about the basic of web technologies (e.g. JavaScript) and TensorFlow.js around how to train the model and import the pre-trained model into TensorFlow.js. If you find the workshop interesting, here are the material used in the day.

A presentation slide material is available in Google slide. We also have a code repository in GitHub, Machine-Learning-Tokyo/tfjs-workshop used for the tutorial of TensorFlow.js. Please give me some feedback if you find something wrong in the tutorial.

Teaching someone itself is often the thing we tend to face the difficulty because of the gap of prerequisite knowledge respectively. As I had no experience to have such kind of workshop, there were several things I need to prepare in advance

Have an Enough Time Buffer

People may take some amount of time to complete the tutorial more than we initially expected. That’s the typical thing if the workshop is open for those who are not familiar with the things we are going to talk. Packing many things in a workshop is not a good idea because a lot of people may not be able to complete the program. That brings the attendees an unsatisfactory experience. Therefore, we should not be greedy about the contents provided in the workshop. It is important to pick up the minimal but essential things to be taught. That will give people meaningful time in the workshop so that they are able to keep learning even after the workshop, I believe.

Touch Everything as much as Possible

Although we should make the content minimal, we should also cover the topic as much as possible so that the learner can review the content afterward. I think introducing the terminology or concept is important even if we do not deep dive into each of them. As far as they remember the word or concept ambiguously, they can restart the learning by researching by themselves. So covering the whole thing is also important in this kind of introductory workshop.

Fun Tutorial

I think this is the most important part to provide a good workshop. People can have a good experience by creating the application actually running in their environment and see the result. The good thing of TensorFlow.js was that it was suitable for that kind of working apparently application because the result of the application can easily be shown in the web browser. Actually, we created demo applications working in the browser so that everyone can easily check the result just by having Chrome or other browsers. Moreover, GitHub pages enables us to publish the application for free so that you can share the application result globally.


I’m not sure taking care of these things contribute to making a workshop better finally, but many people seem to enjoy the workshop. Above all, that made me glad to do the workshop.

Last but not least, thank you for having me as a presenter of the workshop. Google Tokyo office was a great place to have this kind of conference. The workshop itself brought me a nice experience to teach someone having a variety of knowledge and skills. I hope the workshop would be useful for the learner of the TensorFlow.js.