pith. sign in

arxiv: 2509.25055 · v3 · pith:JCKN7HLMnew · submitted 2025-09-29 · 💱 q-fin.CP

AlphaSAGE: Structure-Aware Alpha Mining via GFlowNets for Robust Exploration

Binqi Chen , Hongjun Ding , Ning Shen , Jinsheng Huang , Taian Guo , Luchen Liu , Ming Zhang This is my paper

Pith reviewed 2026-05-21 21:46 UTC · model grok-4.3

classification 💱 q-fin.CP
keywords alpha miningGFlowNetsquantitative financerelational graph convolutional networksautomated trading signalsdiverse alpha portfoliosreinforcement learning alternativespredictive signal generation
0
0 comments X

The pith

AlphaSAGE uses a graph encoder and GFlowNets with dense rewards to generate diverse, novel, and predictive alpha portfolios instead of converging on single signals.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper seeks to fix three core problems in automated alpha mining: rewards that arrive only after a full formula is built, sequential representations that ignore the structure of mathematical expressions, and standard reinforcement learning that collapses to one optimal solution rather than producing varied non-correlated signals. It introduces a framework that represents formulas as graphs, encodes them with a Relational Graph Convolutional Network, samples candidates using Generative Flow Networks, and supplies feedback at every construction step through a multi-faceted reward. A reader would care because quantitative strategies need many weakly correlated signals to remain robust when markets shift, and current methods often fail to deliver that diversity at scale.

Core claim

AlphaSAGE is a framework built on a Relational Graph Convolutional Network encoder that captures structural properties of mathematical expressions, a Generative Flow Network that samples from a distribution over possible formulas to promote diversity, and a dense multi-faceted reward that provides intermediate feedback during expression construction, yielding empirical portfolios of alphas that are more diverse, novel, and predictive than those produced by existing reinforcement learning baselines.

What carries the argument

The GFlowNet sampling process guided by an RGCN structure-aware encoder and a dense multi-faceted reward function, which distributes probability mass across many high-reward expression paths rather than concentrating on a single mode.

If this is right

  • Portfolios assembled from the mined alphas exhibit lower average correlations, allowing better risk reduction in live trading.
  • The generated alphas display higher out-of-sample predictive accuracy than those from standard RL approaches.
  • Training becomes more stable because intermediate rewards replace the single terminal reward signal.
  • The method can produce a larger number of unique candidate formulas without exhaustive enumeration.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same structure-aware flow network could be applied to other tasks that require generating diverse mathematical expressions, such as symbolic regression or program synthesis.
  • In production systems, the diversity might improve adaptation across market regimes where individual signals decay at different rates.
  • Pairing the mined alphas with downstream portfolio optimization that explicitly penalizes residual correlations could amplify the practical benefit.
  • Extending the graph representation to include temporal or cross-asset relations might further improve capture of real-world alpha behavior.

Load-bearing premise

The graph representation of formulas and the multi-faceted reward function are assumed to capture the true factors that make alphas effective and non-correlated in live markets.

What would settle it

A controlled experiment on a held-out market dataset where the alphas mined by AlphaSAGE show no measurable gain in portfolio Sharpe ratio or pairwise correlation reduction compared with RL baselines would falsify the performance claim.

Figures

Figures reproduced from arXiv: 2509.25055 by Binqi Chen, Hongjun Ding, Jinsheng Huang, Luchen Liu, Ming Zhang, Ning Shen, Taian Guo.

