The Masters of Uncertainty

By Nicole Hemsoth

September 13, 2013

According to Dr. Houman Owhadi and Dr. Clint Scovel, both from the California Institute of Technology, Bayesian methods are becoming more prevalent as high performance computing advances continue. In this special audio-based feature interview, we talk with both about what these methods will contribute to a number of research and enterprise endeavors, what computational requirements exist as we move toward more advanced questions, and how the field is evolving—and will continue to evolve with exascale (or even quantum) class systems.

HPCwire: In the context of high performance computing the two of you have argued recently that Bayesian methods are becoming even more popular than they ever were before to quantify uncertainty both in science and industry. What is it about these methods that are necessary and even more necessary as we move towards ever more advanced systems?

Owhadi: Bayesian inference goes all the way back to a formula discovered by Reverend Thomas Bayes, and when he found that formula, Pierre-Simon Laplace took his research and developed further into a field called Baysian Inference. This started a 250 year old controversy. What is the controversy about? In Bayesian inference, you have some prior about what reality could be, then you condition your prior with some data that you observe. This is basically the Bayes’ rule.

Now Pierre-Simon Laplace took that one step further where he said, ok, I don’t really need to have an exact prior we’re representing – an exact measure corresponding to what reality could be. I could just make up a prior, or a belief about what it could be, then I could use Bayes’ formula to update my belief. And this started the field that we know today as Bayesian inference.

In the 50’s, Bayesian inference was mainly looked at as a curiosity because we couldn’t really compute those Bayesian posteriors for complex systems. But now with the advent of high performance computing, we can actually compute those posterior probabilities. Since Bayesian inference is also an elegant and simple way of combining information with beliefs, it’s becoming increasingly popular.

HPCwire: Dr. Scovel to build on what he just said – HPC is so often considered to be about increasing fidelity and resolution, moving as close to reality as possible, so where does uncertaintly fit in in the next generation of systems and applications.

Scovel: Well certainly no one believes that those systems compute exact, anything. There’s always an error, and so having some confidence in what the results are is always going to be of interest. There’s another way that these things all fit together, and that is that not only is uncertainty quantification useful for high performance computing, but high performance computing is useful for uncertainty quantification; because that computation that he was saying that has just come about in the 50’s is basically about our ability to do these numerical computations – in particular Markov chain Monte Carlo simulations to compute these posteriors.

HPCwire: Let’s talk about what areas of industry and scientific computing most important. Where is this most valuable?

Owhadi: Risk analysis. Climate modeling. Take for instance Boeing. So when Boeing is developing a new plane, most of the budget goes into the safety assessment of their new model. What you have to understand with respect to that industry is that they have to certify that their new model of airplane has a probability of catastrophic event that is smaller than 10 to the power of minus nine per hour of flight. Now this is really small.

And of course, they cannot fly one billion airplanes and just see how many crash, so they have to assess the safety of their system and they have limited information. Take another example, you are JPL and you wanted to design your new satellite and you want your satellite to go around the planet in the solar system, and you are spending a lot of money for it. How do you certify that your system is not going to crash? One way is to build 1,000 of these satellites and just count how many crash, but that would be too costly. So you have to do it with a very limited amount of data. This has created a new field called uncertainty quantification, which is an emerging field. It is basically a field that is at the interface between probability and statistics and computer science.

It mainly has to do with engineering systems characterized by low number of samples, and complex information. The way we are seeing that it should be pushed forward is basically to be able to process information in an optimal way, to assess the risk in an optimal way, without making assumptions that may not be true, and without ignoring relevant information.

So let me explain – at the end of the day our point of view is that you cannot really say if a piece of information or piece of data is accurate or not unless you test it. But once this information is given to you, the best thing that you can do is to process it in an optimal way. Basically what we are striving to do is to develop an algorithmic framework to allow us to do just that: process information in an optimal way. Now you can imagine that there are plenty of places where you can apply these things.

HPCwire: Dr. Scovel, I believe this leads into your pet project right now which is the scientific computation of optimal estimators. Where does that fit into this conversation we’re having. Can you describe it more thoroughly.

Scovel: Yes, when he was talking doing this in the optimal way, the first question is what does that mean. Instead of providing solutions, the first thing that we do is actually formulate a problem that incorporates everybody’s – the customers objectives that they’re interested in, the available information, what we know about the information, what the domain experts know about the information et cetera – and then you formulate this optimization problem which essentially defines what it means to be an optimal solution to this question that you’ve asked – like how reliable is that satellite going to be.

Where this is new is that historically what has been done, you provide some model for this process and you see what happens with the model. Ours is different. We’re saying we want to formulate this problem that we’re trying to solve and we’re going to use our computing capability – in particular, high performance computing – to solve these problems. I think historically is very similar to Bayes methods in the past. Historically, the reason why people didn’t go down this path is we didn’t have the computing power to do it, but I think that we now do have the computing power to actually solve these problems defining optimal estimation problems, or optimal prediction problems given optimal being optimal over some set of assumptions that we’re all willing to agree on.

