We feature speakers at Global Predictive Analytics Conference March 27 - 29, 2017 to catch up and find out what he or she is working on now and what's coming next. This week we're talking to Dr. Avidan Akerib, VP Technologies, GSI Technology
1. Tell us about yourself and your background.
I am the VP of the Associative Computing business unit at GSI Technology, and I have a PhD in Applied Mathematics and Computer Science from the Weizmann Institute of Science in Israel. I invented a breakthrough associative processing technology and have more than 30 patents related to In-Place Associative Computing technology.
2. What have you been working on recently?
We have been working on an in-place associative computing technology that changes the concept of computing from standard classical data processing, where data is moved back and forth between the processor and memory, to in-place data processing that utilizes extreme-parallelism (millions of processors per chip), that can compute, and search in-place directly in the memory array. By removing the bottleneck between the processor and memory, a significant improvement in the performance-over-power ratio is achieved.
This breakthrough technology enables extracting information from Big Data and using it to predict trends and behavior patterns. The technology is particularly suited for solving predictive analytics algorithms such as data mining, deep learning, k-NN etc., in real time by moving the processing from CPU to the memory array. In addition to solving the IO bottleneck issue, the in-place associative processor can implement additional functions that do not exist in standard CPU/GPU. For example, a search-query can be presented to the associative processor and a result is generated in O(1). Unlike with current solutions, there is no need to write the result back to memory since the result is already in the associative processor, and in-place flags prepare the data for the next stage of processing.
3. Where are we today in terms of the state of Predictive Analytics, and where do you think we’ll go over the next five years?
The availability of Big Data and the rapid development of deep learning and data mining algorithms is helping to drive the growth of predictive analytics at a rapid pace. However, this progress might start to hit a brick wall because current HW architectures for predictive analytics can’t keep up with the exponential growth in the amount of big data. Our industry will start to realize that we need to break through some of the limitations. “More is better” is good, but it is not enough. Beyond 2017, we will start to see new types of HW solutions that will be able to keep up with orders of magnitude more data. We will see new algorithms and applications that take advantage of new conceptual frameworks.
In 5 years, predictive analytics will play an essential role in our lives. It will assist in identifying diseases, providing multiple types of recommendation, political assessments, optimizing marketing campaigns, shortening drug development, and more.
4. What are some of the best takeaways that the attendees can have from your "In Place Associative Predictive Analytics Computing" talk?
a) GSI’s APU technology allows for massive parallel data processing on a different scale by providing millions of parallel processors on a chip
b) The technology scales to 1M queries of cosine similarity searches/second for billions of records
c) The architecture accesses randomly distributed data by content not by address and allows for real-time predictive analytics
d) The technology provides orders of magnitude improvement for sparse matrix – vector multiplication and allows for finding k-minimum values in O(1)
We are developing a new type of parallel processor for predictive analytics that will change performance and applications from end to end. This requires expertise in all disciplines such as writing primitive language code using only two instructions (associative read and write), writing higher-level libraries by developing new tools such as a compiler for associative processing, and providing high-level customer interfaces, such as Tensorflow. This requires recruiting highly motivated professionals who are determined to change the world of computing and are creative engineers characterized by out-of-the-box thinking.
It is rare that a totally new paradigm challenges the existing order. The potential to benefit humanity in AI in general is unprecedented. Here is an opportunity to create something new that will provide the quantum leap needed to fulfill its promise.
6. What are the top 5 Predictive Analytics Use cases in enterprises?
Fraud detection and credit risk scoring
Sales prediction and customer segmentation for marketing
7. Which company do you think is winning the global Predictive Analytics race?
Many of the large companies in e-commerce and social media, such as Amazon, eBay, Facebook, Netflix and Twitter are winning the predictive analytics race.
8. Why is ‘in-place associative computing' important for the industry?
With in-place associative computing, data is searched efficiently by content rather than address and the IO bottleneck between the processor and memory is removed. This significantly increases performance and lowers power. Processing is integrated directly in the data itself, turning “dumb cache” into “massively parallel intelligent cache” with millions of parallel processors.
Our industry is poised to change the world. In-place processing is a key technology that will help ensure it happens.