Figure 1
Figure 1. Figure 1: Different forms of alpha: (a) For￾mulated alpha; (b) Alpha’s expression tree; (c) Reverse Polish Notation (RPN) for alpha. The direct application of RL to alpha mining is gen￾erally fraught with significant obstacles that limit its efficacy. First, current methodologies often suf￾fer from a severe “cold-start” problem, as the re￾ward signal—typically based on an alpha’s Infor￾mation Coefficient (IC)—is ext… view at source ↗
Figure 2
Figure 2. Figure 2: An Overview of AlphaSAGE. (a) AlphaGenerator. Starting from an empty state, a GNN encoder embeds the partial expression tree; a GFlowNet policy produces masked action logits over valid edits. We sample the next token, apply early stopping via a forced SEP when triggered, and continue until a terminal, valid alpha is formed. The resulting formula αn and its embedding en are retained for evaluation. (b) Alph… view at source ↗
Figure 3
Figure 3. Figure 3: Cumulative return on CSI300 (2022–2024). Comparison among AlphaSAGE (ours), all [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Sensitivity analysis of the weights for RNOV and RSA on CSI300. For the y-axis, IC, RIC, AR, and MDD refer to the axis on the left; ICIR, RICIR, and SR refer to the axis on the right. broad plateau before tapering when novelty dominates. For RSA, improvements are largely mono￾tonic across correlation and portfolio metrics with stable drawdowns. Overall, AlphaSAGE exhibits smooth responses without abrupt pe… view at source ↗
Figure 5
Figure 5. Figure 5: Cumulative return on CSI500 (2022–2024). Comparison among AlphaSAGE (ours), all [PITH_FULL_IMAGE:figures/full_fig_p021_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Cumulative return on S&P500 (2018–2020). Comparison among AlphaSAGE (ours), all [PITH_FULL_IMAGE:figures/full_fig_p021_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Learning dynamics vs. training steps. GFlowNets (GFN) achieve higher plateaus earlier and with less volatility than PPO across IC/RIC and ICIR/RICIR [PITH_FULL_IMAGE:figures/full_fig_p022_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Effect of candidate-pool size. Metrics increase quickly at small–moderate pool sizes and then stabilize, indicating diminishing returns beyond a modest pool. 22 [PITH_FULL_IMAGE:figures/full_fig_p022_8.png] view at source ↗
read the original abstract

The automated mining of predictive signals, or alphas, is a central challenge in quantitative finance. While Reinforcement Learning (RL) has emerged as a promising paradigm for generating formulaic alphas, existing frameworks are fundamentally hampered by a triad of interconnected issues. First, they suffer from reward sparsity, where meaningful feedback is only available upon the completion of a full formula, leading to inefficient and unstable exploration. Second, they rely on semantically inadequate sequential representations of mathematical expressions, failing to capture the structure that determine an alpha's behavior. Third, the standard RL objective of maximizing expected returns inherently drives policies towards a single optimal mode, directly contradicting the practical need for a diverse portfolio of non-correlated alphas. To overcome these challenges, we introduce AlphaSAGE (Structure-Aware Alpha Mining via Generative Flow Networks for Robust Exploration), a novel framework is built upon three cornerstone innovations: (1) a structure-aware encoder based on Relational Graph Convolutional Network (RGCN); (2) a new framework with Generative Flow Networks (GFlowNets); and (3) a dense, multi-faceted reward structure. Empirical results demonstrate that AlphaSAGE outperforms existing baselines in mining a more diverse, novel, and highly predictive portfolio of alphas, thereby proposing a new paradigm for automated alpha mining. Our code is available at https://github.com/BerkinChen/AlphaSAGE.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper introduces AlphaSAGE, a framework for automated alpha mining that combines a Relational Graph Convolutional Network (RGCN) encoder for structure-aware representation of mathematical expressions, Generative Flow Networks (GFlowNets) for exploration, and a dense multi-faceted reward to address reward sparsity, semantically inadequate sequential representations, and mode collapse in prior RL-based alpha mining methods. It claims that the approach yields a more diverse, novel, and predictive portfolio of alphas than existing baselines, supported by empirical results and open-sourced code.

Significance. If the empirical claims hold under rigorous validation, the work could advance automated alpha discovery by offering a mechanism for generating diverse alphas that better support non-correlated portfolios, a practical need in quantitative finance. The integration of GFlowNets to mitigate mode collapse and the structure-aware encoding represent conceptually motivated contributions; the availability of code aids reproducibility.

major comments (2)
  1. [Experimental results / §4] Experimental section (likely §4 or §5): the central claim that AlphaSAGE produces a 'portfolio of non-correlated alphas' relies on structural diversity in expression graphs, but no direct evidence is provided of materially lower average pairwise return correlations or information coefficient correlations on held-out test data compared to RL baselines. Without this measurement, the practical advantage over mode-collapsing methods remains unestablished.
  2. [§3.3] Method section on reward design: the multi-faceted dense reward is described at a high level as providing gradient signal at partial formulas, but the manuscript does not include ablation studies or sensitivity analysis showing that this reward avoids bias toward particular alpha families or that its components are robust across market regimes.
minor comments (2)
  1. [Abstract] Abstract: the sentence introducing AlphaSAGE contains a grammatical issue ('a novel framework is built upon') that should be revised for clarity.
  2. [§3] Notation: ensure consistent use of symbols for GFlowNet flow parameters and RGCN relation types across equations and text.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their detailed and constructive feedback on our manuscript. We have carefully considered the major comments and provide point-by-point responses below. We agree that the suggested additions will enhance the clarity and strength of our claims, and we will incorporate the necessary revisions.

