Machine Learning has had many positive, sometimes unexpected, results and applications in modern times. Industries from Transportation, Finance, Media, Arts to Medicine and Politics can expect to be disrupted by Machine Learning in the not so distant future. Machine Learning is here to stay.

As a pragmatic entrepreneur, I’ve decided to use the ever-growing research, literature, and software to empower businesses and help them embrace the future today.

The result: Picmatix, a Machine Learning company focused on getting things done.

Machine Learning in the wild

These days, Leapfrogging years of previous research and comfortably setting a new state-of-the-art in established industries can be a matter of just plugging in existing Machine Learning models to new domains:

Deep features to classify skin lesions

… Compared to state-of-the-art, our proposed approach achieves (…) noticeable improvements in accuracy for underrepresented classes (e.g., 60% compared to 15.6%).

If these numbers don’t mean much to you, think about the following analogy: the top Olympic swimmers struggle to improve their timings by milliseconds. This is what the state-of-the-art in many industries looks like today. Then, imagine a different swimmer with a completely new swimming technique comes along and wins every competition, beating everyone else’s records by seconds or even minutes, not milliseconds. This is what deep learning models are doing to new verticals all around the world.

In the medical industry, advances in Machine Learning have been used to detect skin cancer, prevent cervical cancer through automated screening (* Kaggle competition in progress), detect Diabetic Retinopathy in Eye Images, and much more.

Machine Learning is successfully used in Amazon’s recommendation engine to increase sales. In the payment processing industry, Stripe’s fraud detection systems are shielding customers from fraudulent charges. In the fashion industry, Zalando is building great customer experiences with the help of deep learning. Tesla uses deep learning as a key component of their self-driving cars.

And the list goes on. The secret sauce: neural nets.

Tesla self-driving car in action

Tesla uses deep learning to power their self-driving cars

Neural networks

Neural networks are mathematical constructs, weakly inspired by our brains, that create a mapping between input and output. Neural nets have been proven to work extremely well for visual recognition tasks and for understanding human language.

While there are still many challenges, both practical and theoretical, with respect to neural net themselves. The main challenge for Machine Learning practitioners is how to best model a problem in such a way that it can be solved by existing technologies.

Coloured Neural Network

Neural Network with one hidden layer. By (CC BY-SA 3.0), Wikimedia Commons

Open Collaboration

An undisputed reason to the success of Machine Learning, in particular, neural networks, is the vast amount of data and computing power available to researchers and practitioners. I would argue that another significant contributor is the openness of the field. Researchers not only publish their papers with their results and discoveries. They make their code, algorithms, and models available to everyone for them to be improved upon.

Similarly, the Distill team, lead by Chris Olah and Shan Carter are creating new mediums to explore and publish Machine Learning research. Enhancing collaboration all around the world.

Open collaboration has been and will remain fundamental to many past and future Machine Learning break-throughs. Openness and collaboration are and will remain at the core of our work at Picmatix.



Picmatix’s goal is simple: to leverage existing computer science and Machine Learning technologies to more effectively solve problems across industries through an open collaboration process, publications and open source contributions.

If you own a business or product you can use Machine Learning today to make it better. Let’s talk: ml

comments powered by Disqus