pySigLib -- Fast Signature-Based Computations on CPU and GPU

2025/09/12 by Shmelev, Daniil, Salvi, Cristopher · 2 citations
#60L10 #65Y05 #68T99 #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Mathematical Software (cs.MS)

paper · doi:10.48550/arxiv.2509.10613

Abstract

Signature-based methods have recently gained significant traction in machine learning for sequential data. In particular, signature kernels have emerged as powerful discriminators and training losses for generative models on time-series, notably in quantitative finance. However, existing implementations do not scale to the dataset sizes and sequence lengths encountered in practice. We present pySigLib, a high-performance Python library offering optimised implementations of signatures and signature kernels on CPU and GPU, fully compatible with PyTorch's automatic differentiation. Beyond an efficient software stack for large-scale signature-based computation, we introduce a novel differentiation scheme for signature kernels that delivers accurate gradients at a fraction of the runtime of existing libraries.

Citations

• Quantum Path Signatures
• Structured Linear CDEs: Maximally Expressive and Parallel-in-Time Sequence Models
• ParallelFlow: Parallelizing Linear Transformers via Flow Discretization
• Pricing American options under rough volatility using deep-signatures and signature-kernels
• Sparse Signature Coefficient Recovery via Kernels
• Exact Gradients for Stochastic Spiking Neural Networks Driven by Rough Signals
• Log-PDE Methods for Rough Signature Kernels
• A path-dependent PDE solver based on signature kernels
• Signature Kernel Conditional Independence Tests in Causal Discovery for Stochastic Processes
• Signature volatility models: pricing and hedging with Fourier
• Signature Methods in Stochastic Portfolio Theory
• A Neural RDE approach for continuous-time non-Markovian stochastic control problems
• Non-adversarial training of Neural SDEs with signature kernel scores
• Optimal Stopping via Distribution Regression: a Higher Rank Signature Approach
• Neural signature kernels as infinite-width-depth-limits of controlled ResNets
• New directions in the applications of rough path theory
• A structure theorem for streamed information
• Neural Stochastic PDEs: Resolution-Invariant Learning of Continuous Spatiotemporal Dynamics
• Higher Order Kernel Mean Embeddings to Capture Filtrations of Stochastic\n Processes
• SigGPDE: Scaling Sparse Gaussian Processes on Sequential Data
• SK-Tree: a systematic malware detection algorithm on streaming trees via\n the signature kernel
• Neural Rough Differential Equations for Long Time Series
• A Data-driven Market Simulator for Small Data Environments
• Distribution Regression for Sequential Data
• Sig-SDEs model for quantitative finance
• Signatory: differentiable computations of the signature and logsignature transforms, on both CPU and GPU
• PyTorch: An Imperative Style, High-Performance Deep Learning Library
• Deep Signature Transforms
• The iisignature library: efficient calculation of iterated-integral signatures and log signatures
• A Primer on the Signature Method in Machine Learning
• Kernels for sequentially ordered data
• The Signature of a Rough Path: Uniqueness
• Extracting information from the signature of a financial data stream

Cited by

• Novelty detection on path space
• Explicit and Effectively Symmetric Schemes for Neural SDEs on Lie Groups

Related

• A User's Guide to KSig: GPU-Accelerated Computation of the Signature Kernel
• Keras Sig: Efficient Path Signature Computation on GPU in Keras 3
• The Signature Kernel
• Scalable Signature Kernel Computations for Long Time Series via Local Neumann Series Expansions
• Embedding and learning with signatures