read point-by-point responses

  1. Referee: [Experimental results / §4] Experimental section (likely §4 or §5): the central claim that AlphaSAGE produces a 'portfolio of non-correlated alphas' relies on structural diversity in expression graphs, but no direct evidence is provided of materially lower average pairwise return correlations or information coefficient correlations on held-out test data compared to RL baselines. Without this measurement, the practical advantage over mode-collapsing methods remains unestablished.

    Authors: We concur that providing direct empirical evidence of lower correlations would more robustly support the claim of generating a portfolio of non-correlated alphas. Our current results emphasize structural diversity via the RGCN encoder and overall portfolio performance metrics. To address this gap, we will compute and report the average pairwise return correlations and IC correlations on the held-out test data for AlphaSAGE compared to the RL baselines in the revised experimental section. This addition will better illustrate the practical benefits for quantitative trading applications. revision: yes

  2. Referee: [§3.3] Method section on reward design: the multi-faceted dense reward is described at a high level as providing gradient signal at partial formulas, but the manuscript does not include ablation studies or sensitivity analysis showing that this reward avoids bias toward particular alpha families or that its components are robust across market regimes.

    Authors: Thank you for highlighting this aspect of the reward design. The multi-faceted reward incorporates terms for partial formula evaluation to address sparsity, along with factors for diversity and predictiveness. We will strengthen this section by including ablation studies that isolate the contribution of each reward component and sensitivity analyses evaluating performance across various market regimes, such as different volatility levels and economic conditions. These will be added to the experimental results in the revision. revision: yes

Circularity Check

0 steps flagged

No circularity: framework components independently motivated with external empirical validation

full rationale

The paper introduces AlphaSAGE by combining three separately motivated components (RGCN encoder for structure, GFlowNets for diverse sampling, dense multi-faceted rewards) to address documented RL limitations in alpha mining. No equation or claim reduces a reported performance metric to a fitted parameter or self-referential definition. Diversity and novelty are evaluated via standard out-of-sample metrics on held-out data rather than being defined circularly by the sampling objective itself. Self-citations, if present, are not load-bearing for the central empirical claims. The derivation chain remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The method relies on standard assumptions from graph neural networks and generative flow networks plus domain-specific choices for reward design and graph construction of formulas; no new physical entities are postulated.

free parameters (2)
  • GFlowNet temperature and flow parameters
    Typical hyperparameters of GFlowNets that control sampling diversity and must be tuned on validation data.
  • RGCN layer dimensions and relation types
    Architectural choices for the structure-aware encoder that are selected to fit the alpha-generation task.
axioms (2)
  • domain assumption Mathematical expressions can be faithfully represented as relational graphs whose node and edge types capture operator semantics.
    Invoked when the RGCN encoder is introduced as the solution to sequential-representation inadequacy.
  • ad hoc to paper A dense multi-faceted reward can be defined that provides useful gradient signal at every partial formula without introducing bias toward particular alpha families.
    Central to overcoming reward sparsity; the precise definition is not given in the abstract.

pith-pipeline@v0.9.0 · 5796 in / 1582 out tokens · 50979 ms · 2026-05-21T21:46:51.854234+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

  • IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear
    ?
    unclear

    Relation between the paper passage and the cited Recognition theorem.

    We reformulate alpha discovery as a problem of learning a generative policy Pθ(α) that directly models the distribution of high-quality alphas over the entire space X. The policy is trained such that the probability of sampling any alpha is proportional to a carefully designed reward function R(α)

  • IndisputableMonolith/Foundation/ArithmeticFromLogic.lean embed_injective unclear
    ?
    unclear

    Relation between the paper passage and the cited Recognition theorem.

    A GFlowNet learns a forward policy PF(st+1|st;θ) for constructing objects and a backward policy PB(st|st+1;θ) for deconstruction. The training objective enforces a flow-matching condition

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. From Feedback Loops to Policy Updates: Reinforcement Fine-Tuning for LLM-Based Alpha Factor Discovery

    cs.CE 2026-05 unverdicted novelty 7.0

    QuantEvolver applies reinforcement fine-tuning to evolve an LLM policy for generating executable alpha factor expressions, yielding higher-quality and more complementary factors than prompt-based baselines on market b...

