IEEE Access
We are thrilled to announce that our paper
“Attractor-Aware Hyperbolic Lipschitz-Constrained Reinforcement Learning for FX Market: LLM-Structured Chain of Thought with Lyapunov–Entropy Dynamics”
has been accepted for publication in IEEE Access Journal (Q1)
Authors: Francesco Rundo, Ph.D., Simone Brancozzi, Massimo Orazio Spata, Ph.D., Michael S. Rundo, and Sebastiano Battiato.
-Innovation & Pipeline: this work introduces a fully explainable, geometry-aware deep learning pipeline for FX market modeling and trading, merging physics of chaos, non-Euclidean learning, and reinforcement control:
-LLM-Structured Chain-of-Thought (s-CoT): A novel reasoning framework where a Large Language Model builds structured, evidence-based s-CoT that connect heterogeneous signals (market data, technical indicators, macro events, CoT reports, Lyapunov and entropy diagnostics) into interpretable causal rules.
These reasoning graphs are then embedded in hyperbolic space, allowing the model to preserve hierarchical dependencies and express multi-scale relationships between causes, effects, and price dynamics, effectively transforming market reasoning into a geometric structure.
-Reinforcement Learning in Hyperbolic Manifolds: The RL policy evolves on curved space, where magnet prices act as dynamic attractors guiding stable, risk-aware actions.
-Lyapunov Exponents & Permutation Entropy: Measure local stability, predictability horizon, and structural disorder, turning chaotic-system diagnostics into quantitative market intelligence.
-Lipschitz-Constrained Learning: Ensures robustness and smooth adaptation under regime shifts and distributional changes, enabling continual learning in volatile financial environments.
The result is an interpretable reasoning-to-action pipeline, capable of producing daily trading decisions with explicit risk budgets, transparent structure, and resilience across market regimes.
Applications:
-Banking Institutions, designing risk-managed structured financial instruments.
-Investment Funds, constructing adaptive, regime-aware portfolio allocation strategies.
-Quantitative Traders & Analysts, understanding markets as nonlinear dynamical systems and enhancing decision timing and risk control.
This research stems from advanced deep learning R&D on non-conventional geometric spaces applied to financial systems, conducted at the Department of Mathematics and Computer Science of the University of Catania (AI4ILF Group@IPLAB).
Paper – Early Access Link:
https://ieeexplore.ieee.org/document/11215735?fbclid=IwY2xjawOBSBBleHRuA2FlbQIxMQBzcnRjBmFwcF9pZBAyMjIwMzkxNzg4MjAwODkyAAEe09EenR2obAGLfEG-0Dls2vDNH9Q-Ax5ceX-RBy7YE5dRBE9HUkuPhuiNTnM_aem_O8BprQV1Y5-hxubt0EmNgA