HPCwire: I noticed that both of you have, just in your body of research, talked about solving exascale class problems. Those are the types of systems you’re looking at to be able to do this at a much higher level, is that correct?

Owhadi: Yes. I could take another analogy here – 200 years ago, if I were to ask you to solve a partial differential equation, you wouldn’t use a computer, you would probably use your brain. You would probably not come with a quantitiative estimate of the solution, but only qualitative estimates.

Now, if ask you the same question today, you will not use your brain to solve the partial differential equation – you will use a computer. But you will still use your brain to program the computer that will crunch numbers for you to solve the partial differential equation. This paradigm shift can be traced back to seminal work by John Von Neumann and (Herman) Goldstein in the 50’s, and humans organized as computers in the beginning of the previous century.

Now today if I asked you to find a statistical estimator, or to find the best possible climate model, or to find a test that will tell me if some data that I’m observing is corrupted or not – you’re not going to use a computer to do that. You’re going to use your brain and guess work. What we want to do here is basically turn this guesswork into an algorithm that we’ll be able to implement on a high performance cluster.

If you look at the mathematics behind this problem of turning the process of scientific discovery into an algorithm, you can basically translate it into an optimization problem, but optimization variables are not discrete. They’re not zeros and ones – they’re basically infinite dimensional objects. What we have found is a new form of calculus that allows us to turn these infinite dimensional optimization problems into finite dimensional optimization problems that we can start solving on computers.

Even after reduction these optimization problems are extremely large, so that’s why we believe that we’ll probably need exascale or petascale machines for solving these kind of problems for complex systems.

HPCwire: What’s interesting in that conversation about the systems required, I recently talked to, I believe the only, quote, “quantum computer company,” D-Wave recently – and this exact type of optimization problem – this best of all worlds solution is exactly the sweet spot for quantum cmputers. Do you believe those actually exist, and if so, are they a good fit for the types of problems you’re seeking to solve?

Owhadi: This is interesting because with respect to exascale machines – some people believe that there are two ways to approach high performance computing. The first way is just to solve the same kind of problems, but bigger problems.  So for instance if you’re interested in climate modeling, you’re still going to do climate modeling. You still are going to run your model, but with a finer mesh, with a finer resolution, and hopefully instead of predicting the weather for four days ahead, you’ll be able to do it for five days out.

What we envision here is some kind of paradigm shift where instead of numerically solving a bigger model, you actually use your computer to find the model itself. So yes, if there are quantum computers out there, that would be a great thing for this new kind of framework.

Scovel: The more computing power we can have the better our success is going to be. 

Owhadi: Information doesn’t necessarily come in the form of zero and ones. If you look at the interlaying optimization problems, they involve optimization variables that are measures of probability and function and these objects live in infinite dimensional spaces. Calculus on a computer is necessarily discrete and finite. So the first step of the technology is mathematical.  You have to be able to manipulate – to come up with a new form of calculus that is able to manipulate these infinite dimensional objects. This is basically what we’ve done. This new form of calculus allows us to take these huge optimization problems and turn them into something discrete and finite that we can start solving on a computer.

Scovel: It’s also more than that in the sense that it’s not just a question of computing power. Part of this program and this paradigm shift that we’re talking about is actually coming up with formulation of what it means to be an optimal solution to these things. That’s actually a big part of the program. It’s not just, I know what I want to compute and I need more computing power. It’s actually, what do we want to compute and what does it mean to be the best.

I think the complexity involved into the formulation of these problems – and these formulations for these problems of what it means to be an optimal predictor or an optimal estimator requires communication from all levels of the effort – from the customer to the project leader to the domain expert, the material science domain experts, the statisticians – instead of for example in many places you do a bunch of runs, you do a bunch of modeling, you do a bunch of stuff, and then you hand all the results off to a statisticians and you want them to put that stuff together.

We’re saying, no you need to do that all together. You need to have everybody communicating so you’re formulating not only what the objectives that you’re interested in but you formulate what pieces of information you have good confidence about, and those establish a quantitative set of realistic paradigms. Then the optimization now proceeds as, ok, now we have this huge optimization problem, how do we reduce it analytically, how do we take those reduced analytic problems, and how do we implement them on the computer, and how do we know we’re done – that’s sort of the picture.

Owhadi: Let me give you two examples here.

The first one will be investing in the stock market. Question: is there an optimal way to invest in the stock market. Currently, it’s not clear how to turn this into an optimization problem, but with the framework that we are developing, we think that we’ll be able to turn this into an optimization problem and reduce it to something that we can start solving with a high performance cluster. The idea is to invest in an optimal way given the limited information that you have at hand.