Reference graph

Works this paper leans on

44 extracted references · 44 canonical work pages · cited by 1 Pith paper · 3 internal anchors

  1. [1]

    Flow network based generative models for non-iterative diverse candidate generation

    Emmanuel Bengio, Moksh Jain, Maksym Korablyov, Doina Precup, and Yoshua Bengio. Flow network based generative models for non-iterative diverse candidate generation. Advances in neural information processing systems, 34: 0 27381--27394, 2021

    work page 2021

  2. [2]

    Gflownet foundations

    Yoshua Bengio, Salem Lahlou, Tristan Deleu, Edward J Hu, Mo Tiwari, and Emmanuel Bengio. Gflownet foundations. Journal of Machine Learning Research, 24 0 (210): 0 1--55, 2023

    work page 2023

  3. [3]

    Geometric Deep Learning: Grids, Groups, Graphs, Geodesics, and Gauges

    Michael M Bronstein, Joan Bruna, Taco Cohen, and Petar Veli c kovi \'c . Geometric deep learning: Grids, groups, graphs, geodesics, and gauges. arXiv preprint arXiv:2104.13478, 2021

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  4. [4]

    Chain-of-alpha: Unleashing the power of large language models for alpha mining in quantitative trading

    Lang Cao, Zekun Xi, Long Liao, Ziwei Yang, and Zheng Cao. Chain-of-alpha: Unleashing the power of large language models for alpha mining in quantitative trading. arXiv preprint arXiv:2508.06312, 2025

    work page arXiv 2025

  5. [5]

    Can i trust the explanations? investigating explainable machine learning methods for monotonic models

    Dangxing Chen. Can i trust the explanations? investigating explainable machine learning methods for monotonic models. arXiv preprint arXiv:2309.13246, 2023

    work page arXiv 2023

  6. [6]

    2016, in Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (ACM), 785–794, doi: 10.1145/2939672.2939785

    Tianqi Chen and Carlos Guestrin. XGBoost : A scalable tree boosting system. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , pp.\ 785--794. ACM , 2016. ISBN 978-1-4503-4232-2. doi:10.1145/2939672.2939785. URL https://dl.acm.org/doi/10.1145/2939672.2939785

    work page doi:10.1145/2939672.2939785 2016

  7. [7]

    An empirical study of financial factor mining based on gene expression programming

    Tianxiang Chen, Wei Chen, and Luyao Du. An empirical study of financial factor mining based on gene expression programming. In 2021 4th International Conference on Advanced Electronic Materials, Computers and Software Engineering (AEMCSE), pp.\ 1113--1117. IEEE, 2021

    work page 2021

  8. [8]

    Cluster-gcn: An efficient algorithm for training deep and large graph convolutional networks

    Wei-Lin Chiang, Xuanqing Liu, Si Si, Yang Li, Samy Bengio, and Cho-Jui Hsieh. Cluster-gcn: An efficient algorithm for training deep and large graph convolutional networks. In Proceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining, pp.\ 257--266, 2019

    work page 2019

  9. [9]

    Alphae- volve: A learning framework to discover novel alphas in quantitative investment

    Can Cui, Wei Wang, Meihui Zhang, Gang Chen, Zhaojing Luo, and Beng Chin Ooi. AlphaEvolve : A learning framework to discover novel alphas in quantitative investment. In Proceedings of the 2021 International Conference on Management of Data, pp.\ 2208--2216. ACM , 2021. ISBN 978-1-4503-8343-1. doi:10.1145/3448016.3457324. URL https://dl.acm.org/doi/10.1145/...

    work page doi:10.1145/3448016.3457324 2021

  10. [10]

    Neural message passing for quantum chemistry

    Justin Gilmer, Samuel S Schoenholz, Patrick F Riley, Oriol Vinyals, and George E Dahl. Neural message passing for quantum chemistry. In International conference on machine learning, pp.\ 1263--1272. Pmlr, 2017

    work page 2017

  11. [11]

    Inductive representation learning on large graphs

    Will Hamilton, Zhitao Ying, and Jure Leskovec. Inductive representation learning on large graphs. Advances in neural information processing systems, 30, 2017

    work page 2017

  12. [12]

