An elegant approach to run existing CUDA analytics on both GPU and CPU, with added benefit of AAD
We know NVIDIA GPU offers a massive number of CUDA Cores, but CPUs are not far behind. See our whitepaper that demonstrates how your CUDA analytics can be accelerated by AADC on a CPU with an option of AAD.
Case Study: How a Major European Bank Revolutionised Their Front-Office Risk Management Using MatLogica AADC
MatLogica’s AADC enabled the client to supercharge their analytics by introducing AAD for risk computations and to accelerate pricing and scenario analysis. The MatLogica-enhanced analytics unlocked new revenue streams, lowered infrastructure costs, and improved risk management.
Automatic Synthesis of Neurons for Recurrent Neural Nets
Prof. Roland Olsson and his team used MatLogica's AADC to design state-of-the-art neural network architectures for time series analysis. It is up-to 3x more accurate than the available cutting-edge methods and the training time is several times lower due to MatLogica’s technology.
Automatic Implicit Function Theorem
The paper demonstrates a way to apply the Implicit Function Theorem in a not widely known way, which is important for practical AAD application and performance, particularly with complex calibration
Adjoint Differentiation for Generic Matrix Functions
No doubt, AAD is amazing. However, implementing it in practice has a lot of subtleties. For instance, how to deal with operations requiring an SVD decomposition? Our researchers have found an elegant solution to this problem.
More Than a Thousand-fold Speedup for xVA Pricing Calculations with Intel® Xeon® Scalable Processors
Intel-led white paper demonstrating an up to 1770x performance increase for XVA pricing (and 830 for XVA risks!) on Intel processors when using Matlogica AADC. It is open-source and available at GitHub.
A New Approach to Parallel Computing Using Automatic Differentiation: Getting Top Performance on Modern Multicore Systems
A paper in Parallel Universe Magazine №40 featuring a new approach that turns object-oriented, single-thread, scalar code into AVX2/AVX512 vectorized multi-thread and thread-safe lambda functions with no runtime penalty
Open-Source Benchmark demonstrating a leap in performance for valuation and AAD risk calculations using AADC on Intel Scalable Xeon CPUs.
AAD and calibration
Remarks on stochastic automatic adjoint differentiation and calibration of financial models.
AAD: Breaking the Primal Barrier
Dmitri Goloubentsev and Evgeny Lakshtanov wrote an article for Wilmott Magazine on how merging Code Transformation and Operator Overloading techniques leads to a major performance boost.