Trigger-Level Multiple Electron Event Classification with LDMX using Artificial Neural Networks
(2023) FYSM33 20231Particle and nuclear physics
Department of Physics
- Abstract
- Artificial neural networks is a powerful tool for classifying and identifying patterns in large amounts of data. One of the possible tasks of these networks is classification of data into categories. LDMX is a fixed target experiment that is designed to search for light dark matter particles using missing energy and momentum of electrons passing through and scattering of a target. In the current design the number of electrons passing through is counted using the trigger scintillator system.
In this thesis, four different types of artificial neural networks were trained to count the number of electrons present in events using simulated data from both the electromagnetic calorimeter and trigger scinitllator to explore the viability of... (More) - Artificial neural networks is a powerful tool for classifying and identifying patterns in large amounts of data. One of the possible tasks of these networks is classification of data into categories. LDMX is a fixed target experiment that is designed to search for light dark matter particles using missing energy and momentum of electrons passing through and scattering of a target. In the current design the number of electrons passing through is counted using the trigger scintillator system.
In this thesis, four different types of artificial neural networks were trained to count the number of electrons present in events using simulated data from both the electromagnetic calorimeter and trigger scinitllator to explore the viability of using neural networks at LDMX. These were convolutional neural networks, recurrent neural networks, graph neural networks, and a combined convolutional and recurrent neural network. Three datasets were generated: one using only data from the electromagnetic calorimeter, one that combined the electromagnetic calorimeter and the trigger scinitllator, and one that combined the electromagnetic calorimeter and a modified design for the trigger scinitllator data. To be viable the accuracy of the models needed to be equal to or exceed 0.99999. None of the models were able to exceed or match this accuracy with the highest accuracy reached by a convolutional and recurrent model training with the electromagnetic calorimeter and a modified design for the trigger scinitllator data of 0.962 and 0.9575. The graph neural network was not able to be trained with data in time and neither was all of the combined convolutional and recurrent neural networks. The different benefits and drawbacks of the models were then evaluated and compared to each other. (Less) - Popular Abstract
- Dark Matter remains an elusive adversary for physicists. Few topics in physics have inspired more theories and debate while simultaneously remaining inscrutable, dodging our collective stabs at an effective explanation. Due to Dark Matter not interacting with light or other forces such as the strong nuclear force that binds protons and neutrons together, the only real way of confirming its effect on our universe is through its gravitational effects. The Standard Model of physics that describes all the matter made up of atoms and other particles only describes 5 \% of our universe with 25 $\%$ of all the matter and energy made out of Dark Matter and the rest made out of some "Dark Energy". This means that a significant portion of our... (More)
- Dark Matter remains an elusive adversary for physicists. Few topics in physics have inspired more theories and debate while simultaneously remaining inscrutable, dodging our collective stabs at an effective explanation. Due to Dark Matter not interacting with light or other forces such as the strong nuclear force that binds protons and neutrons together, the only real way of confirming its effect on our universe is through its gravitational effects. The Standard Model of physics that describes all the matter made up of atoms and other particles only describes 5 \% of our universe with 25 $\%$ of all the matter and energy made out of Dark Matter and the rest made out of some "Dark Energy". This means that a significant portion of our universe is only accessible indirectly meaning the true nature of our universe is obscured to us.
Many theories that go beyond the Standard Model of physics predict the existence of a type of particles called Weakly Interacting Massive Particles, or WIMPs, that have very small interactions with regular matter, if any, where weak refers to the weak nuclear force responsible for nuclear decay. These particles can have masses ranging from 100 TeV to below 1 GeV corresponding to roughly 100000 times the mass of a proton to below the proton mass, respectively. Numerous experiments exist to try to discover these particles. One type of theory that can involve the existence of WIMPs or possibly other types of particles with other interaction types is the Dark Sector. In this hypothesis there can exist entirely different types of forces and particles that cannot interact with us except through some mediator particle or particles. One version of the Dark Sector hypothesis treats one of these sectors analogously to electromagnetism and can interact with regular charged matter through a "dark photon" mediator particle that can then decay to Dark Matter. This would then allow for the existence of Dark Matter particles below 1 GeV in mass a region referred to as "Light Dark Matter". It is in this region that the Light Dark Matter eXperiment is designed to search for and possibly (finally) find Dark Matter.
