Paper: arXiv 2602.11020

Authors: Rui Ma

Abstract

We study same-source multi-view learning and adversarial robustness for next-day direction prediction using two deterministic, window-aligned image views derived from the same time series: an OHLCV-rendered chart (ohlcv) and a technical-indicator matrix (indic). To control label ambiguity from near-zero moves, we use an ex-post minimum-movement threshold min_move (tau) based on realized absolute next-day return, defining an offline benchmark on the subset where the absolute next-day return is at least tau. Under leakage-resistant time-block splits with embargo, we compare early fusion (channel stacking) and dual-encoder late fusion with optional cross-branch consistency. We then evaluate pixel-space L-infinity evasion attacks (FGSM/PGD) under view-constrained and joint threat models. We find that fusion is regime dependent: early fusion can suffer negative transfer under noisier settings, whereas late fusion is a more reliable default once labels stabilize. Robustness degrades sharply under tiny budgets with stable view-dependent vulnerabilities; late fusion often helps under view-constrained attacks, but joint perturbations remain challenging.

Complexity vs Empirical Score

  • Math Complexity: 3.5/10
  • Empirical Rigor: 7.5/10
  • Quadrant: Street Traders — practical and empirical, lighter on theory

Why this score: The paper provides a rigorous empirical evaluation of multi-view image fusion and adversarial robustness for financial time series using a 20-year gold spot dataset. While the mathematical formulations are straightforward standard deep learning and evasion attack definitions, the methodology addresses critical backtesting practices such as embargoed time-block splits and label-ambiguity filtering.

Research Flowchart

  flowchart TD
    A[Research Goal: Assess Multi-View Fusion & Robustness in Next-Day Stock Direction] --> B[Data Inputs: Time-Aligned OHLCV Charts & Technical Indicator Matrices]
    B --> C[Labeling & Splitting: Apply Threshold min_move \n& Leakage-Resistant Time Block Split]
    C --> D[Model Training & Fusion: Compare Early Stacking vs. Dual-Encoder Late Fusion]
    D --> E[Adversarial Evaluation: Apply FGSM & PGD Evasion Attacks]
    E --> F[Findings: Late Fusion Improves Stability & View-Constrained Defense]
    E --> G[Findings: Early Fusion Suffers Negative Transfer; Joint Perturbations Break Models]

Cited References (37)

  1. Convolutional Neural Network Applied to Gold Price Forecasting with an Image Integration Methods Based on Multi-Sources and Heterogeneous Data (2022) — link
  2. Autoregressive Conditional Models for Interval-Valued Time Series Data (Preliminary Version) (2012)
  3. A decade systematic review of fusion techniques in financial market prediction (2016–2025) (2025) — link
  4. Algorithmic financial trading with deep convolutional neural networks: Time series to image conversion approach (2018) — link
  5. Encoding Candlesticks as Images for Patterns Classification Using Convolutional Neural Networks (2020) — link
  6. A novel graph convolutional feature based convolutional neural network for stock trend prediction (2021) — link
  7. (Re-)Imag(in)ing Price Trends (2020) — link
  8. Imaging Time-Series to Improve Classification and Imputation (2015) — link
and 29 more
  1. Forecasting stock prices with a feature fusion LSTM-CNN model using different representations of the same data (2019) — link
  2. A stock time series forecasting approach incorporating candlestick patterns and sequence similarity (2022) — link
  3. Intelligent candlestick forecast system for financial time-series analysis using metaheuristics-optimized multi-output machine learning (2022) — link
  4. Financial Trading Model with Stock Bar Chart Image Time Series with Deep Convolutional Neural Networks (2020) — link
  5. Neural Network and Order Flow, Technical Analysis: Predicting short-term direction of futures contract (2022) — link
  6. Explaining and Harnessing Adversarial Examples (2015) — link
  7. Towards Deep Learning Models Resistant to Adversarial Attacks (2018) — link
  8. Adversarial Robustness of Deep Convolutional Candlestick Learner (2020) — link
  9. The probability of backtest overfitting (2017) — link
  10. The Deflated Sharpe Ratio: Correcting for Selection Bias, Backtest Overfitting, and Non-Normality (2014) — link
  11. Adversarial Attacks on Deep Neural Networks for Time Series Classification (2019) — link
  12. Multimodal machine learning: A survey and taxonomy (2019) — link
  13. Multimodal Co-learning: Challenges, applications with datasets, recent advances and future directions (2022) — link
  14. Incorporating stock prices and text for stock movement prediction based on information fusion (2024) — link
  15. Advances in Financial Machine Learning (2018)
  16. Consistent cross-validatory model-selection for dependent data: hv-block cross-validation (2000) — link
  17. Deep Residual Learning for Image Recognition (2016) — link
  18. ImageNet: A Large-Scale Hierarchical Image Database (2009) — link
  19. Distilling the Knowledge in a Neural Network (2015) — link
  20. Deep Mutual Learning (2018) — link
  21. Comparison of the predicted and observed secondary structure of T4 phage lysozyme (1975) — link
  22. Intriguing Properties of Neural Networks (2014) — link
  23. Adversarial Machine Learning at Scale (2017) — link
  24. Towards Evaluating the Robustness of Neural Networks (2017) — link
  25. Obfuscated Gradients Give a False Sense of Security: Circumventing Defenses to Adversarial Examples (2018) — link
  26. Defending Multimodal Fusion Models Against Single-Source Adversaries (2021) — link
  27. Tushare: Financial Data Interface (2025)
  28. (Re-)Imag(in)ing Price Trends (2023) — link
  29. LastingBench: Defend Benchmarks Against Knowledge Leakage (2025) — link

0 of 37 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2518 linked papers.