Michael.Walker's blog

A new curated list of medical data for machine learning is available here.

In addition, Stanford is developing a petabyte-scale, cloud-based, multi-institutional, searchable, open repository of diagnostic imaging studies for developing intelligent image analysis systems - called Medical Image Net.

These curated data sets are great for experimenting - please respect data usage restrictions for each data set.

Today data scientists are using new techniques with machine learning to solve complex challenges. In applied data science speed kills and the Tesla V100 accelerator is built for speed. Data scientists always make trade-offs between accuracy and time. Thus, more powerful compute systems are required to crunch and analyze diverse data sets, and train exponentially more complex deep learning models in a practical amount of time.

Google's 2nd generation tensor processor has four 45 teraflops chips in a 180 TFLOPS Tensor Processor Unit (TPU) - available for use in the cloud later this year. Google calls them Cloud TPUs - initially available on the Google Compute Engine.

The Data Science Association has posted a wide variety of TensorFlow tutorial videos here.

MapD Core is an in-memory, column store, SQL relational database designed to run on GPUs (also runs on CPUs). It is now available free - licensed under the Apache License, Version 2.0.

Within a single server, MapD can handle datasets 1.5-3TB in raw size in GPU RAM, or 10-15TB in CPU RAM. In distributed deployments, that number scales linearly.

Qualcomm's new Snapdragon 820 chip (used in many high-end smartphones) can be used for deep learning and neural network applications. The Snapdragon Neural Processing Engine is an SDK powered by the Zeroth Machine Intelligence Platform.

The Google Cloud Platform will offer GPUs in 2017 for Google Compute Engine and Google Cloud Machine Learning with the AMD FirePro S9300 x2 and NVIDIA Tesla P100 and K80 GPUs. Up to 8 GPU dies can be attached per VM instance including custom machine types.

The Radeon Open Compute Platform (ROCm) is an open source platform for GPU computing that is language independent and brings modular software development to GPU computing. This provides a real cheaper alternative to Nvidia's CUDA and helps developers in coding compute-oriented software for AMD Radeon GPUs along with converting existing CUDA software to run on GCN hardware.

The Nvidia DGX-1 is a new HPC system (not just a server) that features the Tesla P100 accelerators for GPU computing. It includes 2x Intel Xeon E5-2698 v3 (16 core, Haswell-EP) and 8 P100s for 28,672 CUDA
cores and 128GB of shared VRAM. The DGX-1 is rated to be able to hit 170 FP16 TFLOPs of performance (or 85 FP32 TFLOPs) inside of 3Us.