We present the implementation of binary and ternary neural networks in the hls4ml library, designed to automatically convert deep neural network models to digital circuits with FPGA firmware. Starting from benchmark models trained with floating point precision, we investigate different strategies to reduce the network's resource consumption by reducing the numerical precision of the network parameters to binary or ternary. We discuss the trade-off between model accuracy and resource consumption. In addition, we show how to balance between latency and accuracy by retaining full precision on a selected subset of network components. As an example, we consider two multiclass classification tasks: handwritten digit recognition with the MNIST data set and jet identification with simulated proton-proton collisions at the CERN Large Hadron Collider. The binary and ternary implementation has similar performance to the higher precision implementation while using drastically fewer FPGA resources.

Over the Arctic regions, current conventional altimetry products suffer from a lack of coverage or from degraded performance due to the inadequacy of the standard processing applied in the ground segments. This paper presents a set of dedicated algorithms able to process consistently returns from open ocean and from sea-ice leads in the Arctic Ocean (detection of water surfaces and derivation of water levels using returns from these surfaces). This processing extends the area over which a precise sea level can be computed. In the frame of the European Space Agency Sea Level Climate Change Initiative (http://cci.esa.int), we have first developed a new surface identification method combining two complementary solutions, one using a multiple-criteria approach (in particular the backscattering coefficient and the peakiness coefficient of the waveforms) and one based on a supervised neural network approach. Then, a new physical model has been developed (modified from the Brown model to include anisotropy in the scattering from calm protected water surfaces) and has been implemented in a maximum likelihood estimation retracker. This allows us to process both sea-ice lead waveforms (characterized by their peaky shapes) and ocean waveforms (more diffuse returns), guaranteeing, by construction, continuity between open ocean and icecovered regions. This new processing has been used to produce maps of Arctic sea level anomaly from 18-Hz ENVIronment SATellite/RA-2 data.

We introduce an automated tool for deploying ultra low-latency, low-power deep neural networks with convolutional layers on field-programmable gate arrays (FPGAs). By extending the hls4ml library, we demonstrate an inference latency of 5 µs using convolutional architectures, targeting microsecond latency applications like those at the CERN Large Hadron Collider. Considering benchmark models trained on the Street View House Numbers Dataset, we demonstrate various methods for model compression in order to fit the computational constraints of a typical FPGA device used in trigger and data acquisition systems of particle detectors. In particular, we discuss pruning and quantization-aware training, and demonstrate how resource utilization can be significantly reduced with little to no loss in model accuracy. We show that the FPGA critical resource consumption can be reduced by 97% with zero loss in model accuracy, and by 99% when tolerating a 6% accuracy degradation.

We describe the implementation of Boosted Decision Trees in the hls4ml library, which allows the translation of a trained model into FPGA firmware through an automated conversion process. Thanks to its fully on-chip implementation, hls4ml performs inference of Boosted Decision Tree models with extremely low latency. With a typical latency less than 100 ns, this solution is suitable for FPGA-based real-time processing, such as in the Level-1 Trigger system of a collider experiment. These developments open up prospects for physicists to deploy BDTs in FPGAs for identifying the origin of jets, better reconstructing the energies of muons, and enabling better selection of rare signal processes.

Vaccine development against hepatitis C virus (HCV) is hindered by poor understanding of factors defining cross-immunoreactivity among heterogeneous epitopes. Using synthetic peptides and mouse immunization as a model, we conducted a quantitative analysis of cross-immunoreactivity among variants of the HCV hypervariable region 1 (HVR1). Analysis of 26,883 immunological reactions among pairs of peptides showed that the distribution of cross-immunoreactivity among HVR1 variants was skewed, with antibodies against a few variants reacting with all tested peptides. The HVR1 cross-immunoreactivity was accurately modeled based on amino acid sequence alone. The tested peptides were mapped in the HVR1 sequence space, which was visualized as a network of 11,319 sequences. The HVR1 variants with a greater network centrality showed a broader cross-immunoreactivity. The entire sequence space is explored by each HCV genotype and subtype. These findings indicate that HVR1 antigenic diversity is extensively convergent and effectively limited, suggesting significant implications for vaccine development.