LDMX is a proposed fixed-target missing momentum experiment. The experiment involves measuring if Dark Matter particles have been produced in the wake of an electron beam being fired at a tungsten target. Dark Matter can then be generated when the electron interacts with the target and emits a dark photon. When measuring the total amount of energy and momentum of the electron there will then be a certain amount of missing energy and momentum that can then be used to determine the properties of the mediator particle and the resultant Dark Matter. Dealing with the large amounts of events in the detector requires the use of \emph{triggers}. A trigger is a real-time decision that determines whether an observed event should be saved to be analysed for a possible "signal" event involving the emission of a dark photon later. With a beam rate or incoming event rate of 37.2 MHZ meaning approximately 37 million events per second only 5 KHZ or 5 thousand events will be saved to avoid filling the experiment with data that is not relevant or unlikely to contain a signal event. For single electron events where only one electron is involved triggers have been designed that can veto, or remove, non-relevant events. However, for multiple electron events these vetoes are harder to design and often require more processing than for a single electron event and the possibility of accidentally vetoing a multiple electron signal event remains. Being able to distinguish between different types of events as involving one versus two or more electrons is then something that is of vital importance for ensuring that the proper analysis can be performed and no Dark Matter signal indicators are missed.
Machine Learning has steadily become a reliable tool for doing analysis on large sets of data such as for the identification of objects in images or finding relationships between different data points. With products such as ChatGPT demonstrating the power of large language models, Machine Learning continues to be an active area of research and development in all aspects of society. This thesis aims to see how Machine Learning in the form of Artificial Neural Networks can be trained on the data from the Electromagnetic Calorimeter and the Trigger Scintillator that register the energy of charged particles moving through it and positional data on the beam of incoming electrons. Artificial Neural Networks is a type of Machine Learning structure that relies on a set of nodes arranged in layers that are connected. These connections are controlled with a set of weights and these are continuously updated while training. In this case it would be trained using simulated data on events with up to four electrons in order to determine how an Artificial Neural Network can be used when designing vetos. As numerous types of architecture for these networks exist, this thesis restricted itself to four types: Convolutional Neural Networks, Recurrent Neural Networks, Graph Neural Networks, and a Convolutional and Recurrent Neural Network. Combining Machine Learning with the LDMX could then help end our quest for the true nature of Dark Matter and finally lay our nemesis to rest. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9131950
@misc{9131950,
abstract = {{Artificial neural networks is a powerful tool for classifying and identifying patterns in large amounts of data. One of the possible tasks of these networks is classification of data into categories. LDMX is a fixed target experiment that is designed to search for light dark matter particles using missing energy and momentum of electrons passing through and scattering of a target. In the current design the number of electrons passing through is counted using the trigger scintillator system.
In this thesis, four different types of artificial neural networks were trained to count the number of electrons present in events using simulated data from both the electromagnetic calorimeter and trigger scinitllator to explore the viability of using neural networks at LDMX. These were convolutional neural networks, recurrent neural networks, graph neural networks, and a combined convolutional and recurrent neural network. Three datasets were generated: one using only data from the electromagnetic calorimeter, one that combined the electromagnetic calorimeter and the trigger scinitllator, and one that combined the electromagnetic calorimeter and a modified design for the trigger scinitllator data. To be viable the accuracy of the models needed to be equal to or exceed 0.99999. None of the models were able to exceed or match this accuracy with the highest accuracy reached by a convolutional and recurrent model training with the electromagnetic calorimeter and a modified design for the trigger scinitllator data of 0.962 and 0.9575. The graph neural network was not able to be trained with data in time and neither was all of the combined convolutional and recurrent neural networks. The different benefits and drawbacks of the models were then evaluated and compared to each other.}},
author = {{Lindahl, Jacob}},
language = {{eng}},
note = {{Student Paper}},
title = {{Trigger-Level Multiple Electron Event Classification with LDMX using Artificial Neural Networks}},
year = {{2023}},
}