“Q Entanglement Lab”, using HPC in quantum computing

To paraphrase Mark Twain: “Everyone talks about Quantum Entanglement but no one ever does anything about it“. Q entanglement is at the very heart of quantum computing and quantum information theory. And yet there still doesn’t exist a decent software library for calculating the q entanglement of small to very large quantum systems in either pure or mixed states. In this blog post, I announce that I will soon release such a software library (It’s almost finished. It will be called “Q Entanglement Lab”.)

Our current BayesForge promotion is going very well, thank you. We seem to be making some inroads among users and among investors who want to help us scale our BF up in size. Many more news bulletins about BF will certainly be forthcoming in this blog in the future. This blog post is not about BF, but it’s about an issue that arose in conversations with some potential BF users. Some persons who might have access to HPC ( i.e., classical supercomputers) inquired whether you can use BF as an interface to try to break the existing record for simulating the largest number of qubits. My answer: yes, you could use BF for that purpose, because BF can include heavy duty C++, highly parallelized qc simulators. However, I think it would be wiser to pursue other, kindred records, that haven’t been pursued too vigorously by the qc community yet and would therefore be much easier to break. The record that I have in mind is using HPC for calculating entanglement for the largest number of qubits. Let me explain.

First, a bit of history about the race to use HPC to simulate the largest number of qubits. It’s a very old race. I first wrote about that race in this blog post dated June, 2010. Back then, the record was 42 qubits. It was held by a German supercomputer called Jugene. 7 years later, I reported in this blog post dated Oct 2017 how Google/ProjectQ captured the record with 45 qubits, only to have it snatched away shortly thereafter by IBM with 49 qubits. According to this paper, and also this one, the Chinese have achieved a similar 49 qubit record on their Sunway TaihuLight supercomputer.

An important consideration in pursuing such a record is that it can be quite expensive to pursue. The people at Google have written a paper in which they calculate that Google would charge you more than a million dollars to make such a simulation on their cloud.

To summarize, here are some reasons why trying to break the record for simulating the largest number of qubits might not be a very wise record to pursue.
1. expensive
2. requires access to a supercomputer, which is an extremely limited resource
3. that record has been broken many times before, so it is now very difficult to improve, and, if you do, the improvement will only be by an unsatisfying epsilon
4. improving the record by an epsilon won’t add much light into the underlying physics.

Now let me talk a little bit about my alternative proposal, to pursue the record for using HPC to calculate entanglement for the largest number of qubits.

I believe that as a quantum system evolves, its entanglement changes, sometimes going through phase transition points. This is very interesting stuff, at least to me. I believe in the future we will want to calculate various quantum entanglement measures for very large systems, so the end-product software that arises from pursuing such a record will be very useful for conducting physics studies. Furthermore, such entanglement calculations cannot be done by a qc, they must be done on classical computers, as far as I now, whereas simulations of 49 qubits will someday be done more efficiently on an actual qc.

I like so much the idea of having better software for calculating the quantum entanglement of arbitrary systems, that I have written a software package that does this. My software package is called “Q Entanglement Lab”. (first version already finished, soon to be released). My software package calculates various entanglement measures for pure and mixed states, either exactly or approximately: exactly for small systems, and approximately, using perturbation theory, for larger systems for which exactness is untenable. I have invented several new algorithms for “Q Entanglement Lab”. If you’ve previously used some of my software, you already know that I often have my own, very idiosyncratic way of doing things. Furthermore, the theory of quantum entanglement measures is very diverse. Therefore, I am sure that if others try writing their own version of “Q Entanglement Lab” before looking at what I have done, they will arrive at very different outcomes than mine. That makes this race very exciting. Let many flowers, with different colors and shapes, bloom.

BayesForge, the movie

Just kidding. What is true is that our “BayesForge, the docker” is now available on AWS (the Amazon cloud). Check out

Bayesforge home page: http://www.bayesforge.com

Bayesforge discourse forum: http://discourse.bayesforge.com

A Meetup talk about Bayesforge: http://www.ar-tiste.com/BayesForgeTalk.pdf

Bayesforge will soon be available on multiple clouds (we are aiming for at least 3 clouds, including at least one in the US and one in China). Versions 1 and 1.1 of BF have been available on AWS for 1.5 years, since April, 2017. But those versions were written in a language specific to the Amazon cloud, whereas the new version 2.0 is a complete rewrite in the Docker language which is very portable across clouds. V2.0 also adds many new features.

The Great Power Shift (from West to East)

The title of this blog post is a pun on the term The Great Vowel Shift. Despite the irreverent title of this blog post, I found the following BBC radio show to be very interesting, and full of serious insights into world events during the last ten years.

The 3rd episode of my @bbcworldservice After the Crash series is broadcast tomorrow with @kuminaidoo @JeffDSachs @NOIweala @lordstern1 @Lagarde @martinwolf_ @mahbubani_k & @adam_tooze Details & podcasts of this & the previous episodes are at: https://t.co/hYFPQJpsx8 & BBC Compass

— Ian Goldin (@ian_goldin) October 9, 2018

Caption:
After the Crash, Episode 3 of 5

Professor Ian Goldin examines how the 2008 financial crisis led to a shift in power from West to East.

As I never tire of reminding people, our quantum and classical AI software company artiste-qb.net is incorporated in Ontario, Canada, but it has a branch in Shenzhen, China, headed by our cofounder Dr. Tao Yin who lives in Shenzhen.

Another Western group of qc people with ties to China is the team behind ProjectQ. ProjectQ is a lower quality competitor to my software Qubiter. The two principal authors of ProjectQ wrote it while working for Matthias Troyer at ETH Zurich, but nowadays Matthias works for the Dark Side (Microsoft) and the two authors of ProjectQ work for Huawei, according to their linkedin accounts:

• Steiger linkedin,
• Haener linkedin

How many AI startups exist in the US? What are the best AI grad schools in the US?

