ICYMI: Top machine learning (ML) and deep learning (DL) stories from last week.
— On the AWS Compute Blog, Naveen Swamy and Joseph Spisak explain distributed deep learning with the AWS Deep Learning AMI. The AMI supports Caffe, CNTK/Microsoft Cognitive, MXNet, TensorFlow, Theano, and Torch. Product reviewers are underwhelmed, noting the absence of NVIDIA, CUDA, Anaconda or Python3 support.
— Cray unveils deep learning toolkits for the Cray XC and Cray CS-Storm systems. Kits include Caffe, Microsoft Cognitive Toolkit (CNTK), MXNet, NVIDIA DIGITS (Deep Learning GPU Traning System), TensorFlow, and Torch.
— IBM launches the PowerAI software kit to run on OpenPOWER LC servers for DL/AI workloads.
— Inspur announces the D1000 deep learning appliance based on NVIDIA Tesla GPUs, with support for the Caffe-MPI framework.
— In the Wall Street Journal, Don Clark reports that next year Intel will start shipping deep learning chips based on technology it acquired through its purchase of Nervana Systems. Stephanie Condon elaborates.
— Karl Freund reports on NVIDIA’s announcements at the SC16 show.
— BigML announces its fall release and offers a webinar on November 29 to review the new features.
— DataRobot, a machine learning company, partners with data blending leader Alteryx.
— In The Wall Street Journal, Burton Malkiel reviews Virtual Competition, a book about the darker side of the algorithmic economy. The book describes the use of algorithms to support price discrimination and collusion; it closes with a discussion of potential approaches to regulation.
— Matthew Mayo explains ensemble learners.
— In O’Reilly Radar, Reza Zadeh explains why deep learning is hard. In a word: optimization.
— Separately, S. Zayd Enam explains why machine learning is hard.
— On Slideshare, Makoto Yui describes Apache Hivemall, a machine learning library for Hive, Pig, and Spark.
— Microsoft’s David Smith explains how to call Microsoft Cognitive Services (CNTK) with R.
— Timothy Prickett Morgan examines NVIDIA’s recent growth.
— Also on The Next Platform, Nicole Hemsoth dissects Cray’s new XC50 supercomputer, built on the NVIDIA Pascal GPU-accelerated chip.
— Moor Insights & Strategy releases two white papers pertinent to machine learning. The first details AMD’s Radeon Open Compute Platform (ROCm), which supports GPUs in HPC and DL. The second explains Xilinx’ reconfigurable acceleration stack, an FPGA-based approach for machine learning workloads.
— In Forbes, Moor’s Karl Freund says AMD’s ROCm could be a game-changer.
— NVIDIA touts the energy efficiency of its DGX SATURNV supercomputer.
— Paul Alcorn offers additional detail on Intel’s new Xeon CPU and Deep Learning Inference Accelerator.
— In Fast Company, Christina Farr describes a joint project of General Electric and the University of California – San Francisco to develop a library of machine learning algorithms for diagnostics in primary care, pathology, and radiology.
— Scientists at UCLA and the University of Illinois use machine learning to discover and design alpha-helical membrane-active peptides, which may lead to improved cancer therapies.