4 ms·

Machine learning proliferates in particle physics

44 points by qubitcoder 8y ago

buboard 8y ago
"proliferates" ? skepticism? Afaik, people in CERN have been using neural nets to identify trajectories at least since the 80s.
- xvilka 8y ago
  Exactly! They basically invented modern big data/data science. ROOT[1] is one of the oldest statistical frameworks after all. And they have machine learning in it too[2][3]. [1] https://root.cern.ch/ https://root.cern.ch/ [2] https://root.cern.ch/tmva https://root.cern.ch/tmva [3] http://iopscience.iop.org/article/10.1088/1742-6596/898/7/072046/pdf http://iopscience.iop.org/article/10.1088/1742-6596/898/7/07...
  - xvilka 8y ago
    Note also, they are working heavily on ROOT 7[1], which will do some refactor, not sure probably even better ML/DL support. Not sure why did they target c++14 instead of c++17 though. [1] https://root.cern.ch/root-7 https://root.cern.ch/root-7
thoughtstheseus 8y ago
Check out the podcast Linear Digressions, one of the hosts Katie went from physics to data science/ML work.
abcininin 8y ago
I wish there were concrete examples of how machine learning is being used.
- davesque 8y ago
  Pretty broad but, from one section in the article, sounds like they're using convolutional layers in their networks?
tpaschalis 8y ago
The comments of buboard and xvilka are on point. One thing I'd add though, is that physicists still have a healthy amount of skepticism on ML/NN. They don't like the "black box" approach, where the algorithm is regarded to "provide" the truth, but we're required to fully understand the math behind any implementation, as well as to be able to answer deeper questions on the results we're acquiring. Background : I'm completing my master in Particle Physics, and have had the opportunity to work along these kinds of people. They have wrestled their entire careers with enormous amounts of data, create new solutions to problems the industry had not even formulated, and keep themselves on the forefront of Statistics and ML. I feel that the things I've learned by trotting along grey-haired old-school Unix wizards could not be replicated by any courses, bootcamps or manuals.

44 points by qubitcoder 8y ago

"proliferates" ? skepticism? Afaik, people in CERN have been using neural nets to identify trajectories at least since the 80s.

xvilka 8y ago
Exactly! They basically invented modern big data/data science. ROOT[1] is one of the oldest statistical frameworks after all. And they have machine learning in it too[2][3]. [1] https://root.cern.ch/ https://root.cern.ch/ [2] https://root.cern.ch/tmva https://root.cern.ch/tmva [3] http://iopscience.iop.org/article/10.1088/1742-6596/898/7/072046/pdf http://iopscience.iop.org/article/10.1088/1742-6596/898/7/07...
- xvilka 8y ago
  Note also, they are working heavily on ROOT 7[1], which will do some refactor, not sure probably even better ML/DL support. Not sure why did they target c++14 instead of c++17 though. [1] https://root.cern.ch/root-7 https://root.cern.ch/root-7

thoughtstheseus 8y ago

Check out the podcast Linear Digressions, one of the hosts Katie went from physics to data science/ML work.

abcininin 8y ago

I wish there were concrete examples of how machine learning is being used.

davesque 8y ago
Pretty broad but, from one section in the article, sounds like they're using convolutional layers in their networks?

tpaschalis 8y ago

The comments of buboard and xvilka are on point. One thing I'd add though, is that physicists still have a healthy amount of skepticism on ML/NN. They don't like the "black box" approach, where the algorithm is regarded to "provide" the truth, but we're required to fully understand the math behind any implementation, as well as to be able to answer deeper questions on the results we're acquiring. Background : I'm completing my master in Particle Physics, and have had the opportunity to work along these kinds of people. They have wrestled their entire careers with enormous amounts of data, create new solutions to problems the industry had not even formulated, and keep themselves on the forefront of Statistics and ML. I feel that the things I've learned by trotting along grey-haired old-school Unix wizards could not be replicated by any courses, bootcamps or manuals.