Roundup 10/28/2016

Machine learning (ML) and deep learning (DL) content from the past 24 hours. Plus, some AI stuff.

I’m publishing a three-part series on the state of enterprise machine learning in The Next Platform. Part one is here.

Good Reads

— Helen Beers explains AI in the second part of a series. Part one is here.

— Market Realist publishes a twelve-part series on the outlook for NVIDIA. Don’t worry, they’re short parts.


— In Wired, Clive Thompson whines that AI isn’t accountable, citing an example where a consumer’s application for insurance is declined without explanation. It’s pretty clear that Clive Thompson has no idea what he’s writing about:

  • AI-driven applications are accountable to the people who build them. Banks and insurance companies don’t want to decline your business; they want to approve your application unless you’re a bad risk.
  • A biased model is bad for business, and the data scientists who build them know it. That’s why they constantly monitor models for systematic errors.
  • Any predictive model, including “black box” learners, can be designed to deliver explanations for rejects. In regulated industries like banking and insurance, this is legally required.

Other than that, it’s a fine article.

Methods and Techniques

— Researchers at Google Brain publish a paper on how neural networks can learn to use secret keys to protect information when they communicate with one another.

— Daniel Tunkelang, formerly of LinkedIn and Google, is skeptical that we can automate ML.

— If you’re inclined to feel that convolutional neural networks get way too much buzz and recurrent neural networks are underappreciated, William Vorhies is on your side.



— Kaggle launches a product recommendation challenge for Santander Bank. Submissions due 11:59pm Wednesday, 21 December UTC. First place gets $30K; second place gets $20K; third place gets $10K; fourth place gets a set of steak knives.


— Twitter says it will use ML to make your timeline more relevant. No word on whether they plan to use ML to decide who to fire.

— Netradyne announces three new DL-driven applications for its Drivieri fleet safety platform: traffic light detection, relative speed determination, and pedestrian identification.

— To defend against all of those lawsuits we can expect from the introduction of autonomous vehicles, Intraspexion introduces a DL-driven application to identify potential litigation risks.

— IBM adds “cognitive behavioral biometrics” to its digital banking fraud prevention technology. That’s where they measure how you interact with the keyboard and mouse to create a biometric profile.

— Datadog introduces ML-based anomaly detection to help engineering teams detect abnormal behavior in a cloud environment. When the software detects unusual behavior, it barks. Just kidding.

— Parker Beauchamp surveys AI applications in the insurance industry.

— Purdue researchers develop a DL-driven application to assess disaster damage from images.

— Yahoo Tech reports on DL-driven robots that cut meat.

ML/DL Software and Services

— Databricks adds support for DL with TensorFrames, a Spark package that enables TensorFlow to run on Spark. The managed service includes support for TensorFrames with GPU-accelerated compute instances. Alex Woodie reports, noting that Spark ML runs 10X faster on GPUs.

Hardware and Provisioning

— Scientific Computing World reports on the challenges of managing an HPC cluster for applications like deep learning.

— Rich Brueckner asks if deep learning will scale to supercomputers. The answer seems to be “yes”.

Leave a Reply

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

You are commenting using your 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 )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.