Quantitative Finance Paper Adaptations / 量化金融领域论文创作

Adapting cutting-edge AI conference methodologies to quantitative finance 将前沿AI顶会方法论迁移至量化金融领域

---

Overview / 概述

This repository contains 9 full-length conference papers that adapt core methodologies from top-tier AI/ML conferences (CVPR, ICML, ICLR) to quantitative finance problems. Each paper is accompanied by a complete, runnable Python implementation.

本仓库包含9篇完整的会议论文，每篇论文将顶会AI/ML核心方法论迁移至量化金融问题。每篇论文均附有完整的、可运行的Python实现代码。

Paper Collection / 论文列表

#	Source Paper / 源论文	Venue / 会议	Core Method / 核心方法	Quant Finance Paper / 量化金融论文	Quant Area / 量化方向
1	Thinking with Video	CVPR 2026	Video generation as reasoning / 视频生成推理	Thinking with Time-Series	Multi-horizon market reasoning / 多周期市场推理
2	Markovian Scale Prediction	CVPR 2026	Sliding window multi-scale / 滑动窗口多尺度	Markovian Multi-Resolution Forecasting	Multi-resolution price prediction / 多分辨率价格预测
3	SAM 3D	CVPR 2026	Single-image 3D reconstruction / 单图3D重建	SAM-Vol: Segment Anything for Volatility	Volatility surface reconstruction / 波动率曲面重建
4	GRPO is Secretly a PRM	ICML 2026	Implicit process reward / 隐式过程奖励	GRPO for Trading Execution	Optimal trade execution / 交易执行优化
5	ESamp: Latent Distilling	ICML 2026	Latent exploration sampling / 潜空间探索采样	ESamp for Portfolio Diversification	Portfolio strategy diversity / 组合多样性探索
6	CEO-Bench	arXiv 2026	Long-horizon agent evaluation / 长程Agent评测	QuantBench: Long-Horizon Market Regimes	Long-horizon trading evaluation / 长周期交易评测
7	TROLL Trust Regions	ICLR 2026	Differentiable trust region / 离散可微信任区域	TROLL for Portfolio Risk	Portfolio risk management / 组合风险管理
8	Deep Ignorance	ICLR 2026	Pretraining data filtering / 预训练数据防篡改	Financial Deep Ignorance	Financial model safety / 金融模型安全
9	Societies of Thought	Google 2026	Internal multi-agent reasoning / 内部多智能体推理	Investment Societies of Thought	Multi-expert investment reasoning / 多专家投资决策

Repository Structure / 仓库结构

quant-finance-papers/
├── README.md                              # This file / 本文件
├── papers/
│   ├── thinking-with-time-series/         # Thinking with Video → Multi-horizon market reasoning
│   │   ├── paper.md                       # Full paper (~7,400 words)
│   │   └── code/
│   │       ├── main.py                    # Training & evaluation script
│   │       ├── model.py                   # Diffusion-based trajectory generator
│   │       ├── data.py                    # MarketThinkBench synthetic data
│   │       └── requirements.txt
│   ├── markovian-price-prediction/        # Markovian Scale Prediction → Price forecasting
│   │   ├── paper.md
│   │   └── code/ (main.py, model.py, data.py, requirements.txt)
│   ├── sam-volatility/                    # SAM 3D → Volatility surface
│   │   ├── paper.md
│   │   └── code/ (main.py, model.py, data.py, requirements.txt)
│   ├── grpo-trading-execution/            # GRPO-PRM → Trade execution
│   │   ├── paper.md
│   │   └── code/ (main.py, model.py, data.py, requirements.txt)
│   ├── esamp-portfolio-diversity/         # ESamp → Portfolio diversification
│   │   ├── paper.md
│   │   └── code/ (main.py, model.py, data.py, requirements.txt)
│   ├── quantbench-long-horizon/           # CEO-Bench → Trading agent evaluation
│   │   ├── paper.md
│   │   └── code/ (main.py, model.py, data.py, requirements.txt)
│   ├── troll-portfolio-risk/              # TROLL → Portfolio risk
│   │   ├── paper.md
│   │   └── code/ (main.py, model.py, data.py, requirements.txt)
│   ├── financial-deep-ignorance/          # Deep Ignorance → Financial model safety
│   │   ├── paper.md
│   │   └── code/ (main.py, model.py, data.py, requirements.txt)
│   └── investment-societies/              # Societies of Thought → Investment reasoning
│       ├── paper.md
│       └── code/ (main.py, model.py, data.py, requirements.txt)
└── LICENSE