    Open graph benchmark: Datasets for machine learning on graphs

    Weihua Hu, Matthias Fey, Marinka Zitnik, Yuxiao Dong, Hongyu Ren, Bowen Liu, Michele Catasta, and Jure Leskovec. Open graph benchmark: Datasets for machine learning on graphs. Advances in neural information processing systems, 33: 0 22118--22133, 2020

    work page 2020

  13. [13]

    Biological sequence design with gflownets

    Moksh Jain, Emmanuel Bengio, Alex Hernandez-Garcia, Jarrid Rector-Brooks, Bonaventure FP Dossou, Chanakya Ajit Ekbote, Jie Fu, Tianyu Zhang, Michael Kilgour, Dinghuai Zhang, et al. Biological sequence design with gflownets. In International Conference on Machine Learning, pp.\ 9786--9801. PMLR, 2022

    work page 2022

  14. [14]

    101 Formulaic Alphas

    Zura Kakushadze. 101 formulaic alphas, 2016. URL https://arxiv.org/abs/1601.00991

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  15. [15]

    LightGBM : A highly efficient gradient boosting decision tree

    Guolin Ke, Qi Meng, Thomas Finley, Taifeng Wang, Wei Chen, Weidong Ma, Qiwei Ye, and Tie-Yan Liu. LightGBM : A highly efficient gradient boosting decision tree. In Advances in Neural Information Processing Systems, volume 30. Curran Associates, Inc., 2017. URL https://proceedings.neurips.cc/paper_files/paper/2017/hash/6449f44a102fde848669bdd9eb6b76fa-Abst...

    work page 2017

  16. [16]

    A theory of continuous generative flow networks

    Salem Lahlou, Tristan Deleu, Pablo Lemos, Dinghuai Zhang, Alexandra Volokhova, Alex Hern \'a ndez-Garc a, L \'e na N \'e hale Ezzine, Yoshua Bengio, and Nikolay Malkin. A theory of continuous generative flow networks. In International Conference on Machine Learning, pp.\ 18269--18300. PMLR, 2023

    work page 2023

  17. [17]

    A tutorial on energy-based learning

    Yann LeCun, Sumit Chopra, Raia Hadsell, M Ranzato, Fujie Huang, et al. A tutorial on energy-based learning. Predicting structured data, 1 0 (0), 2006

    work page 2006

  18. [18]

    Distance encoding: Design provably more powerful neural networks for graph representation learning

    Pan Li, Yanbang Wang, Hongwei Wang, and Jure Leskovec. Distance encoding: Design provably more powerful neural networks for graph representation learning. Advances in Neural Information Processing Systems, 33: 0 4465--4478, 2020

    work page 2020

  19. [19]

    Can large language models mine interpretable financial factors more effectively? a neural-symbolic factor mining agent model

    Zhiwei Li, Ran Song, Caihong Sun, Wei Xu, Zhengtao Yu, and Ji-Rong Wen. Can large language models mine interpretable financial factors more effectively? a neural-symbolic factor mining agent model. In Lun-Wei Ku, Andre Martins, and Vivek Srikumar (eds.), Findings of the Association for Computational Linguistics: ACL 2024, pp.\ 3891--3902, Bangkok, Thailan...

    work page doi:10.18653/v1/2024.findings-acl.233 2024

  20. [20]

    Learning gflownets from partial episodes for improved convergence and stability

    Kanika Madan, Jarrid Rector-Brooks, Maksym Korablyov, Emmanuel Bengio, Moksh Jain, Andrei Cristian Nica, Tom Bosc, Yoshua Bengio, and Nikolay Malkin. Learning gflownets from partial episodes for improved convergence and stability. In International Conference on Machine Learning, pp.\ 23467--23483. PMLR, 2023

    work page 2023

  21. [21]

    Trajectory balance: Improved credit assignment in gflownets

    Nikolay Malkin, Moksh Jain, Emmanuel Bengio, Chen Sun, and Yoshua Bengio. Trajectory balance: Improved credit assignment in gflownets. Advances in Neural Information Processing Systems, 35: 0 5955--5967, 2022

    work page 2022

  22. [22]

    Multilayer perceptrons for classification and regression

    Fionn Murtagh. Multilayer perceptrons for classification and regression. Neurocomputing, 2 0 (5-6): 0 183--197, 1991

    work page 1991

  23. [23]

    From Linear to Hierarchical : Evolving Tree -structured Thoughts for Efficient Alpha Mining , August 2025

