Paper: arXiv 2505.03789

Authors: Syoiti Ninomiya, Yuming Ma

Abstract

A new deep-learning neural network architecture based on high-order weak approximation algorithms for stochastic differential equations (SDEs) is proposed. The architecture enables the efficient learning of martingales by deep learning models. The behaviour of deep neural networks based on this architecture, when applied to the problem of pricing financial derivatives, is also examined. The core of this new architecture lies in the high-order weak approximation algorithms of the explicit Runge–Kutta type, wherein the approximation is realised solely through iterative compositions and linear combinations of vector fields of the target SDEs.

Complexity vs Empirical Score

  • Math Complexity: 8.5/10
  • Empirical Rigor: 4.0/10
  • Quadrant: Lab Rats — high math complexity, limited empirical testing

Why this score: The paper proposes a novel neural network architecture inspired by high-order weak approximation algorithms for SDEs, requiring advanced stochastic calculus and numerical analysis (e.g., Stratonovich integrals, Runge–Kutta methods, vector fields). While the derivative pricing application suggests practical relevance, the summary lacks implementation details, backtests, or empirical validation data, focusing instead on theoretical framework and methodology.

Research Flowchart

  flowchart TD
  A["Research Goal: <br>Learn Martingales via Deep Learning"] --> B{"Data: Financial SDEs &<br>Derivative Pricing Problems"}
  B --> C["Methodology: High-Order RK Architecture"]
  C --> D["Process: Iterative Composition &<br>Linear Combination of Vector Fields"]
  D --> E["Outcome: Efficient Pricing of<br>Financial Derivatives"]
  E --> F["Key Finding: High-Order Weak Approximation<br>enables accurate SDE learning"]

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper proposes a new deep‑learning neural network architecture based on high‑order weak approximation algorithms for stochastic differential equations to efficiently learn martingales.

Source sentence
A new deep-learning neural network architecture based on high-order weak approximation algorithms for stochastic differential equations (SDEs) is proposed.

Assumes 4

  • High‑order discretisation refers to discretisation of 2nd order or higher. model
    Source sentence
    Henceforth, what is referred to as high-order discretisation in this paper refers to discretisation of 2nd order or higher.
  • The Euler–Maruyama discretisation achieves 1st order accuracy under certain conditions. model
    Source sentence
    Under certain conditions, the Euler–Maruyama discretisation achieves 1st order accuracy.
  • The Adam optimiser updates parameters using moment estimates of gradients. statistical
    Source sentence
    Adam optimiser is an optimiser that updates parameters via dynamically scaling the update step based on moment estimation along each parameter’s gradient vector.
  • Quasi‑Monte Carlo integration uses 5000 Sobol sequence points. other
    Source sentence
    All numerical integrations included are proceeded by using Quasi-Monte Carlo method with KBIN = 5000 sample points generated from the generalised Sobol’ sequence.

Data

  • Quasi‑Monte Carlo Sobol sample points synthetic unknown
    Source sentence
    All numerical integrations included are proceeded by using Quasi-Monte Carlo method with KBIN = 5000 sample points generated from the generalised Sobol’ sequence.

Costs

Not applicable The paper reports no trading or execution result.

Finds 2

  • The proposed architecture enables efficient learning of martingales by deep learning models. theoretical
    Source sentence
    The architecture enables the efficient learning of martingales by deep learning models.
  • Theorem 2 allows the value of equation (18) to be used directly as the loss function for training. theoretical
    Source sentence
    The result of Theorem 2 allows for the utilisation of the value of (18) itself as the LOSS function.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 12 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 1 paper in the archive

Code

No repository by the authors found in the paper.

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.