🍺 AI-powered beer brewing, OpenAI builds a hide and seek playground for their agents, machine bias as art, and more — This Week in Machine Learning!
This week was quite a slow one. So lay back, grab a beer and learn how Budweiser is using artificial intelligence to get better tasting beer and better customer relationships.
They explain how human tasters are still included in the equation in order to effectively “label” the data after various new recipes are tried out. This is one scenario where active learning techniques, in general, and Cognifeed, in particular, can really help out with.
Waiting for the beer to brew and having people taste it makes for a very slow and costly iterative scenario. Using Cognifeed they would be able to identify the most “interesting” recipes and only try out the ones that help the algorithm learn the most. If you need a refresher on active learning, be sure to check out our article.
OpenAI trained a bunch of agents to play hide and seek. Various emergent behaviours were observed and documented. You can watch the most notable ones in action in the video below.
We feel important to note that the agents’ behavior is not due to some sort of causal inference going on in the background. They simply observed correlations, playing many enough times, between certain actions and their outcomes. They weren’t surfing the box because they knew other agents were hiding behind walls and they needed a higher ground to spot them. They simply learned from the previous adversarial rounds that surfing the cube increases their chance of getting to their goal.
Another important note is that the agents operated directly on the real state of their environment (accurate positions and rotations of objects in their line of sight) and didn’t have to process any sort of visual input.
It’s easy to anthropomorphize such agents, especially when they are cutely animated and appear to display intelligent behaviours, but we feel that this distracts from the real state of the art when it comes to reinforcement learning and AI in general. What’s your take on this? Let us know in the comments!
Ken Burns effects are noting new, but getting high quality parallax-animated pictures from a single image is quite difficult with off the shelves methods. Well, that’s about to change!
While this application is quite eye-candy and cool, the generated depth maps can have even more impactful applications in image processing, computer graphics (depth maps from RGB textures) and robotics (when depth data is not available).
Curious how an algorithm sees you? Head to ImageNet Roulette and find out! While it is fun to mess around with it, ImageNet Roulette is in fact quite a serious art installation. It is currently on display in Milan and it’s supposed to underline the problem of bias and discrimination in data and data driven learning systems. Don’t take it personally if the results insult you ;)
ImageNet Roulette is a provocation designed to help us see into the ways that humans are classified in machine learning systems. It uses a neural network trained on the “Person” categories from the ImageNet dataset which has over 2,500 labels used to classify images of people.
It is currently on show as part of the Training Humans exhibition by Trevor Paglen and Kate Crawford at the Fondazione Prada museum in Milan. For more context, see their investigative article excavating.ai.
Staying in the realm of art, check out this amazing interpolation generated by a Generative Adversarial Network (GAN) starting from a few portraits by the artist Bay Raitt.
We talked in a previous week about a similar tool, ganbreeder, in the context of generating cool morph animation:
No startups in our spotlight this week. If you know any cool startups that use machine learning or artificial intelligence in innovative crazy ways or if you founded one and want to promote it, be sure to reach out!
We’re preparing a two-part article on how you can use Cognifeed to better target influencers for your influencer marketing campaign. Make sure to tune in next Tuesday to find out more!
Don’t forget to follow us on Twitter!
