Differentiable Vertex Fitting for Jet Flavour Tagging
Authors:
Rachel E. C. Smith,
Inês Ochoa,
Rúben Inácio,
Jonathan Shoemaker,
Michael Kagan
Abstract:
We propose a differentiable vertex fitting algorithm that can be used for secondary vertex fitting, and that can be seamlessly integrated into neural networks for jet flavour tagging. Vertex fitting is formulated as an optimization problem where gradients of the optimized solution vertex are defined through implicit differentiation and can be passed to upstream or downstream neural network compone…
▽ More
We propose a differentiable vertex fitting algorithm that can be used for secondary vertex fitting, and that can be seamlessly integrated into neural networks for jet flavour tagging. Vertex fitting is formulated as an optimization problem where gradients of the optimized solution vertex are defined through implicit differentiation and can be passed to upstream or downstream neural network components for network training. More broadly, this is an application of differentiable programming to integrate physics knowledge into neural network models in high energy physics. We demonstrate how differentiable secondary vertex fitting can be integrated into larger transformer-based models for flavour tagging and improve heavy flavour jet classification.
△ Less
Submitted 19 October, 2023;
originally announced October 2023.
Grazing Incidence Optics for Wide-field X-ray Survey Imaging: A Comparison of Optimization Techniques
Authors:
Peter W. A. Roming,
John C. Liechty,
Jared R. Shoemaker,
David H. Sohn,
William B. Roush,
David N. Burrows,
Gordon P. Garmire
Abstract:
Utilizing a ray-tracing program, we have modeled the angular resolution of a short focal length (~2m), large field-of-view (3.1 square degrees), grazing incidence mirror shell. It has been previously shown in the literature that the application of a polynomial to the surface of grazing incidence mirror shells enhances the global performance of the mirror over the entire field-of-view. The object…
▽ More
Utilizing a ray-tracing program, we have modeled the angular resolution of a short focal length (~2m), large field-of-view (3.1 square degrees), grazing incidence mirror shell. It has been previously shown in the literature that the application of a polynomial to the surface of grazing incidence mirror shells enhances the global performance of the mirror over the entire field-of-view. The objective of this project was to efficiently locate the optimal polynomial coefficients that would provide a 15 arcsec response over the entire field-of-view. We have investigated various techniques for identifying the optimal coefficients in a large multi-dimensional polynomial space. The techniques investigated include the downhill simplex method, fractional factorial, response surface (including Box-Behnken and central composite) designs, artificial neural networks (such as back-propagation, general regression, and group method of data handling neural networks), and the Metropolis-Coupled Markov-Chain Monte-Carlo (MC-MCMC) method. We find of the methods examined, the MC-MCMC approach performs the best. This project demonstrates that the MC-MCMC technique is a powerful tool for designing irreducible algorithms that optimize arbitrary, bounded functions and that it is an efficient way of probing a multi-dimensional space and uncovering the global minimum in a function that may have multiple minimums.
△ Less
Submitted 3 June, 2004;
originally announced June 2004.
