An amateur to coding but trying to build a Deep learning tool? That may have sounded impossible had it been last year. So what’s up with this year then? It’s the Lobe. Heard of it? Some of us might have, for it’s causing quite a buzz among the AI community since earlier this year. How would you feel if you’re told that you could design a Deep learning tool, without having to write even one line of code? As much as it sounds impossible, it has become possible with the Lobe.

Lobe more or less boils down all the applications of DL down to a simple interface, which enables the user to

  • Build- Input training data set and Lobe builds a custom DL model
  • Train-Can keep an eye on the training process using the interactive interface
  • Ship- Export the model to TensorFlow or CoreML and run it directly on Android or iOS

All without even a line of code typed in by the user. That is quite a feat!

This is currently based on Machine Vision. If you want to build a tool that recognized emoji, listen to the sound of a musical instrument and recognize the instrument, measure the quantity of a substance in a container or anything that requires the use of machine vision can be successfully built and run using Lobe.  Makes you wish you had heard of it sooner, right? Same here.

At present Lobe is in Beta version and though it needs some slight tinkering, it has a simple interface which makes it a sweet heaven for the amateurs.  The same procedure as you would follow in any machine learning applications, collect a dataset, label the dataset by sorting them, and let the neural network do the work. The thing with Lobe is that it has got its own default templates for some general tools. Let’s say we’re creating a tool that identifies the emotion of a person in the image and shows the correct emoji that suits the emotion. In this case, we might be going with, Matching Images with predefined labels template.

We all know how deep and intricate the work is with applications like TensorFlow and that is the advantage and the disadvantage with the lobe. Your options of customizing the tool are heavily limited when compared to TensorFlow. You have no means of looking at the bare code in Lobe and modifying it. They’re already predefined and that makes it a little bit rigid for those who are looking for more. But on the other hand, for those who’ve never written a code in their life, this is pretty much an alluring gift. Deep learning is no longer a foreign untouchable concept for the amateurs and that doesn’t mean we can consider what Lobe represents as the whole extent of AI.

Lobe has indeed opened up a sliver of AI and presented it as easy on the eye enigma, though there’s still more left to be explored.