    Junji Ren, Junjie Zhao, Shengcai Liu, and Peng Yang. From Linear to Hierarchical : Evolving Tree -structured Thoughts for Efficient Alpha Mining , August 2025. URL http://arxiv.org/abs/2508.16334. arXiv:2508.16334 [cs]

    work page arXiv 2025

  24. [24]

    Markov chain monte carlo and variational inference: Bridging the gap

    Tim Salimans, Diederik Kingma, and Max Welling. Markov chain monte carlo and variational inference: Bridging the gap. In International conference on machine learning, pp.\ 1218--1226. PMLR, 2015

    work page 2015

  25. [25]

    The graph neural network model

    Franco Scarselli, Marco Gori, Ah Chung Tsoi, Markus Hagenbuchner, and Gabriele Monfardini. The graph neural network model. IEEE transactions on neural networks, 20 0 (1): 0 61--80, 2008

    work page 2008

  26. [26]

    Modeling relational data with graph convolutional networks

    Michael Schlichtkrull, Thomas N Kipf, Peter Bloem, Rianne Van Den Berg, Ivan Titov, and Max Welling. Modeling relational data with graph convolutional networks. In European semantic web conference, pp.\ 593--607. Springer, 2018

    work page 2018

  27. [27]

    Alphaforge: A framework to mine and dynamically combine formulaic alpha factors

    Hao Shi, Weili Song, Xinting Zhang, Jiahe Shi, Cuicui Luo, Xiang Ao, Hamid Arian, and Luis Angel Seco. Alphaforge: A framework to mine and dynamically combine formulaic alpha factors. In Toby Walsh, Julie Shah, and Zico Kolter (eds.), AAAI-25, Sponsored by the Association for the Advancement of Artificial Intelligence, February 25 - March 4, 2025, Philade...

    work page doi:10.1609/aaai.v39i12.33365 2025

  28. [28]

    Navigating the alpha jungle: An llm-powered mcts framework for formulaic factor mining

    Yu Shi, Yitong Duan, and Jian Li. Navigating the alpha jungle: An llm-powered mcts framework for formulaic factor mining. arXiv preprint arXiv:2505.11122, 2025 b

    work page arXiv 2025

  29. [29]

    Alphaagent: Llm-driven alpha mining with regularized exploration to counteract alpha decay

    Ziyi Tang, Zechuan Chen, Jiarui Yang, Jiayao Mai, Yongsen Zheng, Keze Wang, Jinrui Chen, and Liang Lin. Alphaagent: Llm-driven alpha mining with regularized exploration to counteract alpha decay. In Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V. 2, pp.\ 2813--2822, 2025

    work page 2025

  30. [30]

    Graph Attention Networks

    Petar Veli c kovi \'c , Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Lio, and Yoshua Bengio. Graph attention networks. arXiv preprint arXiv:1710.10903, 2017

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  31. [31]

    \ text\ Alpha \ 2\ : Discovering Logical Formulaic Alphas using Deep Reinforcement Learning , June 2024

    Feng Xu, Yan Yin, Xinyu Zhang, Tianyuan Liu, Shengyi Jiang, and Zongzhang Zhang. \ text\ Alpha \ 2\ : Discovering Logical Formulaic Alphas using Deep Reinforcement Learning , June 2024. URL http://arxiv.org/abs/2406.16505. arXiv:2406.16505 [cs, q-fin]

    work page arXiv 2024

  32. [32]

    How powerful are graph neural networks? In International Conference on Learning Representations

    Keyulu Xu, Weihua Hu, Jure Leskovec, and Stefanie Jegelka. How powerful are graph neural networks? In International Conference on Learning Representations

    work page

  33. [33]

    Qlib: An ai-oriented quantitative investment platform, 2020

    Xiao Yang, Weiqing Liu, Dong Zhou, Jiang Bian, and Tie-Yan Liu. Qlib: An ai-oriented quantitative investment platform, 2020. URL https://arxiv.org/abs/2009.11189

    work page arXiv 2020

  34. [34]

    Graph convolutional networks for text classification

    Liang Yao, Chengsheng Mao, and Yuan Luo. Graph convolutional networks for text classification. In Proceedings of the AAAI conference on artificial intelligence, volume 33, pp.\ 7370--7377, 2019

    work page 2019

  35. [35]

