I’ve got two links I’d like to share. This recent New York Times article on bias in face recognition caught my eye/ire. The article highlights work by Joy Buolamwini (MIT) and Timnit Gebru (Microsoft Research) described here on the persistent gender and racial bias in artificially intelligent facial recognition. It’s tiring, really tiring. As if the Big Artificial Intelligence companies (Google, Facebook, etc.) are saying “Yeah, our artificial intelligence is sexist and racist, deal with it.” We can take solace that there are phenomenal researchers who just won’t let it go. That’s perhaps how change happens anyway.
While we’re on the topic of artificial intelligence (AI), Viktorya Krakovna has a great blog on AI Safety that you should visit.
I too have two links to share. In The human problem of machine learning, the blogger says:
…like many market revolutions that have come before it, machine learning (ML) stands to make a lot of people a lot of money. So, granted that, the market must be flooded with newly minted AI experts, right? Actually, that’s not quite the case. In fact, according to this New York Times article, there is a severe shortage of talent, so much so that Google has sponsored a division to create ML programs to, in turn, create ML programs.
That blogger then discusses the issue of supply and demand in machine learning, with several interesting links. For the past few years, I have been living under the assumption that the huge demand for machine learning experts will calm down, as more and more commodity products become available. So far I’ve been proven wrong.
Additionally, AnnMaria De Mars, who holds a Ph.D. in statistics, is a world judo champion and is CEO and a founder of a 7 Generations Games, a company that produces educational games for children. I’ve been following her statistics blog for at least six years, and try not to miss any new post she publishes.
I’ve got three links to share. Firstly, Google is using convolutional neural networks to analyze images from a person’s retina to predict age, blood pressure, and smoking status. Secondly, are you interested in Deep Learning? Here’s a great resource to help you brush up on your matrix calculus. And last but not least, the precious truth will be in high demand in the future as technology continues to blur the lines between what is real and fake. Students from Washington University have used recurrent neural networks to re-create photorealistic lip syncing from audio files.
My education is in physics and before I left academia, I used machine learning to filter the data collected by the IceCube neutrino telescope. This article gives a number of examples of how particle physicists use machine learning to analyze the large amounts of data that are collected by their experiments today.