Research Topic

Machine Learning for Interpretable, Efficient Reconstruction of Physics in Big Data

About this Research Topic

Recent advancements in the field of machine learning and artificial intelligence (ML/AI) have found a variety of applications including high-level analysis of big data in science. While many of the early applications in experimental physics are found in high-level analysis tasks toward the end of a pipeline, recently they are expanded into data reconstruction tasks in order to efficiently and maximally extract physics information from the low-level data, ultimately the raw data recorded by a particle detector. Data reconstruction is causal, hierarchical, and heterogeneous, where a wider variety of ML/AI techniques will thrive while it also serves as a perfect ground to develop science-specific development of ML/AI methods. Beyond the development of algorithms for individual reconstruction tasks, a study of efficient, end-to-end optimizable chain of reconstruction algorithms and software tools to interface distributed computing system may dramatically change the turn-around time for experiments’ data production. The development of ML/AI methods in all aspects of data reconstruction is, therefore, a topic of critical interest for experimental particle physics.

Reconstruction assumes an underlying physics model, and thus ML/AI models developed for data reconstruction are expected to incorporate physics knowledge in their design and architecture. This includes conservation laws, causality such as particle decays, symmetry, geometrical and topological properties, constraints from detector response, and more. Furthermore, data reconstruction as a tool to support generic physics analyses means it has to provide a comprehensive list of physics observables in the output. This heterogeneity motivates a study of a wide range of algorithms to extract a whole spectrum of physics observables directly, as well as a research directed toward an efficient optimization of chained algorithms, such as multi-task cascade models, using big data and computing power. Finally, for the outputs to be fully interpretable for physics analyses, reconstruction algorithms need to be compatible with methods of uncertainty estimation. Two particularly important features include propagation of input systematic uncertainties (e.g. an uncertainty on detector response parameter) and also a model-intrinsic uncertainty on reconstructed parameters. Furthermore, optimization methods specialized towards making the reconstruction process robust against input systematic uncertainties and mis-modeling of underlying physics processes (i.e. domain adaptation) are also relevant. In summary, three key challenges include incorporation of physics knowledge in ML/AI models, a study of multi-task cascade models and optimization techniques, and quantification of uncertainties from both internal and external origins are the core achievables in fully integrating ML/AI models in physics data reconstruction pipeline.

A wide range of research work related to ML/AI models applied for data reconstruction will be welcomed and highly valued. This includes the development of a specific reconstruction task, optimization of end-to-end reconstruction chains, physics-informed (i.e. causality, inductive bias, symmetry, conservation laws, etc.) models, integration of uncertainty quantification into reconstruction algorithms, optimization of models against external systematic uncertainties, implementation of models in a distributed computing environment, and more.

Topics that can be included are:
• Interpretable data reconstruction methods to support high-level physics analysis
• Optimization of an end-to-end (chain of) data reconstruction algorithms
• Quantification of systematic uncertainty through and from the reconstruction models
• Domain adaptation and systematics-aware optimization methods to make reconstruction algorithms more robust.


Keywords: Machine Learning, Interpetable, Efficient Reconstruction, Physics, Big Data


Important Note: All contributions to this Research Topic must be within the scope of the section and journal to which they are submitted, as defined in their mission statements. Frontiers reserves the right to guide an out-of-scope manuscript to a more suitable section or journal at any stage of peer review.

Recent advancements in the field of machine learning and artificial intelligence (ML/AI) have found a variety of applications including high-level analysis of big data in science. While many of the early applications in experimental physics are found in high-level analysis tasks toward the end of a pipeline, recently they are expanded into data reconstruction tasks in order to efficiently and maximally extract physics information from the low-level data, ultimately the raw data recorded by a particle detector. Data reconstruction is causal, hierarchical, and heterogeneous, where a wider variety of ML/AI techniques will thrive while it also serves as a perfect ground to develop science-specific development of ML/AI methods. Beyond the development of algorithms for individual reconstruction tasks, a study of efficient, end-to-end optimizable chain of reconstruction algorithms and software tools to interface distributed computing system may dramatically change the turn-around time for experiments’ data production. The development of ML/AI methods in all aspects of data reconstruction is, therefore, a topic of critical interest for experimental particle physics.

Reconstruction assumes an underlying physics model, and thus ML/AI models developed for data reconstruction are expected to incorporate physics knowledge in their design and architecture. This includes conservation laws, causality such as particle decays, symmetry, geometrical and topological properties, constraints from detector response, and more. Furthermore, data reconstruction as a tool to support generic physics analyses means it has to provide a comprehensive list of physics observables in the output. This heterogeneity motivates a study of a wide range of algorithms to extract a whole spectrum of physics observables directly, as well as a research directed toward an efficient optimization of chained algorithms, such as multi-task cascade models, using big data and computing power. Finally, for the outputs to be fully interpretable for physics analyses, reconstruction algorithms need to be compatible with methods of uncertainty estimation. Two particularly important features include propagation of input systematic uncertainties (e.g. an uncertainty on detector response parameter) and also a model-intrinsic uncertainty on reconstructed parameters. Furthermore, optimization methods specialized towards making the reconstruction process robust against input systematic uncertainties and mis-modeling of underlying physics processes (i.e. domain adaptation) are also relevant. In summary, three key challenges include incorporation of physics knowledge in ML/AI models, a study of multi-task cascade models and optimization techniques, and quantification of uncertainties from both internal and external origins are the core achievables in fully integrating ML/AI models in physics data reconstruction pipeline.

A wide range of research work related to ML/AI models applied for data reconstruction will be welcomed and highly valued. This includes the development of a specific reconstruction task, optimization of end-to-end reconstruction chains, physics-informed (i.e. causality, inductive bias, symmetry, conservation laws, etc.) models, integration of uncertainty quantification into reconstruction algorithms, optimization of models against external systematic uncertainties, implementation of models in a distributed computing environment, and more.

Topics that can be included are:
• Interpretable data reconstruction methods to support high-level physics analysis
• Optimization of an end-to-end (chain of) data reconstruction algorithms
• Quantification of systematic uncertainty through and from the reconstruction models
• Domain adaptation and systematics-aware optimization methods to make reconstruction algorithms more robust.


Keywords: Machine Learning, Interpetable, Efficient Reconstruction, Physics, Big Data


Important Note: All contributions to this Research Topic must be within the scope of the section and journal to which they are submitted, as defined in their mission statements. Frontiers reserves the right to guide an out-of-scope manuscript to a more suitable section or journal at any stage of peer review.

About Frontiers Research Topics

With their unique mixes of varied contributions from Original Research to Review Articles, Research Topics unify the most influential researchers, the latest key findings and historical advances in a hot research area! Find out more on how to host your own Frontiers Research Topic or contribute to one as an author.

Topic Editors

Loading..

Submission Deadlines

15 June 2021 Abstract
15 September 2021 Manuscript

Participating Journals

Manuscripts can be submitted to this Research Topic via the following journals:

Loading..

Topic Editors

Loading..

Submission Deadlines

15 June 2021 Abstract
15 September 2021 Manuscript

Participating Journals

Manuscripts can be submitted to this Research Topic via the following journals:

Loading..
Loading..

total views article views article downloads topic views

}
 
Top countries
Top referring sites
Loading..