Resources

Publications & Resources

Technical papers, how-to guides, and presentations on automatic differentiation and high-performance quantitative computing

Featured Comparison

Comprehensive guide to choosing the right AAD approach

AAD Technologies Comparison

Comprehensive comparison of Tape-based, Code Transformation, and Code Generation AAD approaches. Make an informed decision for your quantitative finance systems with detailed analysis of integration complexity, memory usage, performance characteristics, and scalability.

Banking & Finance

AAD research and whitepapers for quantitative finance

AAD for Functions Involving Expectations

Methods for computing gradients of functions involving expectations, commonly used in stochastic model calibration. Presents faster implementation approaches with parallelization, demonstrated on European option calibration.

  • Gradients of functions involving expectations
  • Faster AAD implementation approaches
  • Parallelization strategies for calibration
Download PDF

Stochastic AAD for Financial Model Calibration

Explores automatic adjoint differentiation for calibration functions involving expectations, enabling effective parallel processing with detailed computational cost analysis and numerical validation.

  • AAD for calibration functions with expectations
  • Parallel processing strategies for gradient computation
  • Computational cost analysis for stochastic AAD
Download PDF

Run Legacy CUDA Code on CPU with Comparable Performance + AAD

Run legacy CUDA code on CPU with comparable performance + added AAD using minimal changes.

  • Minimal code changes for dual GPU/CPU
  • Free AAD on CPU
Download PDF

Adjoint Differentiation for Matrix Functions

Derives formulas for computing adjoints of square-matrix operations C=f(A) where f is holomorphic. Applies to spectral decomposition, spectrum cut-off, correlation matrices, and regularized regression.

  • Formula for adjoints of matrix operations C=f(A)
  • Spectral decomposition adjoint computation
  • Correlation matrix routine adjoints
Download PDF

Independent Review: 300x Speedup on 10K OIS Swap Portfolio

Independent blog showing ~300x speedup on 10K OIS swap portfolio revaluation with AADC.

  • Simple kernel compilation example
Download PDF

Accelerate Python Models with AADC

Prototype in Python, achieve C++ performance. Use AADC to accelerate your existing Python pricing models with minimal code changes — no rewrites required.

  • How to integrate AADC into existing Python models
  • Step-by-step methodology for model acceleration
  • Performance comparison: Python vs AADC vs hand-optimised C++
  • Why AADC guarantees mathematically exact results
  • Practical workflow for accelerating quant models
Download PDF

Novel Hybrid AAD That Removes Per-Sample Overhead

Novel hybrid operator overloading + code generation AAD that removes per-sample overhead and enables multithreading/vectorization.

  • Hybrid OO×CT AAD design
  • Elimination of tape interpretation overhead
  • Static adjoint function generation
Download PDF

Using Claude AI to Accelerate Python Models 345× with AADC

A practical guide to using Claude AI for AADC integration: accelerate your existing Python pricing models 345× with minimal code changes — in under half a day.

  • How to use Claude AI to integrate AADC into existing Python models
  • Step-by-step methodology for AI-assisted model acceleration
  • Performance comparison: Python vs AADC vs hand-optimised C++
  • Why AADC eliminates AI hallucination risks in numerical results
  • Practical workflow for accelerating quant models with AI assistance
Download PDF

Graph-Based Automatic Parallelization with MatLogica

Paper describing MatLogica's graph-based parallelization and kernel generation approach.

  • Novel automatic parallelism from valuation graphs
Download PDF

Intel & MatLogica: High-Performance AAD on Modern Xeon CPUs

Intel/MatLogica collaboration whitepaper on achieving high-performance AAD and pricing on modern Xeon CPUs.

  • High-performance AAD on Intel hardware
  • Benchmark results vs competitors
Download PDF

Technical Guide: Integrating AADC with Open Risk Engine

Technical article explaining how to transform ORE into a streamlined LiveRisk service using AADC, including code integration points and handling control flow branches.

  • How to identify input/output points in ORE
  • How to hook AADC into CSVLoader for market data
  • How to mark trade NPVs as outputs
  • How to handle control flow branches in AADC recording
Download PDF

Transform ORE into a High-Performance LiveRisk Service

Jupyter notebook demonstrating how to use AADC to transform the Open Risk Engine into a fast LiveRisk service for 1M FX linear trades with sub-second pricing and risk calculations.

  • Transform ORE into a LiveRisk service using AADC
  • Record AADC kernels for ultra-fast repricing
  • Calculate delta sensitivities in under 1 second
  • Handle intraday portfolio changes without re-recording
  • Perform market scenario analysis
Download PDF

Neural Networks

Machine learning and deep learning research

ARN: Automatic Synthesis of Neurons for Recurrent Neural Nets

Novel ARN neurons achieve up to 3x better accuracy than LSTMs on time series tasks through automatic synthesis of recurrent architectures with skip connections and novel activation functions.

  • How ARNs outperform LSTMs by up to 3x on cross entropy
  • Novel skip connections across time steps
  • Multiple internal memory states per neuron
  • Quadratic activation function components
Read Paper

Deep Learning for Oil Well Monitoring and Anomaly Detection

IEEE Computer Society publication on advanced computer vision techniques for monitoring oil wells using deep learning.