Let me give you another example. Consider the game of chess. We have computers playing chess, and they play chess very well. Question: can you have a computer play an information game – an information war? So for instance, you have two armies and they are playing this information war – or you have two companies, and each company has limited information about the other company and they have some decisions to make with respect to what the other company could or could not know and could or could not do. If you look at this as just an information game, but the board is not just composed of 64 squares, and the moves are not discrete, they could be anything.

If you want to start addressing these kind of problems, you need some new mathematics that allows you to manipulate these pieces of information that do not live on a chess board. This is basically the first step in our technology to develop this new mathematics.

HPCwire: Where do you see this maybe revolutionizing certain approaches to computing down the road?

Owhadi: I think that eventually we will use computers to help the process of scientific discovery, but not just to solve mathematical equations, but to develop the mathematical equations themselves. We see this going into the field of machine learning. In the field of machine learning, if you want to develop an intelligent agent, you basically decompose the tasks that you ask the computer to do into small steps, and you chew the work for the computer.

What this technology that we are developing will allow to do – this is basically a long term vision – is give the ability to a computer to develop a model of reality, act on that model of reality, and update the model based on feedback that it is receiving from reality. This is basically a change in machine learning.

Subscribe to HPCwire's Weekly Update!

Be the most informed person in the room! Stay ahead of the tech trends with industry updates delivered to you every week!

Q&A with Nvidia’s Chief of DGX Systems on the DGX-GB200 Rack-scale System

March 27, 2024

Pictures of Nvidia's new flagship mega-server, the DGX GB200, on the GTC show floor got favorable reactions on social media for the sheer amount of computing power it brings to artificial intelligence.  Nvidia's DGX Read more…

Call for Participation in Workshop on Potential NSF CISE Quantum Initiative

March 26, 2024

Editor’s Note: Next month there will be a workshop to discuss what a quantum initiative led by NSF’s Computer, Information Science and Engineering (CISE) directorate could entail. The details are posted below in a Ca Read more…

Waseda U. Researchers Reports New Quantum Algorithm for Speeding Optimization

March 25, 2024

Optimization problems cover a wide range of applications and are often cited as good candidates for quantum computing. However, the execution time for constrained combinatorial optimization applications on quantum device Read more…

NVLink: Faster Interconnects and Switches to Help Relieve Data Bottlenecks

March 25, 2024

Nvidia’s new Blackwell architecture may have stolen the show this week at the GPU Technology Conference in San Jose, California. But an emerging bottleneck at the network layer threatens to make bigger and brawnier pro Read more…

Who is David Blackwell?

March 22, 2024

During GTC24, co-founder and president of NVIDIA Jensen Huang unveiled the Blackwell GPU. This GPU itself is heavily optimized for AI work, boasting 192GB of HBM3E memory as well as the the ability to train 1 trillion pa Read more…

Nvidia Appoints Andy Grant as EMEA Director of Supercomputing, Higher Education, and AI

March 22, 2024

Nvidia recently appointed Andy Grant as Director, Supercomputing, Higher Education, and AI for Europe, the Middle East, and Africa (EMEA). With over 25 years of high-performance computing (HPC) experience, Grant brings a Read more…

Q&A with Nvidia’s Chief of DGX Systems on the DGX-GB200 Rack-scale System

March 27, 2024

Pictures of Nvidia's new flagship mega-server, the DGX GB200, on the GTC show floor got favorable reactions on social media for the sheer amount of computing po Read more…

NVLink: Faster Interconnects and Switches to Help Relieve Data Bottlenecks

March 25, 2024

Nvidia’s new Blackwell architecture may have stolen the show this week at the GPU Technology Conference in San Jose, California. But an emerging bottleneck at Read more…

Who is David Blackwell?

March 22, 2024

During GTC24, co-founder and president of NVIDIA Jensen Huang unveiled the Blackwell GPU. This GPU itself is heavily optimized for AI work, boasting 192GB of HB Read more…

Nvidia Looks to Accelerate GenAI Adoption with NIM

March 19, 2024

Today at the GPU Technology Conference, Nvidia launched a new offering aimed at helping customers quickly deploy their generative AI applications in a secure, s Read more…

The Generative AI Future Is Now, Nvidia’s Huang Says

March 19, 2024

We are in the early days of a transformative shift in how business gets done thanks to the advent of generative AI, according to Nvidia CEO and cofounder Jensen Read more…

Nvidia’s New Blackwell GPU Can Train AI Models with Trillions of Parameters

March 18, 2024

Nvidia's latest and fastest GPU, codenamed Blackwell, is here and will underpin the company's AI plans this year. The chip offers performance improvements from Read more…

Nvidia Showcases Quantum Cloud, Expanding Quantum Portfolio at GTC24

March 18, 2024

Nvidia’s barrage of quantum news at GTC24 this week includes new products, signature collaborations, and a new Nvidia Quantum Cloud for quantum developers. Wh Read more…

Houston We Have a Solution: Addressing the HPC and Tech Talent Gap

March 15, 2024

Generations of Houstonian teachers, counselors, and parents have either worked in the aerospace industry or know people who do - the prospect of entering the fi Read more…

Alibaba Shuts Down its Quantum Computing Effort

November 30, 2023

In case you missed it, China’s e-commerce giant Alibaba has shut down its quantum computing research effort. It’s not entirely clear what drove the change. Read more…

Nvidia H100: Are 550,000 GPUs Enough for This Year?

