Jet-Images: Computer Vision Inspired Techniques for Jet Tagging (1407.5675v3)

Abstract: We introduce a novel approach to jet tagging and classification through the use of techniques inspired by computer vision. Drawing parallels to the problem of facial recognition in images, we define a jet-image using calorimeter towers as the elements of the image and establish jet-image preprocessing methods. For the jet-image processing step, we develop a discriminant for classifying the jet-images derived using Fisher discriminant analysis. The effectiveness of the technique is shown within the context of identifying boosted hadronic W boson decays with respect to a background of quark- and gluon- initiated jets. Using Monte Carlo simulation, we demonstrate that the performance of this technique introduces additional discriminating power over other substructure approaches, and gives significant insight into the internal structure of jets.

• The paper presents a novel computer vision approach by treating jets as images to enhance the classification of hadronic decays.
• It introduces a systematic preprocessing pipeline—including noise reduction, alignment, and equalization—to standardize jet-images for feature extraction.
• Using Fisher’s Linear Discriminant, the method achieves robust discrimination of jet types, outperforming traditional techniques like N-subjettiness.

An Overview of Jet-Images: Computer Vision Inspired Techniques for Jet Tagging

The paper "Jet-Images: Computer Vision Inspired Techniques for Jet Tagging" introduces a novel methodology for classifying jet types, notably between hadronically decaying boosted $W$ bosons and QCD jets. This approach leverages techniques from the domain of computer vision, treating jets as images to facilitate a more intuitive and potentially potent strategy for jet tagging in the context of TeV-scale colliders such as the LHC.

Methodology and Techniques

The researchers propose representing jets as "jet-images," where calorimeter towers are akin to pixels in an image. This laid the foundation for employing computer vision techniques traditionally used for tasks like facial recognition. The authors detail a preprocessing pipeline specifically designed for jet-images:

1. Noise Reduction: Applied through trimming to manage pileup effects.
2. Point of Interest Finding: By identifying the leading subjets.
3. Alignment: Aligning images through rotation, translation, and reflection to account for variations like decay angles.
4. Equalization: Standardizing the energy scale across jet-images.
5. Binning: Categorizing jet-images based on parameters such as transverse momentum ($p_T$) and the angular separation of subjets ($\Delta R_{jj}$).

The preprocessing ensures that the jet-images are in a consistent format, allowing for effective application of feature extraction techniques similar to those used in identifying features in facial images.

Discrimination Approach

The authors apply Fisher's Linear Discriminant (FLD) analysis, emphasizing its speed and interpretability compared to complex non-linear methods, to determine a discriminant referred to as the Fisher-jet. This method projects jet-images into a high-dimensional space where class separation is optimized, allowing for effective discrimination between jet types. Complementing the preprocessing, FLD provides both a systematic categorization method and insights into the attributes distinguishing different jet classes.

Performance Analysis

The technique's efficacy was assessed using Monte Carlo simulations focused on distinguishing boosted hadronic W boson decays from QCD jets. Results show that this method offers robust discrimination power, outperforming traditional approaches such as N-subjettiness across the tested momentum ranges. The methodology maintains its performance even under varying pileup conditions or when evaluated on samples from different event generators, affirming its applicability in diverse experimental contexts.

Implications and Future Directions

This paper indicates significant promise in transferring methodologies from computer vision to high-energy physics, opening avenues for more nuanced jet classification schemes. By framing jets within an image-processing paradigm, the approach can potentially be expanded to other physics problems characterized by spatial energy distributions. The insights into jet substructure gained through visual analysis of Fisher-jets could inform the development of more refined analytical models, enhancing our understanding of particle interactions in high-energy environments.

In future developments, integrating non-linear classifiers or employing deeper learning methods on these consistent jet-images may enhance the discrimination effectiveness further. The paper underscores a paradigm shift in jet tagging, fostering cross-disciplinary synergies that could yield richer exploration and identification capabilities for high-energy particle physics.