  • Application of CNNs to well imagery
  • Handling noisy industrial data
  • Integration with IoT sensors
Read Paper

Automatic Synthesis of Recurrent Neurons for CNC Imitation Learning

IEEE paper exploring innovative automatic synthesis of recurrent neurons to enable imitation learning directly from expert CNC machine operators.

  • Synthesis of recurrent neural architectures
  • Behavioral cloning from operator demonstrations
  • Application to sequential control tasks
Read Paper

How-To Guides

Step-by-step guides for implementing AADC in your projects

Accelerate QuantLib 6-100x with Minimal Changes

Step-by-step guide to accelerating QuantLib and ORE with AADC integration requiring less than 1% code changes.

  • Integration steps required
  • Performance gains achievable
  • AAD enablement process
  • Minimal code changes needed (<1%)
View Guide

Accelerate Python Models with AADC

Prototype in Python, achieve C++ performance. Use AADC to accelerate your existing Python pricing models with minimal code changes — no rewrites required.

  • How to integrate AADC into existing Python models
  • Step-by-step methodology for model acceleration
  • Performance comparison: Python vs AADC vs hand-optimised C++
  • Why AADC guarantees mathematically exact results
  • Practical workflow for accelerating quant models
View Guide

Using Claude AI to Accelerate Python Models 345× with AADC

A practical guide to using Claude AI for AADC integration: accelerate your existing Python pricing models 345× with minimal code changes — in under half a day.

  • How to use Claude AI to integrate AADC into existing Python models
  • Step-by-step methodology for AI-assisted model acceleration
  • Performance comparison: Python vs AADC vs hand-optimised C++
  • Why AADC eliminates AI hallucination risks in numerical results
  • Practical workflow for accelerating quant models with AI assistance
View Guide

Step-by-Step Guide to Accelerating Legacy Quant Libraries

Transform ORE/QuantLib into a LiveRisk service: 1M FX trades priced in 0.4 seconds (100x faster) with complete delta risk. Black box approach - no C++ refactoring needed.

  • Black box approach - no C++ hierarchy knowledge needed
  • 100x+ speedup for FX pricing with complete delta risk
  • Hook inputs with markAsInput(), outputs with markAsOutput()
  • Pre-market recording, sub-second intraday updates
  • Works with ORE, QuantLib, and proprietary libraries
View Guide

Technical Guide: Integrating AADC with Open Risk Engine

Technical article explaining how to transform ORE into a streamlined LiveRisk service using AADC, including code integration points and handling control flow branches.

  • How to identify input/output points in ORE
  • How to hook AADC into CSVLoader for market data
  • How to mark trade NPVs as outputs
  • How to handle control flow branches in AADC recording
View Guide

Conference Presentations

Presentations from industry conferences and technical events

Automatic IFT for Live Risk Transition

QuantMinds presentation demonstrating how Automated Implicit Function Theorem enables transition from overnight batch risk to real-time Live Risk without multi-year IT transformation.

  • Transform overnight batch risk into real-time Live Risk
  • Automatic differentiation of calibration routines using AIFT
  • No hidden variable tracking or large Jacobian computation required
  • QuantLib integration achieving 20ms pricing + Greeks for 1000 swaps
Watch

WBS Talk: AAD for Longstaff-Schwartz American Monte Carlo

Conference talk on AAD for Longstaff-Schwartz American Monte Carlo with memory-efficient implementation.

  • Pathwise + regression AAD
  • Memory only 2x primal
  • Better stability than bumping
Watch

WBS2025: Greeks for Autocallables

Presentation from the 21st WBS Quantitative Finance Conference in Palermo demonstrating production-ready autocallable Greeks calculation with 99.8% cloud cost reduction.

  • Production-ready autocallable Greeks methodology
  • How to achieve 99.8% cloud cost reduction
  • Smoothing techniques for discontinuous payoffs
  • Complete open-source implementation
Watch

ING Analysis: Cross-Gamma Performance Comparison

ING comparison of bumping-over-AAD for second-order cross-gamma (CVA portfolio).

  • Scalability of tape vs code-gen AAD for large Hessians
Watch

ING Presentation: AAD Engine Scalability Analysis

ING presentation comparing AAD engine scalability for Monte Carlo and higher-order sensitivities.

  • Linear vs quadratic scaling in AAD engines
  • When code-generation wins
Watch

WBS2023: Expected Shortfall with AAD

Presentation on Expected Shortfall and AAD computational costs.

  • AAD application to regulatory risk metrics
Watch

Learn More

Understand the technology behind our publications

How AADC Works

AADC extracts a computational graph (DAG) from existing analytics at runtime and compiles it to optimized machine code, achieving 100-1000× for Python and 6-100× for C++ with automatic derivatives.

Learn More

Performance Benchmarks

Independent benchmarks demonstrating AADC's performance advantages over traditional AAD approaches, GPU implementations, and ML frameworks.

Learn More

Have Questions About Our Research?

Our team can discuss technical details and provide custom presentations

Schedule Discussion Contact Us