September 11, 2019
Machine Learning has always been big at YND. Take CleanAI, for example, our AI-powered dirt & hazard detection solution, or Smartbar, the facial-recognition beer dispenser in partnership with Wirecard. It’s a field that constantly develops, at a fast pace. So, to stay on top of things, we went to the Deeplearn conference in Warsaw, a 5-day Summer School revolving around the topic of Deep Learning. This technique in the field of AI is mainly used for advanced tasks such as the recognition of objects in images and automated text translation.
This year’s Deeplearn conference consisted of a variety of sessions about different aspects of the field. Most of the talks were mainly technical, for example, a talk from Facebook’s Tomas Mikolov on Using Neural Networks for Natural Languages and Deep Generative Models by Aaron Courville (University of Montréal).
You can watch the Facebook talk here (filmed at different event):
Now, let’s have a look at what stood out for us during this year’s conference.
Generative Adversarial Networks
The main takeaway from this year's Deeplearn is that there’s great progress on Generative Models (GANs). For example, we saw several new methods of training popping up and the general output is more accurate. New Machine Learning models also show impressive results in speech generation (such as WaveNets) and artificial background generation.
Deep Neural Networks
The regularisation and generalization methods in DNN. This helps to optimize the architecture of the neural network and can be used to achieve better scores and improve accuracy.
We saw more variation in the type of attacks on Machine Learning APIs, like KYC, facial recognition and similar systems. We had a chance to find out more about poisoning training and test sets, attacks on online learning models, single-pixel attacks and other computer vision spoofing approaches. Also make sure to check our post on face spoofing or download our very own Face Secure app here.
I believe this part of AI is getting more and more popular but on the other hand, people are now aware that it is one of the most challenging domains and still there are lots of weaknesses. From an academic perspective, many of the challenges in this area are already addressed, and many papers are in progress. Events like Deeplearn are vital to keeping up-to-date with developments in methods and technologies in the field of Machine Learning. As always, we’ll report back on the YND blog on any other conference we’ll attend. Thanks to people at Deeplearn for setting up this great conference, we’ll be back next year!
This post was written by Artur Baćmaga, one of YND’s AI experts. With over 6 years of experience in Python, Artur works as a ML/Python Developer at YND. He leads AI-powered processes for projects such as Car Detection & SmartBar. In need of some brain power? Reach out to us via email@example.com with your questions about ML/AI projects.