This Week in Machine Learning
We’re exposed to tens of “look at this cool thing I made using ML” daily and a few of them leave us impressed. So impressed that we want to share them with you. In order to do this we’re kicking off a new weekly series!
This series is going to be showcasing a mix of applied machine learning projects, useful resources and promising start-ups that leverage ML in various ways. We’ll try to stay away from research, but if something really gets us excited we will list it here, putting emphasis on its applications rather than the theory.
Expect lots of links and only a little bit of text. But first, let’s start with a weekly status update on a pretty significant issue.
Are robots ruling the world yet?
No.
Okay, let’s move on.
Microsoft is showing off what P3sportsicence (peek performance project) has been doing using machine learning and their hardware… and probably their cloud?
There’s surprisingly little information on their cooperation and the exact technologies used. But it sure looks cool.
The video reminded me of MIRA Rehab, a start-up that’s gamifying physical therapy through similar body tracking methods.
As we discussed in our first article, most machine learning models are used to find and learn correlations in data sets and have no notion of causal effects. One of the main obstacles in adopting causality in ML is the need of an a priori causal model of the problem on top of the data set.
In his article, George Lawton is hinting towards how a model-free causal deep learning setup could work and touches on the subject of explainability. It’s been retwitted by Judea Pearl, the father of causality and the author of The Book of Why. You know it’s gonna be good!
Talking about ML explainability, IBM Research introduced Explainability 360. A toolkit containing various explainable algorithms. While the push towards explainable AI is laudable, the algorithm zoo (that you can fined it here) is quite limited for now.
This week in “pretending we’re doing AI to get some sweet funding”: Engineer.ai, as it was revealed by Wall Street Journal.
More and more reports about startups and companies that pretend to use ML and AI appear in the news as of late. A previous report, this time by Forbes, showed that 40% of Europe’s AI startups don’t actually use any AI.
In related news, two new reports came out in the last week that take a deep dive into how AI is applied in various organisations and what are the pains and risks of implementing AI and ML.
Most companies use AI to “ Enhance the features, functions, and/or performance of our products and services” and find trouble with integrating ML within existing systems and the cost and scarcity of talent. In other words, ML is still hard to do — we haven’t launched yet ;).
You can find both studies below.
This pretty much covers it for this week! But before we go, two more things…
Startup Spotlight
Each week we’ll put a startup into our proverbial spotlight. This week the startup is Keigo.
Keigo uses AI assisted psychometrics to compute the personality of your future conversation partner by analyzing their social media feed or other text they’ve written. Based on that it will offer you insights on how to approach the meeting so that you’ll better connect with your partner.
Here’s a Medium post in which manne pyykkö, one of Keigo’s co-founders, goes through how he’d use the app to prepare for a meeting with Trump.
Quite a fun read!
Meme of the week
And talking about fun, here’s a caption that got us laughing hard!
That’s all for this week. Don’t forget that you can still join our closed alpha waiting list. Also, follow us on Medium and Twitter! See you next time!
