Industry News Details

Interview with Victor Jakubiuk , Chief Science Officer, OnSpecta Inc - Speaker at 5th Annual Global Big Data Conf Posted on : Jul 20 - 2017

We feature speakers at 5th Annual Global Big Data Conference - August 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 Victor Jakubiuk , Chief Science Officer, OnSpecta Inc (Topic : Debunking The Myth Of 100x GPU Vs. CPU)

Interview with Victor Jakubiuk

1. Tell us about yourself and your background.
I am a co-founder and Chief Scientist of, an early stage deep learning startup. We’ve created a neural networks optimization platform for CPUs and custom chips.  Previously I was a computational neuroscience researcher at MIT and a founder of a fintech startup.

2.  What have you been working on recently?
Recently, Deep Learning Server (DLS), which is our just-in-time neural network compiler for Caffe, that speeds up inference 10x. We’ve released DLS for Intel CPUs, and our ARM implementation is in private beta now. I’ve been working with a number of customers integrating our solution into their processes.

3. Tell me about the right tool you used recently to solve customer problem?
I try to use the right tool for the problem at hand. Analyzing a problem from the first principles oftentimes surfaces beautiful and simple solutions, not requiring industry-standard tools. Recently, I’ve been working a lot with Caffe and Tensorflow.

4. Where are we now today in terms of the Big data, and where do you think we’ll go over the next five years?
Most of my work is at the intersection of big data and machine learning. I might be somewhat biased, but I think the world is shifting away from just gathering data, and moving towards providing “intelligent” and actionable outputs. I believe we’ll see a lot of interesting applications of supervised (and unsupervised) deep learning and reinforcement learning on large, already accumulated data sets. Particularly in certain enterprise verticals.

5. You’ve already hired Y number of  people approximately. What would be your pitch to folks out there to join your Organization? Why does your organization matter in the world?
We’re an early stage startup, so each hire has tremendous influence on the product’s direction and our company’s culture. We’re doing a lot of fundamental research in software performance engineering, which is not something readily done elsewhere. With a high-velocity product roadmap and the market fit we’re seeing, this is a very exciting ride!

Our thesis is somewhat counter-intuitive to the industry’s outlook: looking into the future, we see the world with heterogeneous chips for deep learning (CPUs, ASICs etc.) that doesn’t include the GPU computational model (not as it is now, at least). That’s why we’re building our virtualization layer, which will become the unifying deep learning platform that abstracts away hardware differences and inefficiencies.

6. What are some of the best take aways that the attendees can have from your workshop on "Debunking The Myth Of 100x GPU Vs. CPU"?
You’ll learn how to optimize Caffe (and Tensorflow) across various hardware architectures. I’ll show you why the GPUs are a commonplace in deep learning these days, and also where they are not necessarily. You’ll learn how to re-purpose your commodity CPU hardware for deep learning.

7. What are the top 5 Big data Use cases in enterprises?
Finance, e-commerce, healthcare, autonomous-navigation and agriculture. But there are many, many more.

8. Which company do you think is winning the global Big Data race?
That depends on the vertical one is trying to win! The top, big tech companies (Google, Facebook, Tencent etc.) are certainly sitting on troves of consumer data. New startups will win certain niche markets, especially with the combination of big data and machine learning feedback-loop.

9. Any closing remarks
Thank you for giving me the opportunity to share OnSpecta’s view on the deep learning and big data space. I’m hoping for an exciting and interactive session!