LHCb

The reconstruction menu, and the gates that prune it

The inner OGTS loop. A named family tree of calorimeter architectures, pruned by throughput and resolution, not by a flat NAS scan.

Aug 31, 2026

Three talks at CHEP 2026 in Bangkok

Presented lightweight GNNs for LHCb's PicoCal, PyTorch-to-ONNX inference optimization, and RAG-enhanced agentic revival of a legacy Λb → Λγ analysis at the 28th Conference on Computing in High Energy and Nuclear Physics.

May 29, 2026

When Less is More: Towards Lightweight and Distilled Graph Neural Networks for Efficient Particle Reconstruction in LHCb's Next-Generation Calorimeter

Attention-enhanced node-centric GarNet GNNs (UzzieNet) with 8× faster inference, plus knowledge distillation down to ~95% smaller models that beat the teacher on energy resolution, for the PicoCal of LHCb Upgrade II.

May 25, 2026

Optimizing GNNs for the Wild: PyTorch-to-ONNX Acceleration of GarNet on CPUs and GPUs

Benchmarking GarNet in PyTorch versus ONNX Runtime on multi-core CPUs and NVIDIA A100 GPUs: up to 5× CPU and ~2× GPU speedup with FP32 parity at 10⁻⁷, targeting the LHCb GPU trigger (HLT1/Allen).

May 25, 2026

Future-Ready Restoration: A Case Study on AI RAG-Enhanced Agentic Revival of a Run-2 2016 Λb → Λγ Analysis

A RAG and agentic-LLM tool (Claude, GPT-family models with a vector database over analysis scripts, notes, and documentation plus the LHCb software stack) to reproduce and accelerate a legacy LHCb analysis.

May 25, 2026

UzzieNet / ecal-reco-gnn

Attention-enhanced node-centric GNN (UzzieNet) for LHCb PicoCal reconstruction, with distillation and PyTorch-to-ONNX deployment for the GPU trigger.

May 25, 2026

When Less is More: Towards Lightweight and Distilled Graph Neural Networks for Efficient Particle Reconstruction

Talk on attention-enhanced lightweight GNNs (UzzieNet) for real-time particle reconstruction in LHCb's next-generation calorimeter.

Sep 8, 2025

When Less is More: Lightweight and Distilled GNNs for Efficient Particle Reconstruction

Poster on optimizing graph neural networks and knowledge distillation for efficient particle reconstruction in LHCb's next-generation calorimeter.

Sep 1, 2025