Paper: arXiv 2505.09551

Authors: Liexin Cheng, Xue Cheng, Shuaiqiang Liu

Abstract

A critical factor in adopting machine learning for time-sensitive financial tasks is computational speed, including model training and inference. This paper demonstrates that a broad class of such problems, especially those previously addressed using deep neural networks, can be efficiently solved using single-layer neural networks without iterative gradient-based training. This is achieved through the extreme learning machine (ELM) framework. ELM utilizes a single-layer network with randomly initialized hidden nodes and output weights obtained via convex optimization, enabling rapid training and inference. We present various applications in both supervised and unsupervised learning settings, including option pricing, intraday return prediction, volatility surface fitting, and numerical solution of partial differential equations. Across these examples, ELM demonstrates notable improvements in computational efficiency while maintaining comparable accuracy and generalization compared to deep neural networks and classical machine learning methods. We also briefly discuss theoretical aspects of ELM implementation and its generalization capabilities.

Complexity vs Empirical Score

  • Math Complexity: 6.5/10
  • Empirical Rigor: 4.0/10
  • Quadrant: Lab Rats — high math complexity, limited empirical testing

Why this score: The paper introduces advanced neural network theory and convex optimization with several mathematical proofs and bounds, scoring high in math complexity, but lacks detailed code, datasets, or backtest results, resulting in moderate empirical rigor.

Research Flowchart

  flowchart TD
  A["Research Goal: Develop a computationally efficient alternative to DNNs for time-sensitive financial tasks"] --> B["Methodology: Extreme Learning Machine<br>(Single-layer network with random weights)"]
  B --> C["Data: Financial time-series & surfaces<br>(Option prices, volatility, intraday returns)"]
  C --> D["Computational Process<br>1. Randomly initialize hidden nodes<br>2. Convex optimization for output weights<br>3. Rapid training & inference"]
  D --> E["Key Outcomes<br>• High computational efficiency<br>• Comparable accuracy to DNNs<br>• Strong generalization<br>• Applicable to option pricing, PDEs, forecasting"]

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 using extreme learning machines (ELM), a single‑layer neural network with random hidden nodes and convex‑optimised output weights, as a fast alternative to deep neural networks for quantitative finance tasks.

Source sentence
This paper demonstrates that a broad class of such problems, especially those previously addressed using deep neural networks, can be efficiently solved using single-layer neural networks without iterative gradient-based training.

Assumes 2

  • Hidden weights are randomly initialised and output weights are obtained via linear least squares (convex optimisation). model
    Source sentence
    ELM employs a single-hidden-layer architecture in which the hidden weights are randomly initialized and the output weights are determined via linear least squares.
  • No hyper‑parameter tuning is performed for ELM during training. model
    Source sentence
    We highlight the methodological difference that no parameter is tuned in ELM during the training process.

Data

  • S&P 500 options data · OptionMetrics options commercial
    January 4 – February 28, 2023 · S&P 500 options
    Source sentence
    This experiment evaluates ELM for reconstructing implied volatility surfaces (IVS) using S&P 500 options data sourced from OptionMetrics (January 4 – February 28, 2023).

Recognised sources: OptionMetrics IvyDB

Costs

Not applicable The paper reports no trading or execution result.

Finds 5

  • ELM achieves orders‑of‑magnitude faster training and inference than deep neural networks and Gaussian process regression while maintaining comparable accuracy and generalisation. performance
    Source sentence
    Across these examples, ELM demonstrates notable improvements in computational efficiency while maintaining comparable accuracy and generalization compared to deep neural networks and classical machine learning methods.
  • In high‑frequency stock return forecasting, ELM classifiers outperform logistic regression in both accuracy and training speed (approximately 20× faster). predictive-accuracy
    Source sentence
    ELM classifiers outperform logistic regression in both accuracy and training speed (20× faster), which demonstrates its advantages for intraday trading data analysis.
  • ELM constructs arbitrage‑free implied volatility surfaces with smoother derivatives and fewer arbitrage violations than Gaussian process regression. pricing-calibration
    Source sentence
    ELM generates arbitrage‑free surfaces with smoother derivatives and fewer arbitrage violations than GPR, making it more robust and practical in applications.
  • ELM‑based physics‑informed neural networks solve Black‑Scholes‑type PDEs orders‑of‑magnitude faster than traditional DNN implementations. performance
    Source sentence
    ELM‑based PINNs solve Black‑Scholes‑type equations orders‑of‑magnitude faster than traditional DNN implementations, which results from its formation of the problem as a linear equation system.
  • ELM achieves training speed orders of magnitude faster than DNNs and GPR for learning pricing functions in stochastic volatility models, with comparable pricing accuracy. performance
    Source sentence
    of magnitude faster than DNNs and GPR. This efficiency advantage persists even when comparing the inference speed, with ELM several times faster while maintaining comparable pricing accuracy.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 50 related papers). Coverage is partial, so this is neither replication nor refutation.

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.