Playing Tag with ANN: Boosted Top Identification with Pattern Recognition (1501.05968v1)

Abstract: Many searches for physics beyond the Standard Model at the Large Hadron Collider (LHC) rely on top tagging algorithms, which discriminate between boosted hadronic top quarks and the much more common jets initiated by light quarks and gluons. We note that the hadronic calorimeter (HCAL) effectively takes a "digital image" of each jet, with pixel intensities given by energy deposits in individual HCAL cells. Viewed in this way, top tagging becomes a canonical pattern recognition problem. With this motivation, we present a novel top tagging algorithm based on an Artificial Neural Network (ANN), one of the most popular approaches to pattern recognition. The ANN is trained on a large sample of boosted tops and light quark/gluon jets, and is then applied to independent test samples. The ANN tagger demonstrated excellent performance in a Monte Carlo study: for example, for jets with p_T in the 1100-1200 GeV range, 60% top-tag efficiency can be achieved with a 4% mis-tag rate. We discuss the physical features of the jets identified by the ANN tagger as the most important for classification, as well as correlations between the ANN tagger and some of the familiar top-tagging observables and algorithms.

• The paper proposes a novel approach using Artificial Neural Networks (ANNs) to identify boosted top quarks by treating hadronic calorimeter data as digital images and applying pattern recognition techniques.
• Applying the ANN to testing samples achieved a 60% top-tagging efficiency with a 4% mis-tag rate for 1100-1200 GeV jets, outperforming some existing top-tagging strategies.
• This ANN method can enhance LHC experiments' ability to identify new particles and suggests promising future directions for integrating machine learning in particle physics analysis.

Playing Tag with ANN: Boosted Top Identification with Pattern Recognition

The paper, "Playing Tag with ANN: Boosted Top Identification with Pattern Recognition," introduces a novel algorithm utilizing Artificial Neural Networks (ANNs) for discerning boosted top quarks amidst light-quark and gluon-initiated jets at the Large Hadron Collider (LHC). The paper proposes a distinct methodology, treating the hadronic calorimeter (HCAL) data as "digital images," and consequently reframing top tagging as a pattern recognition challenge—a domain where ANNs have demonstrated significant efficacy.

Summary of Methodology

The authors leverage a supervised learning approach, training the ANN on a substantial dataset generated via Monte Carlo simulations. These simulations comprise both signal (boosted top quarks) and background (light quark/gluon jets) events. HCAL provides energy depositions within its cells, envisioned as pixel intensities in an image, forming the input for the ANN. This innovative perspective posits boosted top quark identification as a conventional image-recognition task, harnessing the powerful pattern recognition capabilities inherent in ANNs.

Performance and Results

Applying the ANN to testing samples, the paper reports promising performance metrics. Notably, for jets within the 1100-1200 GeV transverse momentum ($p_T$) range, the ANN achieves a 60% top-tagging efficiency, closely linked to a manageable 4% mis-tag rate. These results distinctly surpass several contemporary top-tagging strategies, underscoring the algorithm's robustness and potential utility in particle physics analyses at elevated energy scales.

The ANN's advantage lies in its capacity to capture and generalize jet substructure patterns. It identifies the spatial correlation of energy deposits—a feat demonstrated as beneficial when compared to other top-tagging criteria. The observed correlations between the ANN's classification results and established tagging observables suggest that the ANN extracts features complementary to existing methods, aligning but also extending beyond them due to its non-linear, data-driven nature.

Implications and Future Directions

The implications of this research are twofold. Practically, the adoption of ANN-based tagged could enhance the capability of LHC experiments to identify new particles predicted by extensions of the Standard Model, particularly those involving electroweak symmetry breaking and naturalness. Theoretically, this work strengthens the argument for cross-disciplinary approaches in high-energy physics, utilizing machine learning to solve complex classification tasks.

Considering future developments, enhancing the ANN architecture by integrating more sophisticated neural networks, such as convolutional networks, or augmenting the input with additional data, like sub-jet $b$-tags, could further refine performance. The research also hints at broader applications, such as identifying other exotic particles, including boosted $W$ and Higgs bosons.

Conclusion

This paper robustly establishes the ANN-based method as a viable and effective tool for boosted top quark identification in high-energy experiments. By framing the problem as one of image recognition, the authors not only showcase impressive performance improvements but also illustrate the potential within collaborative interdisciplinary efforts between particle physics and advanced computational techniques. As the LHC and similar experiments continue to explore uncharted territories, the role of such advanced algorithms is poised to be increasingly pivotal.