Spiking Neural Network inference on FPGAs with hls4ml

Spiking Neural Networks (SNNs) provide a naturally temporal machine-learning framework. Their neurons maintain an internal state and propagate information through discrete spikes, enabling low-latency temporal inference. Although SNNs are often associated with asynchronous neuromorphic processors, many scientific real-time inference systems rely on conventional synchronous field-programmable gate arrays (FPGAs) and high-level synthesis (HLS) workflows. In this paper we present an extension of hls4ml that enables clock-driven deployment of SNNs trained in pytorch onto FPGA firmware. We demonstrate the workflow using a dense quantised SNN trained on the Heidelberg Spiking Digits dataset where it achieves inference latencies of approximately $34μ$s. We validate the generated design through software reference comparisons, HLS C simulation, HLS synthesis, export, and Vivado synthesis reports. This work opens up the hls4ml toolkit to neuromorphic computing, allowing streamlined optimisation, synthesis, and deployment of SNN models for real-time inference.

Anomaly detection with spiking neural networks for LHC physics

Anomaly detection offers a promising strategy for discovering new physics at the Large Hadron Collider (LHC). This paper investigates AutoEncoders built using neuromorphic Spiking Neural Networks (SNNs) for this purpose. One key application is at the trigger level, where anomaly detection tools could capture signals that would otherwise be discarded by conventional selection cuts. These systems must operate under strict latency and computational constraints. SNNs are inherently well-suited for low-latency, low-memory, real-time inference, particularly on Field-Programmable Gate Arrays (FPGAs). Further gains are expected with the rapid progress in dedicated neuromorphic hardware development. Using the CMS ADC2021 dataset, we design and evaluate a simple SNN AutoEncoder architecture. Our results show that the SNN AutoEncoders are competitive with conventional AutoEncoders for LHC anomaly detection across all signal models.

Anomalies, Representations, and Self-Supervision

We develop a self-supervised method for density-based anomaly detection using contrastive learning, and test it using event-level anomaly data from CMS ADC2021. The AnomalyCLR technique is data-driven and uses augmentations of the background data to mimic non-Standard-Model events in a model-agnostic way. It uses a permutation-invariant Transformer Encoder architecture to map the objects measured in a collider event to the representation space, where the data augmentations define a representation space which is sensitive to potential anomalous features. An AutoEncoder trained on background representations then computes anomaly scores for a variety of signals in the representation space. With AnomalyCLR we find significant improvements on performance metrics for all signals when compared to the raw data baseline.

Better Latent Spaces for Better Autoencoders

Autoencoders as tools behind anomaly searches at the LHC have the structural problem that they only work in one direction, extracting jets with higher complexity but not the other way around. To address this, we derive classifiers from the latent space of (variational) autoencoders, specifically in Gaussian mixture and Dirichlet latent spaces. In particular, the Dirichlet setup solves the problem and improves both the performance and the interpretability of the networks.

Bump Hunting in Latent Space

Unsupervised anomaly detection could be crucial in future analyses searching for rare phenomena in large datasets, as for example collected at the LHC. To this end, we introduce a physics inspired variational autoencoder (VAE) architecture which performs competitively and robustly on the LHC Olympics Machine Learning Challenge datasets. We demonstrate how embedding some physical observables directly into the VAE latent space, while at the same time keeping the classifier manifestly agnostic to them, can help to identify and characterise features in measured spectra as caused by the presence of anomalies in a dataset.