August 17, 2023

The GPU Squeeze continues to place a premium on Nvidia H100 GPUs. In a recent Financial Times article, Nvidia reports that it expects to ship 550,000 of its lat Read more…

Shutterstock 1285747942

AMD’s Horsepower-packed MI300X GPU Beats Nvidia’s Upcoming H200

December 7, 2023

AMD and Nvidia are locked in an AI performance battle – much like the gaming GPU performance clash the companies have waged for decades. AMD has claimed it Read more…

DoD Takes a Long View of Quantum Computing

December 19, 2023

Given the large sums tied to expensive weapon systems – think $100-million-plus per F-35 fighter – it’s easy to forget the U.S. Department of Defense is a Read more…

Synopsys Eats Ansys: Does HPC Get Indigestion?

February 8, 2024

Recently, it was announced that Synopsys is buying HPC tool developer Ansys. Started in Pittsburgh, Pa., in 1970 as Swanson Analysis Systems, Inc. (SASI) by John Swanson (and eventually renamed), Ansys serves the CAE (Computer Aided Engineering)/multiphysics engineering simulation market. Read more…

Choosing the Right GPU for LLM Inference and Training

December 11, 2023

Accelerating the training and inference processes of deep learning models is crucial for unleashing their true potential and NVIDIA GPUs have emerged as a game- Read more…

Intel’s Server and PC Chip Development Will Blur After 2025

January 15, 2024

Intel's dealing with much more than chip rivals breathing down its neck; it is simultaneously integrating a bevy of new technologies such as chiplets, artificia Read more…

Baidu Exits Quantum, Closely Following Alibaba’s Earlier Move

January 5, 2024

Reuters reported this week that Baidu, China’s giant e-commerce and services provider, is exiting the quantum computing development arena. Reuters reported � Read more…

Leading Solution Providers

Contributors

Comparing NVIDIA A100 and NVIDIA L40S: Which GPU is Ideal for AI and Graphics-Intensive Workloads?

October 30, 2023

With long lead times for the NVIDIA H100 and A100 GPUs, many organizations are looking at the new NVIDIA L40S GPU, which it’s a new GPU optimized for AI and g Read more…

Shutterstock 1179408610

Google Addresses the Mysteries of Its Hypercomputer 

December 28, 2023

When Google launched its Hypercomputer earlier this month (December 2023), the first reaction was, "Say what?" It turns out that the Hypercomputer is Google's t Read more…

AMD MI3000A

How AMD May Get Across the CUDA Moat

October 5, 2023

When discussing GenAI, the term "GPU" almost always enters the conversation and the topic often moves toward performance and access. Interestingly, the word "GPU" is assumed to mean "Nvidia" products. (As an aside, the popular Nvidia hardware used in GenAI are not technically... Read more…

Shutterstock 1606064203

Meta’s Zuckerberg Puts Its AI Future in the Hands of 600,000 GPUs

January 25, 2024

In under two minutes, Meta's CEO, Mark Zuckerberg, laid out the company's AI plans, which included a plan to build an artificial intelligence system with the eq Read more…

Google Introduces ‘Hypercomputer’ to Its AI Infrastructure

December 11, 2023

Google ran out of monikers to describe its new AI system released on December 7. Supercomputer perhaps wasn't an apt description, so it settled on Hypercomputer Read more…

China Is All In on a RISC-V Future

January 8, 2024

The state of RISC-V in China was discussed in a recent report released by the Jamestown Foundation, a Washington, D.C.-based think tank. The report, entitled "E Read more…

Intel Won’t Have a Xeon Max Chip with New Emerald Rapids CPU

December 14, 2023

As expected, Intel officially announced its 5th generation Xeon server chips codenamed Emerald Rapids at an event in New York City, where the focus was really o Read more…

IBM Quantum Summit: Two New QPUs, Upgraded Qiskit, 10-year Roadmap and More

December 4, 2023

IBM kicks off its annual Quantum Summit today and will announce a broad range of advances including its much-anticipated 1121-qubit Condor QPU, a smaller 133-qu Read more…

  • arrow
  • Click Here for More Headlines
  • arrow
HPCwire