31 comments

[ 3.2 ms ] story [ 73.1 ms ] thread
[flagged]
Just like act like it doesn’t exist to begin with, so you’re not disappointed.
Teachable Machine has existed for years: https://www.theverge.com/tldr/2017/10/9/16447006/google-teac...

Its last real update (AFAIK) was in 2019: https://www.theverge.com/2019/11/7/20953095/google-ai-traine...

Like a number of Google projects, this one lives on without any clear direction. It probably will get axed some day, but the technology in Teachable Machine today is so “old school” already that I don’t think it would be that hard for someone to recreate or improve upon.

oh wow, I thought this was new. Given that this has received no attention in 3 years I'd assume this is largely abandoned internally inside company.
This is rad. Perfect snow day activity for the kids.
Isn’t this basically what multimodal LLM does as well… it can do anything on the fly it can understand.

What’s different here?

These are smaller scale model that you can export and run anywhere.

Even the smallest multimodal LLM would be wayyyyyyy bigger than an exported model from this

How small are these models? can I export a model here and embed it in an Android/ios app?
You can supposedly embed them for Arduinos, so an app should be no problem.
I think this is more like fine tuning an existing model to recognize features you specifically intend it to, and be light enough to run locally in a browser.
It's not even fine tuning, it's creating a model from scratch. This isn't like our modern huge models either, these tiny single-purpose models have been around for ages and are quite versatile. They're so small you can't just easily run them in the browser, but train them effectively, which is what this project lets you play around with! Super cool stuff.
The teaching part is what matters, it’s training (tuning in this case) a model, not just using a model already trained for inference (which is what I assume you mean). You’re providing new data that is used to update the model. Inference across an existing multimodal model doesn’t change how it classifies in any way.
Wow I actually have a perfect use case for this in a hobby project. Great timing.

I considered the older version but it's very limited:

> The original Teachable Machine only let you train 3 classes, whereas now you can add as many classes as you like.

I'm curious to see how far this scales, for example can I have a few hundred thousand classes? If so, what are the consequences, if any?

v1 used a very limited (albeit very easy and already quite impressive) form of transfer learning, e.g. take a pretrained network's 1000dim vector outputs given a bunch of images belonging to three sets (since the original was trained on Imagenet), and then just use K-NN to predict what a set "new" image falls into.

v2 does actually finetune weights of a pretrained network. At the time, it was a nice showcase how fast fast JS ML libraries were evolving.

Who owns the model? Does google get to reuse them?
do we have any other self hostable, open source alternatives to this?
It is self-hostable.

It runs locally in your browser, without sending your training data to any servers.

Unless you choose to save it to Google Drive.

If you choose to host the model with Google, they get a copy of your weights, but they still don't see your training data.

Or you can host it yourself with tensorflow.js

And you can also download everything in a zip file, training data and weights, and Google never sees any of it.

If you want the source, it's here -> https://github.com/googlecreativelab/teachablemachine-commun...

FYI, this is not a new project. Here’s an HN discussion from 6 years ago: https://news.ycombinator.com/item?id=15399132
(comment deleted)
The new link mentions "the first version from 2017", so I'm assuming this release is what Google considers version 2.
Yes I agree - the (2017) tag in the title doesn't seem right here given the update.
I've done my share of research on MediaPipe[1], but had never heard of Teachable Machine. I'm curious if these efforts are related, as these products looks like they were almost intended to be used together.

I am definitely excited to see that Google is investing into more "ML at the edge" use cases, especially in the browser. If you've never heard of MediaPipe before, but this caught your eye, definitely check it out. It has seen large uptake in the VTubing community especially as it has a very performant implementation of body + face + hand pose tracking driven by BlazePose.

1: https://developers.google.com/mediapipe