Neural network biased corrections: Cautionary study in background corrections for quenched jets

Summary

This study investigates biases in neural network (NN) background corrections for quenched jets in heavy-ion collisions. NNs are trained on pp jets and used to correct quenched jets, resulting in significant biases dependent on jet-pT.

Highlights

• NNs are trained on pp jets to correct quenched jets in heavy-ion collisions.
• Biases in NN background corrections are significant and dependent on jet-pT.
• The study uses JETSCAPE to generate hydrodynamically modeled QGP and quenched jets.
• The biases are observed to be jet-pT dependent, unlike area-based background corrections.
• The study demonstrates the magnitude of the error introduced by these biases on actual measurements.
• NNs trained on different parameters result in systematically lower RLeadJet values.
• The study suggests using an iterative method to refine the value of modeled quenching.

Key Insights

• The use of NNs for background corrections in quenched jets introduces significant biases, which are dependent on jet-pT. This is a critical issue in accurately measuring jet quenching in heavy-ion collisions.
• The biases in NN background corrections are qualitatively different from those in area-based background corrections, which are independent of jet substructure.
• The study highlights the importance of considering the effects of jet quenching on NN background corrections, as these biases can lead to significant errors in actual measurements.
• The use of different NNs trained on various parameters results in systematically lower RLeadJet values, indicating that the choice of NN and training parameters is crucial in accurately measuring jet quenching.
• The study suggests that using an iterative method to refine the value of modeled quenching may help mitigate the biases introduced by NN background corrections.
• The use of ML to distinguish between jet-like objects and background particles may be a promising approach to reduce the abundance of fake jets in measurements.
• The study emphasizes the importance of considering the limitations and biases of NN background corrections in accurately measuring jet quenching in heavy-ion collisions.

Mindmap

Citation

Stewart, D., & Putschke, J. (2024). Neural network biased corrections: Cautionary study in background corrections for quenched jets (Version 1). arXiv. https://doi.org/10.48550/ARXIV.2412.15440