Machine Learning Roundup 10/10/2016
Machine learning (ML) and deep learning (DL) content from Friday and the weekend. Here is a summary of last week’s top stories.
Note to readers: Big Analytics will rebrand as ML/DL soon, with a new format. The new name reflects the increased focus on machine learning and deep learning.
NVIDIA versus Intel
Wall Street has a fever, and the prescription is more NVIDIA. Reinhardt Krause profiles the company, noting its consistent, sustained leadership and strategy. A licensing agreement between NVIDIA and Intel may expire next year, provoking an all-out war in the high-performance computing market. Merrill Lynch is bullish.
And you thought it was hype An engineer uses DL to resurrect her dead friend. Casey Newton reports, in The Verge.
Brandon Ballinger describes three key challenges for ML/DL and AI in medicine, arguing that if we can use DL to identify cat videos, we can use it to heal. I don’t know why he has a problem with cat videos.
Ryan Kh explains how ML affects consumer behavior: supporting real-time decisions; matching customers and products; anticipating customer needs.
Jeffrey Schwartz summarizes the different approaches to ML/DL/AI of Microsoft and Google.
Algorithmic bias is in the news. Writing in TechCrunch, Devin Coldewey writes that Google researchers aim to prevent machine learning from discriminating. That may be difficult to do since ML is supposed to discriminate; for example, between good and bad credit risks. The issue, of course, isn’t discrimination per se, but bias that has a disparate impact on classes of people.
In The Guardian, whose circulation has markedly declined in the past fifteen years, Julia Powles whines that technology companies are just too darned big. Nobody forces anyone to use Big Tech; delete your accounts if you want to be free.
Jessica Davis worries that an AI boom will cause a jobs bust. It’s called the Luddite Fallacy, a phenomenon that has accompanied most innovations. Think of it as a rent-seeking argument deployed by folks who depend on industries at risk of disruption. Newspapers, for example, like The Guardian.
Google engineers use DL to teach AI how to write songs, Tina Amirtha reports. Meanwhile, in Motherboard, Michael Byrne profiles a computer scientist who has published a manifesto for algorithmic music. Iannis Xenakis anticipated him by 60 years with Pithoprakta, for two trombones, 46 string instruments, xylophone, and woodblock. Xenakis based the work on the statistical properties of gasses, modeling each instrument as a molecule obeying the Maxwell-Boltzmann distribution law, with Gaussian distribution of temperature fluctuation. If you’ve never heard of Iannis Xenakis, you have some idea about the potential for algorithmic music.
In LiveScience, Tia Ghose explains the “spooky secret” behind the power of AI. Spoiler: the secret is DL.
Writing in CIO, Mitch De Felice discovers that you can’t extract value from big data with Big Data. (Doh!) For insight, you need machine learning. He provides a pretty good summary of what neural networks can do.
Also in CIO, Greg Simpson lists five trends enabling the growth of AI:
- Cheap computing in the cloud
- Big data
- Sensors and distributed intelligence nodes
- Natural language processing
- Agile software development
Hey, someone should write a book about this.
Methods and Techniques
Allen Downey explains how to use Bayesian analysis to answer the “Rain in Seattle” question. Of course, he could just check Weather.com.
On Data Science Central Aravindakumar Venugopalan — yes, that Aravindakumar Venugopalan — offers a pretty good rubric for choosing ML platforms.
The editors of Techseen report on Mitsubishi’s automated DL algorithm.
Early bird registration deadline for the IBM Watson AI Xprize is October 15. You don’t have to use IBM software.
Tesla claims 222 million miles for its ML-driven Autopilot capability.
Dylan Furness describes how DL-driven chatbots could serve as a “doctor in your pocket.”
Researchers at Duke use DL to diagnose malaria.
MedicalResearch interviews Dr. Saul Blecker of New York University, who explains how ML analysis of free-text notes improves identification of patients with heart failure.
Marion Marking profiles recent developments in ML-driven language translation.
NASA engineers build DL-driven image recognition into asteroid-fighting drones.
Argyle Data’s Padraig Stapleton describes Ml-driven fraud detection for telcos.
Researchers at the University of St. Andrews in Scotland develop RadarCat, a sensor that can recognize objects and materials.
Software and Services
Katherine Noyes touts Watson.
Optimove launches a “CRM Optimization Bot,” aka a spambot.
Srini Penchikala reports from Reactive Summit 2016. Some speakers made fun of Spark Streaming’s micro-batch architecture, which gave some Spark contributors the sadz.