Tuesday, February 6, 2018

Software "no more accurate than untrained humans"

AI and machine learning has been in the news almost constantly for the last year, so you'd be forgiven for thinking that the era of decisions being made by computers is now here and people should just let the AIs take over. However, we must be careful. A program in use since 1998 in the US, called Compas (Correctional Offender Management Profiling for Alternative Sanctions), used to assess more than a million US defendants may not be accurate enough for potentially life-changing decisions. The accuracy of the program used for bail and sentencing decisions has been called into question after it was found to be no more accurate at predicting the risk of reoffending than untrained people. Read more about this in this Guardian article,

Wednesday, December 13, 2017

AlphaZero learns chess in four hours

You may have heard of Google's deep learning system AlphaGo that learned to play Go better than the best human. Well, the latest incarnation, AlphaZero, has learned to play chess better than the best computer chess playing system, Stockfish. However, what is most remarkable is that AlphaZero took only four hours to learn how to play chess! Read all about it in Chess.com. Machine learning is advancing very rapidly.

Monday, December 11, 2017

Mozilla’s Open Source Speech Recognition Model and Voice Dataset

Mozilla has just announced the initial release of Mozilla’s open source speech recognition model that has an accuracy approaching what humans can perceive when listening to the same recordings. They are also releasing the world’s second-largest publicly available voice dataset, which was contributed to by nearly 20,000 people globally. This looks like it will be a very useful resource for researchers.

Saturday, November 25, 2017

Republicans are about to kill the open internet and this town will pay the price

You've probably heard of "net neutrality" but maybe unsure what it is. Put simply it's the idea that every packet of information crossing the Internet has equal priority. Whether you're watching Netflix, listening to Spotify or playing Minecraft all the data is treated equally. Some people who argue for the "free market" claim that Internet service providers (ISPs) should be allowed to prioritise data from certain sources and to specific customers, of course for a fee. Apple, which has huge cash reserves, could for example pay ISPs to prioritise data from Apple Music giving users a better service at no visible cost to them. However, many argue that this would go against the founding principles of the Internet. To see what the future may hold in an unequal Internet look no further than the small rural town of Winlock, Washington – where the Internet is dead slow, if available at all – the residents there are major proponents of net neutrality and argue the Internet is a basic necessity. The Guardian has an interesting article on this.

Tuesday, November 7, 2017

Computers in the kitchen

Is there nowhere now that computers aren't being used? You'd have thought perhaps that the kitchen would be relatively free of computers. Well, you'd be wrong. Recent innovations in 3D printing are letting innovate chefs create wonderful looking (and hopefully great tasting) dishes. This website illustrates some of the fabulous things that can be made from 3D printers and computer-assisted lathes.

Tuesday, October 17, 2017

Which machine learning algorithm should I use?

Which machine learning (ML) algorithm should I use? It's a common question I get asked and usually, once I know something about the application and the data I can make an educated guess: a clustering algorithm, a neural net, k-nearest neighbour... But, I've been working in ML for decades. For ML newbies this is a hard problem because there are so many ML algorithms to choose between. SAS have created a resource designed primarily for beginner to intermediate data scientists or analysts who are interested in identifying and applying machine learning algorithms to address the problems of their interest. Read their blog post to learn how to navigate their flow chart.

Thursday, October 12, 2017

Machine learning: the power and promise of computers that learn by example

Machine Learning is becoming more important in many aspects of our daily lives. However, most of the general public and importantly politicians and policymakers are quite ignorant of its scope, strengths and weaknesses. To better inform people the UK's prestigious Royal Society has recently released a report titled Machine learning: the power and promise of computers that learn by example.