The interest in possible applications of machine learning in physics has been growing exponentially for a while now and there seems to be a sea of literature. Couple of months ago we decided to review the literature and have weekly seminar about the papers we found interesting. There is a dedicated blog post for each of these on our group website.
Recently I have contributed with the discussion of the paper by E. M. Stoudenmire and D. J. Schwab: Supervised Learning with Quantum-Inspired Tensor Networks (arXiv:1605.05775 (2017)). In this paper the authors propose to train a tensor network with DMRG-like sweeps. You can read my full post here.
I can’t even begin to describe the impact that reading The Power by Naomi Alderman had on me (and I am sure I am not alone). I would even go as far as to predict that this book will become a classic that will keep on being discussed for many decades.
The main premise is the following: at some point teenage girls start experiencing the ability to strike someone with the electricity thanks to the new organ they developed. This new development than leads to complete re-definition of the gender norms on unprecedented scale.
The point that the book is making in the incredible clever and spot-on manner is that the inequality is about power and it has absolutely nothing to do with race or gender by default. I read lots of angry counter arguments to the book stating that if power dynamics were switched women would never ever be so violent or power hungry. But I believe that truth of the matter is when there is an opportunity, means and societal approval people are ready shift their morals surprisingly quickly.
In my opinion, the problem we have nowadays is not there is a better race or gender for politics, computer science or battle against climate change, it is the disproportional representation of people with the certain kind of opinion in the position of power. And what Ms Alderman illustrates wonderfully by the means of extremely readable fiction is that true equality is not that women and minorities can in principle do whatever they want, true equality is the proportional representation of all humans in all the relevant position of power.
For me 2017 was all about reading bit out of my comfort zone and expanding my genre horizons. Even though it was fun and I found some delightful books, I am looking forward to go back to my line-up of contemporary fiction in 2018.
The most remarkable fiction I read last year was definitely Paul Beatty’s The Sellout (the book that takes some of the most twisted racial prejudices to extreme consequences in a rather satirical manner – even though the author himself doesn’t like the book to be labelled as a satire). Apart from the fact that The Sellout will make you laugh and cry at the same time, this book is one of the best ways of making a point I have ever seen in any context. Another fiction book I liked, as disturbing as it is, is Han Kang’s Vegetarian (I wrote bit about it here).
I tried to get into sci-fi this year, and some classics like Dune took my breath away, some others, like 2001: A Space Odyssey, completely annoyed me by its blatant sexism (sure in 50 years the travel across the galaxy will be possible, but having a woman engineer would be absurd). The sci-fi book published in 2017 that I absolutely loved is John Scalzi’s Collapsing Empire, which is very political and seriously funny space opera with a cast of diverse super interesting characters. I am already looking forward to the follow-up to be published in 2019.
I also read number of biographies (I really wanted to mention Al Franken here, but well… I guess that the book itself is still interesting enough if you feel like reading it). I loved Trevor Noah and Tiffany Haddish books that are both super funny and inspiring.
The annoyed-me-the-most in 2017 is split between Yuval Noah Harari: Homo Deus and Brian Green’s Light Falls. Yuval Harari wrote a book about a species that Homo Sapiens will develop into due to the artificial intelligence revolution. While the book is making some good points and somehow landed on the reputable recommended reading lists, I was constantly annoyed by the over-simplification and superficiality of many of the arguments. To give Mr. Harari the benefit of the doubt, I think that often he plays the devil’s advocate on purpose (he likes to make a point that the problem with Capital not becoming reality is that the capitalists can read). So I guess that his argument that liberal values only make sense in the world where each member of the society is needed to contribute to GDP might be meant as a provocation of the similar type. Then again, it was a week of rage at the over-generalization that is a direct consequence of selectively picking certain claims about artifical intelligence and creating a false narrative.
When comes to Brian Green’s book, let me just say is it really okay in 2000something write stuff like: Einstein was stressed because Hilbert was closing in on him and he had to deal with angry outbursts of his wife while facing the pressure to marry his mistress. Really?
Let me finish with honorable mentions of other great books that I warmly recommend: Hope Jahren: Lab Girl (thoughtful and beautiful memoir of a biologist), Neil deGrasse Tyson: Astrophysics for People in a Hurry (super cool popular science book), Andy Weir: The Martian (perfect nerdy sci-fi).
In our latest work, now on arXiv, we show how to use a convolutional neural network to extract physical parameters (even the quantum ones!) from experimental currents.
In my PhD I was generally concerned with monitoring and parameter estimation of quantum systems. These elements are crucial for efficiently functioning quantum devices, and, in difference from on-chip quantum operations, there is still a long way to go in terms of getting efficient readout at reasonable times. The ability to extract the maximum amount of information from an experimental record is therefore essential.
In practice, the experimental noise is sometimes so stubborn and viciously correlated that it may be really hard if not impossible to construct a quantum model that describes it. In our work we show that even for the cases where traditional parameter estimation methods do not work the convolutional network is a great solution to find the parameters governing the dynamics of the system.
Lately I have been working a lot with Google’s TensorFlow library for machine learning. It has a really nice tool for data visualisation, TensorBoard, which can be very useful to understand how the training and evaluation of your model is working. One small bottleneck though is that it has a built-in tool for data export that only works for the scalar functions and unfortunately not for more complex visualisation means like histograms. I find especially the histograms to be particularly useful because they show how is your probability distribution narrowing as a function of learning steps, so it is a really useful figure of merit for understanding the training/evaluation. This is also the reason why I thought it would be useful to export the histograms and customise them for example in Matlab. I would like to share here the code for exporting the histograms from the TensorFlow model. Hopefully you will find it useful! You can download it here.
Saas-Fee is a wonderful car-free Swiss village with the proximity to a glacier and several 4000 peaks. I just took a short holiday there and here is a few pictures. If you are ever going to hike there I can especially recommend going to Spielboden and Langflüh, from where you have excellent view of the glacier, and walking from Felskinn to Plattjen, which starts way above the snow line and is one of the most beautiful hikes I have ever went to.
Recently I finished my latest work that has been done in collaboration with my wonderful supervisor and Oxford experimental team and I would like to use this post to advertise it a bit in general terms. You can read it in full at arXiv.
The past quantum state method relies on a simple assumption: since in practical experimental situations you would like to monitor your system continuously and collect as much data as possible it makes sense to condition your probability not only what happened to your system BEFORE the time t (that is any given time for which you would like to make you probability prediction), but also AFTER the time t. In other words you use both the PAST and the FUTURE (from the point of the time, t, you are interested in) to make a probability prediction. This might sound a little bit sci-fi but as in general in quantum reality it is nothing too fancy, you basically just need to modify the Born rule a bit. The method was first proposed here and we used this kind of reasoning to argue stuff about correlation functions and improve fidelity of the teleportation protocol.
Here we took on the challenge to improve the experimental readout of the single electron quantum dot as well as modify existing techniques for parameter estimation. As it turns out, for typical experimental parameters, we are able to remove most of the noise and we are able to find time of each tunnelling event with super high precision. In addition to that we modified the Baum-Welch parameter estimation method and combined it with good old Bayesian to estimate both coherent and incoherent parameters under the same footing. So if you like quantum dots or you are just interested in quantum measurement theory in general, please have a look!