# hls4ml
**Domain:** Machine Learning / FPGA / Real-Time Inference
**Doc Type:** Technical Toolchain Node
**Maturity:** Deployed and Evolving
**Primary Source:** [[articles/The Art is Long|The Art is Long: Vespucci of Immortality]]
**Related:** [[wiki/Observer Stack|Observer Stack]], [[wiki/LFADS|LFADS]], [[wiki/Closed-Loop BCI|Closed-Loop BCI]], [[wiki/CERN|CERN]]
---
## Definition
**hls4ml** is a toolchain for translating trained machine-learning models into high-level-synthesis representations that can be implemented as FPGA or ASIC firmware for extremely low-latency inference.
## Cross-Domain Migration
The toolchain emerged from particle-physics trigger requirements, where inference must occur before detector buffers expire. It was later used to deploy an LFADS neural-dynamics model on FPGA hardware for real-time neuroscience. This is a direct migration of an observer technology across scientific domains, not merely a metaphorical resemblance.
## Sources / Provenance
- [Fast inference of deep neural networks in FPGAs for particle physics](https://arxiv.org/abs/1804.06913).
- [FPGA Deployment of LFADS for Real-time Neuroscience Experiments](https://arxiv.org/abs/2402.04274).