Domain Adaptation Methodology / 领域迁移方法论

Each paper follows a systematic adaptation pipeline:

每篇论文遵循系统化的迁移流程：

1. Extract Core Mechanism / 提取核心机制 — Abstract the mathematical/algorithmic essence from the source paper, stripping away domain-specific details.
2. Map to Quant Finance Problem / 映射量化金融问题 — Identify structurally analogous unsolved or under-explored problems in quantitative finance.
3. Domain-Specific Constraints / 领域特定约束 — Introduce finance-specific constraints: liquidity constraints, transaction costs, regulatory requirements, non-stationarity of financial time series.
4. Finance Datasets & Metrics / 金融数据集与指标 — Replace generic benchmarks with financial datasets (S&P 500, options chains, institutional orders) and evaluation metrics (Sharpe ratio, maximum drawdown, Calmar ratio, implementation shortfall).
5. Bilingual Presentation / 中英双语撰写 — Bridge the ML community and quantitative finance community terminology systems.

Paper Specifications / 论文规格

Each full paper includes:

• Bilingual Abstract / 双语摘要 (~300 words each language)
• Introduction / 引言 with problem motivation, gap analysis, 4-5 contributions
• Related Work / 相关工作 with 15-35 citations across 2-3 subsections
• Method / 方法 with formal mathematical definitions, theorems with proof sketches, algorithm pseudocode
• Experiments / 实验 with main results tables, ablation studies, hyperparameter sensitivity analysis
• Discussion & Conclusion / 讨论与结论 including limitations and broader impact
• References / 参考文献 (25-35 per paper)

Code Specifications / 代码规格

Each implementation includes:

• data.py — Synthetic data generation with realistic financial properties (regime switching, intraday volume patterns, multi-factor returns)
• model.py — Core PyTorch model architecture with proper docstrings and type hints
• main.py — End-to-end training and evaluation script with argparse CLI (supports train/eval/ablation modes)
• requirements.txt — Dependencies (numpy, torch, pandas, scipy, scikit-learn)

All code generates synthetic data and runs end-to-end without requiring external datasets.

Statistics / 统计

Metric / 指标	Value / 数值
Total Papers / 论文总数	9
Total Words / 总字数	~57,800
Total Code Lines / 总代码行数	~10,228
Python Files / Python文件数	27 (+ 9 requirements.txt)
Avg Words per Paper / 每篇均字数	~6,400
Avg Citations per Paper / 每篇均引用	~28

Quick Start / 快速开始

# Clone the repository / 克隆仓库
git clone https://github.com/sjkncs/quant-finance-papers.git
cd quant-finance-papers

# Run any paper's code / 运行任意论文的代码
cd papers/grpo-trading-execution/code
pip install -r requirements.txt
python main.py --mode train

# Or run evaluation only / 或仅运行评估
python main.py --mode eval

License / 许可证

MIT License

Citation / 引用

If you find this work useful, please cite:

@misc{quant-finance-papers-2026,
  title  = {Quantitative Finance Paper Adaptations: Bridging AI Conference Methods and Quant Finance},
  author = {AI Research Writing Community},
  year   = {2026},
  url    = {https://github.com/sjkncs/quant-finance-papers}
}