Machine Learning Roundup (September 29, 2016)

Machine learning (ML) and deep learning (DL) content from the past 24 hours. ICYMI.

Issues

On the Electronic Frontier Foundation’s Deeplinks blog, Gennie Gebhart and Erica Portnoy ponder privacy issues raised by facial recognition technology.In Fortune, Jeff John Roberts enumerates what companies get wrong about ML.

Also in Fortune, Roger Parloff explains why DL is suddenly changing your life.

Opinions

Katyanna Quach quotes Oren Etzioni, CEO of the Allen Institute of Artificial Intelligence, who cautions against overhyping this stuff.

In Bloomberg, Saijel Kishan quotes quant hedge fund Two Sigma’s founder David Siegel, who thinks AI is stupid. Scroll down to the bottom of this page, and you’ll see that Two Sigma is hiring.

People

IBM’s Ginny Rometty lectures bankers, touts Watson. Under Rometty’s leadership, IBM’s revenue and profit have declined 20% so far while bank profits soared.

Research

Tiffany Fox profiles the Pattern Recognition Lab at the University of California, San Diego.

The Penn State Daily Collegian profiles Alayna Kennedy, who will present a paper on the use of neural networks in human prosthetics.

In the EURASIP Journal on Information Security, a paper about Hidost, which uses ML to detect malicious files.

Applications

Kadaxis’ Chris Sims describes how her company uses machine learning to predict book sales. Speaking of books, Disruptive Analytics is available on Amazon.

Leslie Ellis reports on a machine learning workshop for cable and network execs.

MIT Technology Review describes how Los Alamos uses ML to track disease outbreaks around the world.

ML outperforms inexperienced radiologists differentiating malignant and benign thyroid nodules, according to a report in Diagnostic Imaging.

The Japan Times reports on AI-assisted medical diagnosis.

Adam Stone describes how ML transforms customer care.

Professional Planner describes ML in quant investing.

In Fierce Biotech, Stacy Lawrence summarizes “Surgery 4.0”, robotic surgery guided by AI.

Mohan Devie surveys potential uses of ML/DL/AI in financial services.

In Ars Technica UK, Bob Dormon describes how grocers seek to use ML/DL to sort groceries.

Methodology

In RTInsights, Kai Waehner explains how to avoid the anti-pattern. Read the article to see what that is.

Facebook’s AI guru Yann LeCun touts unsupervised learning.

Software and Services

Google announces the Google Neural Machine Translation System (GNMT), a neural network for machine translation at production scale. Linkapalooza here.

Also, Google adds ML tools to Google Analytics.

Splunk adds machine learning for anomaly detection, adaptive thresholding, event analytics and threat detection. Linkapalooza here.

Sight Machine announces release 2.0 of its ML-driven manufacturing software.

Clarifai introduces two ML services for image and video recognition.

Tessera Technologies’ FotoNation business unit announces DL-driven facial recognition, iris recognition solutions.

Hardware

In ExtremeTech, Graham Templeton explains the potential of neuromorphic ‘brain’ chips.

CEVA introduces DSP-based hardware accelerator, software framework for DL on low-power embedded systems (such as smartphones, robots, drones and autonomous vehicles.)

NVIDIA announces ‘Xavier’ System-on-Chip (SoC) designed for DL on autonomous vehicles.

E4 Computer Engineering announces the availability of E4 OP206, a system for DL with the latest IBM POWER8 chip, NVIDIA NVLink and up to four Tesla P100 GPU-accelerated platforms.

Companies

Amazon, Facebook, Google, IBM and Microsoft launch an AI partnership which they describe as an initiative to “advance public understanding and formulate best practices,” but which some might describe as an oligopolistic combination to restrain competition. Linkapalooza here.

ML startup DataRobot celebrates its 110 millionth predictive model.

Security vendor Webroot acquires Cyberflow, which specializes in anomaly detection.

Skymind raises $3M to develop an open-source deep-learning library for Java. Because we don’t have enough of those already.

Chris Pash reports surging interest in ML-driven security vendor BrainChip following an announced deal with a Vegas casino.

Sales conversion specialist Apptus secures $88 million in late-stage funding.

Careers

Two Sigma is hiring.

For PhDs, some open positions in Switzerland, which doesn’t suck.

Airbus Toulouse has an internship in ML, for European youth who can’t find permanent jobs.

Advertisements

2 comments

  • Hi Thomas, thanks for the Splunk mention. (Yes, I work there). Please weigh in – one of the things I like about Splunk is that the machine learning capabilities are either built in or available as an add on at no extra cost AND as such training, testing, and putting into production are all part of the same ecosystem. One of the stories I tell internally is about working for one of the other big predictive analytics companies and how we would help customers extract data for training and testing – but putting the model into production was a separate IT process. I know it’s a generalization but frankly I’m not aware of any of the big “ML” vendors who can tell the integration from modeling to implementation. Your take?

    • Brian,

      Thanks for reading. There is a difference between general-purpose machine learning engines and embedded machine learning engines, as you note. General-purpose tools operate independently of data sources and consuming applications, and the interfaces tend to be bottlenecks. That’s not a new thing, however; the Unica Marketing platform (now an IBM product) also includes embedded ML, and that product dates back to the 1990s. There are many other vertical and horizontal applications that include some level of ML capability.

      So, which is better? Given that embedded ML is not a new thing, one has to ask: why do enterprises still use general-purpose machine learning tools? One reason: because in most enterprises ML is still the domain of expert data scientists, who prefer to use general purpose tools. Another reason is a concern about creating silos; unless there are open standards that support enterprise model management, embedded ML makes it more difficult to share insights across business processes.

      That said, embedded ML in an app like Splunk offers better time to value and an immediate solution. For those reasons, I’ll think we’ll see more of it in the future, not less.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s