At our current scale (small and scrappy but hoping to grow), I don't see Cake having much use for Cloud AutoML.
It looks like AutoML simplifies the process of creating custom machine learning models, but we'd still need to be able to provide enough data to those models to train them to do something useful. Right now Cake doesn't have a large corpus of data that could be used for this.
What's more appealing to me right now are pre-trained services like Amazon Rekognition and Google Cloud Vision, which have already been trained with massive amounts of data and could be useful to Cake even at our current small scale.
Hopefully someday we'll be successful enough that it will make sense to create and train our own custom ML models, but we're not quite there yet. 🙂
I had to tell my wife that I now have a stash of nudes on my laptop... If anyone discovers it, they're going to wonder what I'm into. Men? Women? Statues? It's to test the various image screening services like Amazon Rekognition and Google Cloud Vision. My hypothesis is Google would be superior in detecting nudity than Amazon because they must have a much larger training set.
The reality is neither one of them must be tuned for what we're looking for. The success rate on my 20 images is close to zero.
That service will be interesting to follow and see its influence in the next few years.
What about Facebook? It seems adept at identification.