Economical Jet Taggers -- Equivariant, Slim, and Quantized

TL;DR

This paper introduces a slim, quantized version of jet-tagging transformers that significantly reduces energy consumption and model size, potentially enabling real-time trigger-level applications at the LHC.

Contribution

It presents a novel, resource-efficient approach to jet tagging by slimming and quantizing Lorentz-equivariant transformers, improving their practicality for high-energy physics experiments.

Findings

Order-of-magnitude energy reduction with moderate performance loss

Effective quantization methods for standard and equivariant transformers

Feasibility of small, 1000-parameter jet taggers for real-time use

Abstract

Modern machine learning is transforming jet tagging at the LHC, but the leading transformer architectures are large, not particularly fast, and training-intensive. We present a slim version of the L-GATr tagger, reduce the number of parameters of jet-tagging transformers, and quantize them. We compare different quantization methods for standard and Lorentz-equivariant transformers and estimate their gains in resource efficiency. We find an order-of-magnitude reduction in energy cost for an moderate performance decrease, down to 1000-parameter taggers. This might be a step towards trigger-level jet tagging with small and quantized versions of the leading equivariant transformer architectures.