Safe Harbor Statement: The information on IBM products is intended to outline IBM's general product direction and it should not be relied on in making a purchasing decision. The information on the new products is for informational purposes only and may not be incorporated into any contract. The information on IBM products is not a commitment, promise, or legal obligation to deliver any material, code, or functionality. The development, release, and timing of any features or functionality described for IBM products remains at IBM's sole discretion.
Tony Pearson is a an active participant in local, regional, and industry-specific interests, and does not receive any special payments to mention them on this blog.
Tony Pearson receives part of the revenue proceeds from sales of books he has authored listed in the side panel.
Tony Pearson is not a medical doctor, and this blog does not reference any IBM product or service that is intended for use in the diagnosis, treatment, cure, prevention or monitoring of a disease or medical condition, unless otherwise specified on individual posts.
IBM Technical University in Orlando 2018 Day 2 Morning Sessions
This week, I am in Orlando, Florida for the [IBM Technical University], with focus on IBM storage, IBM Z mainframes and IBM Power servers. Here is my recap for morning break out sessions on Day 2.
A Survey of Deep Learning Techniques
Nin Lei, IBM Distinguished Engineer, presented a sample of Deep Learning techniques used today. CNN, RNN, and GAN.
Basic decision making: gather data, reviewed by subject matter expert, have an outcome. This is done for a variety of situations: fraudulent vs. legitimate credit card transaction, approve or reject loan application, tumor is benign or malignant. Machine Learning effectively replaces SME with a mathematical function.
Various tools are available for this: Tensorflow, SnapML, SAS, SPSS, are just a few.
Deep Learning is based on "Neural Networks", a subset of Machine Learning. There are input layer, one or more hidden layers, and then an output layer. For example, for a photo, each pixel could be an input feature. A 200x200 pixel photo represents 40,000 input values. In the past, there weren't more than three hidden layers. Today, we can have 20 to 50 layers, because we now have more computational power, with 95-97 percent accuracy.
For each connection between input layers and hidden layers, and output layers, you identify weights and biases. A research paper by Hornik 1989 posits that any machine learning can be performed by a sufficiently large neural network.
Convolution Neural Network (CNN) is often used for image recognition, for object classification or detection.
Some features are invariant. Location invariant means it doesn't matter where it is located within the photo. Color invariant means it does not matter what color it is, and can work with black-and-white or grayscale photos.
For example, for facial recognition, earlier layers are focused on identifying edges, and later layers identify facial features like eyes, nose and mouth.
Image recognition is used with self-driving cars, drones to determine power line maintenance or crop inspection, social media, video surveillance, medical image diagnosis, car racing, and ripeness of fruits and vegetables.
CNN is used for auto-encoding. This takes detailed photos, compresses them, and then can be used to decode back to something similar. It can takes weeks to train a model with a million photo images.
Recurrent Neural Network (RNN) is focused on time sequence.
This is useful for predicting sequences of letters or words. However, since mathematics are involved, a long sequence of multiplications will either get to zero or infinity, this is known as the "vanishing gradient problem".
The solution is "Long Short Term Memory" (LSTM) cells. Basically, the model selectively remembers information from previous steps, which reduces the number of multiplications.
RNN need to know related words. For example, men-women, king-queen, walking-walked, swimming-swam, Spain-Madrid. These are referred to as "embeddings", which are stored in the hidden layers for quick lookup.
Generative Adversarial Networks (GAN) are used to generate fake photos to train other models.
Sometimes, you do not have enough photos in each category for training, so you can generate fake images to help with the training system. Noise is fed into a "Generator" model, and then the results are evaluated by a "Discriminator" model, comparing the fake with real photos. Repetition allows each model to improve so that the fake photos become more realistic for training purposes.
The death of the one-size-fits-all cloud: The mainstreaming of multi-arch
Elise Spence and Drew Thorstensen, IBM Power Systems for Software Defined Cloud Infrastructure, presented this topic. The session was on IBM Cloud Private, and the multiple architectures supported by Docker and Kubernetes.
There are actually six different architectures supported for Docker containers:
While containers are "portable" between systems, the binaries are typically only written for a single architecture, typically Linux-x86 or Windows-x86, and won't run on POWER or IBM Z.
The solution is to create a multi-arch manifest file, and port all the binaries to all of these different architectures. This way, when the containerized application is run on POWER, the manifest will identify the POWER-based binaries.
Introduction to IBM Cloud Object Storage (powered by Cleversafe)
Before 2015, IBM offered two "Object Storage" products: IBM Spectrum Scale and IBM Spectrum Archive, and I was constantly having to compare and contrast IBM products to Cleversafe.
Not any more! With the IBM acquisition of Cleversafe, IBM now offers all three!
This session explained all of the features and functions of IBM Cloud Object Storage System, available as software, as pre-built systems, including a VersaStack CVD, and as Storage-as-a-Service (STaaS) in the IBM Cloud.
(IBM renamed Cleversafe DSnet to "IBM Cloud Object Storage System". I joked that if IBM ever acquired Coca-Cola, they would probably rename their signature soft drink as the "Brown Carbonated Sugar Liquid", or BroCarb SugarLiq for short!)
I provided a general overview, as well as the latest features of Concentrated Dispersal Mode and Compliance Enabled Vaults.
You can follow along with Twitter hashtag #IBMtechU, or follow me at @az990tony.