Better Machine Learning When Data is Noisy and Incomplete Using Quantum Computers

D-Wave Systems had a presentation that summarized the utility of their quantum adiabatic system. It is a specialized QUBO solver. QUBO stands for quadratic unconstrained binary optimization and is a problem type traditionally used in computer science.

There is online documentation and training to understand how to use the D-Wave 2048 qubit system.

Probablistic Machine Learning

Most of the transformation that AI has brought to-date has been based on deterministic machine learning models such as feed-forward neural networks. The real world has a lot more uncertainty. Probabilistic models explicitly handle this uncertainty by accounting for gaps in our knowledge and errors in data sources.

Probabilistic modeling is a practical approach for designing machines that:

* Learn from noisy and unlabeled data
* Define confidence levels in predictions
* Allow decision making in the absence of complete information
* Infer missing data and latent correlations in data

D-Wave’s Quadrant’s algorithms enable accurate discriminative learning (predicting outputs from inputs) using less data by constructing generative models which jointly model both inputs and outputs. Quadrant offers the services of its in-house experts to help customers get the benefit of leading-edge machine learning solutions.

In May 2018, Quadrant announced working with Siemens Healthineers, a global leader in medical technology company. Siemens Healthineers and D-Wave took first place in the CATARACTS medical imaging grand challenge, using Quadrant’s generative machine learning algorithms to identify surgical instruments in videos with high accuracy. These algorithms are being researched as a way to improve patient outcomes through better augmented surgery and ultimately computer-assisted interventions (CAI).

“Machine learning has the potential to accelerate efficiency and innovation across virtually every industry. Quadrant’s models are able to perform deep learning using smaller amounts of labeled data, and our experts can help to choose and implement the best models, enabling more companies to tap into this powerful technology,” said Handol Kim, Sr. Director, Quadrant Machine Learning at D-Wave.

8 thoughts on “Better Machine Learning When Data is Noisy and Incomplete Using Quantum Computers”

  1. “Peter Fairley is a freelance energy journalist based in San Francisco and Victoria, BC.”

    Total bullshit from a obvious greenpeace hack.

    A real expert indeed.

    Note he forgot to mention the purchase of two Russian units a few months ago.

    No doubt one of the Pulser’s colleagues working out off Big Oil’s propaganda headquarters.

  2. “then why is their fleet capacity factor quoted as 77% in this article?”

    I don’t know but I would point out that their reactor fleet is a grab bag of reactors. They have two of everything from the nuclear power buffet.

  3. Unless you have other information, the condenser had a structural failure, which conflicts with your statement about it being cannibalized.

    “[S]ome of the structural supports within the condenser did not operate as designed,” Hopson said. Those supports must be repaired and reinforced as part of a plan to return the condenser to service, he said.”

    So, if the Chinese, who are pretty opaque with their reporting, are not having teething problems with new equipment, then why is their fleet capacity factor quoted as 77% in this article?

  4. At Watts Bar Unit 2, the situation existed that the plant was being canibalized continually to keep Watts Bar Unit 1 and the 2 units at Sequoya running. The constructors reported work completion with it being complete. That is why their condenser partially collapsed. Most places in the world don’t have the issues that those plants and the ones farther south and east have.
    Much of the problem with the U. S. effort is that the management teams that build them are driven by accelerating cost to cover unrealistically low bids. In other places, cost overruns are less accepted and legally tolerable.

  5. I’m sure their operator training is just fine. There is an established curriculum and simulator based training and high standards. That said, I heard by word-of-mouth that the Chinese did an ARI (all rods inserted) dilution to critical test in one of the units, which is indescribably, fundamentally, organizationally idiotic (if true).

    They will improve their 77% capacity factor over time. There are a lot of new units coming online and some learning happening. When Watts Bar 2 (Tennessee) came online in 2016 the condenser imploded after 5 months and the unit was offline for repairs for 4 months. There was a transformer fire before it went commercial too. Expect problems with new equipment.

  6. PISA tests show their high schoolers graduating three years ahead of ours in STEM subjects and add to that their five point IQ advantage and you get a labor force comfortable in the 21st century. A friend, a director-level employee with an engineering background who has worked with multiple multinational companies in various capacities, but has been primarily based in the US sees it this way (his emphasis):
    Most people don’t realize that the Toyota factory churning out cars has only half of its staff on the manufacturing floor. The other half is engineers and supply chain guys, supervisors etc…. the engineers at these facilities are responsible for fixing daily technical issues and working with R & D. The vast majority of modern manufacturing is done by machines.
    American manufacturing moved to China not because of dumb labor, but because you could hire high IQ people for dirt cheap. If your machine broke down, no problem; some Chinese guy (with basically a masters in EE) would pull out the circuit boards and using probes and other instrumentation determine what board needed replacing and he would work annually for a fraction of the salary of his equivalent in the US.
    Manufacturing in the US is a nightmare: at our facility our only requirement for a assembler was a high school degree, US citizenship, passing a drug and criminal background check and then passing a simple assembly test: looking at an assembly engineering drawing and then putting the components together.
    The vas

  7. I wonder (and Scaryjello may have useful input on this) how much of an issue is training the nuke power plant staff and retaining the trained workforce.

    When you are introducing new powerplants every year that means you need a lot of new, trained, staff. Which is an issue even for less critical skill areas in China (according to factory and service industry employers I deal with).

  8. From the article summary: “China has been improving the operational performance of their nuclear power.”

    Actually things have been sliding backwards on that basis. So when they sort that out, they should be able to hit closer to USA levels of capacity factors.

    “With pressure from both directions, even the nuclear plants now operating are underutilized. On average they used 81% of their generating capacity in 2017, 10% less than five years earlier, making the electricity they produce even more expensive.”

Comments are closed.