Monday, January 21, 2013

Physicists At CERN Get Help To Process Experimental Data

A Russian search engine ends up partnering with CERN and not google. Aivars Lode

Yandex, Russia‘s largest online search engine, has collaborated with the European Organization for Nuclear Research, CERN, to help physicists and engineers process experimental data more accurately with their machine learning system, MatrixNet.
MatrixNet is already being tested with CERN’s data of B-meson decay analysis. MatrixNet allows physicists to filter huge datasets in order to find extremely rare events. Increasing the precision of these events, gives physicists the ability to confirm or refute physical models and theories. Yandex is the only Russian company working with CERN.
Forbes caught up with Dr. Andrey Ustyuzhanin, Head of the MatrixNet project with CERN, to find out why a search engine can help some of the smartest physicists in the world. Ustyuzhanin says working with CERN, gives Yandex a chance to help answer some of the most interesting questions about physics and what our universe is made of.
Forbes: Why is it important to help scientists process experimental data more accurately?
Andrey Ustuzhanin: One of primary goals of scientists at CERN is the confirmation of various theories and models that were created by theoretical physicists. If have more precise tools to do that, they can increase their confidence level of their findings using the same amount of data. Given that CERN is going to close for an upgrade soon, it’s essential to have such tools at their disposal.
What does a search engine company have to offer some of the best scientists in the world?
AU: One person can very rarely be ‘the best’ in different areas, even if he/she is a respected as genius. So scientists at CERN are really good at physics, but computer science is totally different area. And, algorithms usually belongs to domain of computer science which is the domain of a search engine company. We hope that data analysis, which is about algorithms and crucial for CERN, can be improved by our technology and techniques.
Why did CERN choose Yandex and say, not Google?
AU: Russia has a very specific cultural background and attitude towards technology. Sometimes when a question is too fascinating, like a question about ‘universe, life and everything’ we can start working on it without any specific budget and revenue expectations, we just want to be a part of that process.
What type of results can the scientists at CERN expect?
AU: If our tools turn out to be helpful, then physicists can expect to speed up discoveries or improve the confidence level of their findings (say, rejection of the next candidatie for ‘theory of everything’).
How about some examples for non-physicists? 
AU: In November 2012, CERN announced a discovery of a rare decay of B-meson to muon-antimuon pair. The confidence level of this decay was ’3.5 sigma’ and rate of this decay was in line with the prediction in the Standard Model of particle physics. This effectively closed some marginal supersymmetry theories. So increasing the confidence level from 3.5 sigma to say five sigma, would make it a discovery, instead of just evidence.
How much experimental data are we talking?
AU: One of CERN big experiments (LHCb) encompasses tens of billion of events per year (which is what they have after pre-filtering), which makes it around a petabyte per year, and that’s not even the biggest CERN experiment. All of CERN’s experiments are reported to register around 15PB per year.
Note: Let’s put that into perspective:  The data flow from all of the experiments will be about 700 MBs, that’s around 15,000 000 GB (=15 PB) per year and that’s about the same as a stack of CDs about 20 km tall piled up each year.
What’s the long-term benefit to Yandex? 
AU: Currently we are thinking in terms of helping find answers to some of the most difficult questions about understanding our universe. Besides working on these questions alongside the best physicists in the world, it will help us improve our methodology of machine learning.
What do you want to achieve with the MatrixNet at CERN?
AU: Our mission is to give answers to users’ questions, so helping answering tough physics questions is just another part of our mission at Yandex.
Yandex says its machine learning system, MatrixNet, will make search smarter because they will be able to train/teach the system based on the massive amount of data collected by CERN and of course apply that to it’s business – internet search.

No comments:

Post a Comment