    Generating synergistic formulaic alpha collections via reinforcement learning

    Shuo Yu, Hongyan Xue, Xiang Ao, Feiyang Pan, Jia He, Dandan Tu, and Qing He. Generating synergistic formulaic alpha collections via reinforcement learning. In Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2023. doi:10.1145/3580305.3599831

    work page doi:10.1145/3580305.3599831 2023

  36. [36]

    Let the flows tell: Solving graph combinatorial problems with gflownets

    Dinghuai Zhang, Hanjun Dai, Nikolay Malkin, Aaron C Courville, Yoshua Bengio, and Ling Pan. Let the flows tell: Solving graph combinatorial problems with gflownets. Advances in neural information processing systems, 36: 0 11952--11969, 2023

    work page 2023

  37. [37]

    AutoAlpha : an efficient hierarchical evolutionary algorithm for mining alpha factors in quantitative investment, 2020

    Tianping Zhang, Yuanqi Li, Yifei Jin, and Jian Li. AutoAlpha : an efficient hierarchical evolutionary algorithm for mining alpha factors in quantitative investment, 2020. URL http://arxiv.org/abs/2002.08245

    work page arXiv 2020

  38. [38]

    QuantFactor REINFORCE : Mining Steady Formulaic Alpha Factors with Variance -bounded REINFORCE , October 2024

    Junjie Zhao, Chengxi Zhang, Min Qin, and Peng Yang. QuantFactor REINFORCE : Mining Steady Formulaic Alpha Factors with Variance -bounded REINFORCE , October 2024. URL http://arxiv.org/abs/2409.05144. arXiv:2409.05144

    work page arXiv 2024

  39. [39]

    Learning from expert factors: Trajectory-level reward shaping for formulaic alpha mining, 2025

    Junjie Zhao, Chengxi Zhang, Chenkai Wang, and Peng Yang. Learning from expert factors: Trajectory-level reward shaping for formulaic alpha mining, 2025. URL https://arxiv.org/abs/2507.20263

    work page arXiv 2025

  40. [40]

    Distdgl: Distributed graph neural network training for billion-scale graphs

    Da Zheng, Chao Ma, Minjie Wang, Jinjing Zhou, Qidong Su, Xiang Song, Quan Gan, Zheng Zhang, and George Karypis. Distdgl: Distributed graph neural network training for billion-scale graphs. In 2020 IEEE/ACM 10th Workshop on Irregular Applications: Architectures and Algorithms (IA3), pp.\ 36--44. IEEE, 2020

    work page 2020

  41. [41]

    Alphaqcm: Alpha discovery in finance with distributional reinforcement learning

    Zhoufan Zhu and Ke Zhu. Alphaqcm: Alpha discovery in finance with distributional reinforcement learning. In Forty-second International Conference on Machine Learning, 2025

    work page 2025

  42. [42]

    @esa (Ref

    \@ifxundefined[1] #1\@undefined \@firstoftwo \@secondoftwo \@ifnum[1] #1 \@firstoftwo \@secondoftwo \@ifx[1] #1 \@firstoftwo \@secondoftwo [2] @ #1 \@temptokena #2 #1 @ \@temptokena \@ifclassloaded agu2001 natbib The agu2001 class already includes natbib coding, so you should not add it explicitly Type <Return> for now, but then later remove the command n...

    work page

  43. [43]

    \@lbibitem[] @bibitem@first@sw\@secondoftwo \@lbibitem[#1]#2 \@extra@b@citeb \@ifundefined br@#2\@extra@b@citeb \@namedef br@#2 \@nameuse br@#2\@extra@b@citeb \@ifundefined b@#2\@extra@b@citeb @num @parse #2 @tmp #1 NAT@b@open@#2 NAT@b@shut@#2 \@ifnum @merge>\@ne @bibitem@first@sw \@firstoftwo \@ifundefined NAT@b*@#2 \@firstoftwo @num @NAT@ctr \@secondoft...

    work page

  44. [44]

    @open @close @open @close and [1] URL: #1 \@ifundefined chapter * \@mkboth \@ifxundefined @sectionbib * \@mkboth * \@mkboth\@gobbletwo \@ifclassloaded amsart * \@ifclassloaded amsbook * \@ifxundefined @heading @heading NAT@ctr thebibliography [1] @ \@biblabel @NAT@ctr \@bibsetup #1 @NAT@ctr @ @openbib .11em \@plus.33em \@minus.07em 4000 4000 `\.\@m @bibit...

    work page