I’ve been told by people who want to start their own Artificial Intelligence startup: “I can do this, this is not rocket science”. My answer to them is, “Luke, that is why you fail”, that is precisely why it will be super difficult for your startup to succeed. Let us do some due diligence to expose the high degree of difficulty of the task:

Check out this Forbes article:

25 Machine Learning Startups To Watch In 2018 (Aug 26, 2018, Forbes, by Louis Columbus)

Excerpt:

• Crunchbase lists over 5,000 startups who are relying on machine learning for their main and ancillary applications, products and services today.
• 81% of machine learning startups Crunchbase tracks have had two funding rounds or less with seed, angel and early-stage rounds being the most common.
• According to KPMG’s Venture Pulse Report, venture capital (VC) investment in artificial intelligence almost doubled in 2017, attracting $12B compared to $6B in 2016.
• Q2’18 was a second-straight record quarter for total Artificial Intelligence (AI) funding with total investments exceeding $2.3B including eight mega-rounds over $100M according to the latest PwC/CB Insights MoneyTree Report from Q2 2018.

So the amount of American VC funding for AI is close to ten billion dollars per year. But the field is HIGHLY CROWDED already, and the competition is FIERCE. Turns out you are not the first one to think: “I can do this, this is not rocket science”. Tens of thousands of people had the same thought long before you did, and they’ve already built a profitable multi-million dollar company. All you have is a vague, possibly non-functioning, pipe dream.

It gets worse.

According to this article in “US and World News Report”, the 5 best American grad schools in AI are:

1. Carnegie Mellon University
2. Massachusetts Institute of Technology
3. Stanford University
4. University of California–Berkeley
5. University of Washington

This is a highly subjective list, but it sort of makes sense. Stanford and Berkeley provide AI worker ants for Silicon Valley, whereas Univ. of Washington does so for Microsoft and Amazon. MIT in the last few decades has been known most of all for its prowess in Genetic Engineering, but that field is closely connected to AI and Big Data. Today MIT announced that it is going to fight Carnegie Mellon for AI supremacy by starting a 1 billion dollar AI school. Check it out:

MIT is investing $1 billion in an AI college(The Verge, Oct 15, 2015, by James Vincent)

So you see chump, it does get worse. That 1 billion dollar school will soon start churning out a continuous stream of AI rocket scientists eager to start their own AI startup that will crush yours.

And China has 4 times the population of the US but it graduates 8 times more people than the US with college degrees in STEM. [Supporting citation for these numbers] And China is investing heavily in AI.

You lose.

Poor Dwave, No longer the only game in town

Quantum Computing Report has recently published a webpage that they call “scorecard on qubit-technology”. Their webpage contains the following spreadsheet, which they will probably update in the future, but as of Oct 8, 2018, it looks like this (it lists 55 qc hardware contenders):

This blog has been around for a long time (since Aug 2008). I remember a time when Scott Aaronson and Geordie Rose were in pitched battle for the hearts and minds of qc fandom. In Nov 2011, I declared Geordie Rose the winner of that contest:

Il Trionfo Di King Kong (The Triumph of King Kong)

If you search this blog under “D-Wave” or “Dwave”, you will see that I’ve had much fun reporting about Dwave through out the years, sometimes positively, sometimes not. But much has happened in qc land since Nov 2011, and one is forced to conclude from the above scorecard that Dwave is no longer the only game in town, now it’s more like one out of 55 games in town. Geordie Rose is not even working at Dwave anymore. He left Dwave in 2014, according to his Linkedin page. So I must declare Scott Aaronson, a.k.a. The Forrelator (Save me mamma!), as the belated winner of that brawl. Of course, Scott hasn’t been super successful since then either, so maybe it’s more of a draw. Scott did win a Cheesy Italian Prize last month (prize must have been a stack of cheese pizzas), but his much publicized Super Boson Sampler failed due to, as was first pointed out by Lubos Motl, a mistake in the assumptions and presumptions of Complexity Theory. Also, Scott’s blog in recent times is as popular as the Maytag repairman’s shop. Nowadays, most of the comments in his blog are either by him, or by right-wing crackpots who enjoy torturing him.

谢谢 (Xièxiè, Thank you), CAS and Tsinghua Univ.

This week was very important for our company artiste-qb-net. Our cofounder Dr. Tao Yin had the honor of giving a short presentation at a workshop entitled: “Quantum Software: from theory to realization” in Beijing. The workshop was organized by the “Institute of Software: Chinese Academy of Sciences (CAS)“. Our most sincere thanks to the organizers of the workshop for allowing us to speak at the last minute.

Here is a conference agenda:
​https://mp.weixin.qq.com/s/3zyzc_ng7A-Re8nbJSAHiA

Here is a photo of Tao giving his presentation:

and here are links to his presentation

Click to access Tao_CAS-sept-2018.pdf

Click to access Tao_CAS_sept_2018_jupyter_nb.pdf

While in Beijing, Tao also had the opportunity to visit and do some networking with qc and Bayesian Networks researchers and programmers at Tsinghua Univ (TU), which is considered one of the best technical universities in China, sort of the MIT of China. TU has a student enrollment 3 times as big (11,000 vs, 36,000) as that of MIT, which reminds me that the population of China is about 4 times that of the USA.

TU is a prominent player in China’s $100 billion push to become a quantum computing giant. In 2015, Turing award winner Andrew Yao moved from Princeton Univ. to TU, to head several Chinese academic institutes, some of which are devoted to quantum information and quantum computing. Recently, TU opened a joint quantum institute at the Univ. of Michigan, and second one at the Univ. of Waterloo and the Waterloo-IQC (Institute for quantum computing)