We’re working on something that will change the way we learn, share and explore Machine Learning - Introducing MachineLabs.
If you work anywhere close to the tech industry you surely heard about Machine Learning. These days many of the most impressive advances in technology rely on trained neural nets. The idea is simple: Where traditional programming is all about writing code to solve problems, Machine Learning is about feeding enough data to a neural net to have the machine figure out the complicated stuff on its own.
Companies such as Google, Tesla or Cruise are using Machine Learning to build self-driving cars. Applications such as PRISMA use it to turn regular photos into amazing art inspired by real artists. Many people believe that Machine Learning and artificial intelligence will have a similar impact on our daily life as the invention of the internet.
But there’s a problem though…
Despite still being a very academical field, Deep Learning and building our own neural networks isn’t a very easy task. That’s why frameworks like TensorFlow, Keras or Theano exist - to help making these things easier.
However, even with these great frameworks available it’s still unnecessarily hard for traditional, web developers (like us) to dive into this amazing field of software technology. Using Machine Learning to solve problems requires a certain infrastructure and computational power to get things going. Coming from a field like as web development, the barrier of entry can seem very high.
To mention some of the bigger pain points:
- Setting up the environment is tedious - Users need to install things like Python, Tensorflow, Keras or Theano, additional librabries and of course, all of this in versions that work in combination.
- Consumer hardware lacks power - In order to perform serious deep learning, a lot of hardware power and resources are needed. That’s why companies like NVIDIA offer things like high performance data center GPUs.
- Training is time consuming - Often, training a neural net can take several hours or even days. We sure don’t want to use our own local machines for that.
- Exploration and Sharing - While there are plenty resources, articles and tutorials on deep learning, there doesn’t seem to be a way to easily consume, share, demo or even fork experiments with the community.
We think Machine Learning should be simple and fun. We believe that, if we manage to help people getting started easier and faster with Machine Learning, everyone will benefit.
With MachineLabs we are building a platform that tries to solve all the problems mentioned and could probably be best described as the CodePen for Machine Learning.
MachineLabs enables you to write and execute your experiments right from within the browser - no installation hassle, no maintenance. Neural net training may run for hours or even several days and you’ll be able to conveniently come back any time to check its status or watch it learning - right, in the browser! We make sure your experiments run on blazingly fast hardware to crack real world problems.
We also believe that for the benefit of everyone it is crucial to accelerate knowledge sharing in the Machine Learning field. With MachineLabs, labs (that’s what we call our experiments) can be forked and shared! Imagine having a lab with a steering prediction for a self-driving car right in your browser and have it forkable and hackable for everyone in your community!
Just like people share code snippets and runnable examples for other languages and platforms on websites like Stack Overflow or GitHub, we want everyone to be able to share or embed labs on other websites to improve the overall accessibility of this exciting technology.
We will use this blog to keep you posted about latest updates and features landed in the MachineLabs code base so you have a chance to get a better picture of what’s happening behind the scenes as the private beta is ongoing.
In the future you’ll be able to:
- Persist your trained model - Labs will have access to a simple API to save trained models and other artifacts
- Build on top of other trained models - Trained models that were exposed from another labs can be consumed to build on top of them
- Easy access to exciting datasets - We’ll make it simple to consume public datasets to enable people to tinker with self-driving cars and more
- Visualizations - Visualizations will make it easier to understand what’s going on with your machine learning code
We launch our private beta very soon to finish implementing the most important features with the valuable feedback of selected people that are into Machine Learning, before we roll it out for the public. Head over to our registration page, tell us about your background and why you want to use MachineLabs and request your personal privat beta invite!
Seats are limited!
Also, make sure to follow us on Twitter and don’t miss any updates!