MARL Provenance DAG
Research paper establishing deterministic credit assignment for multi-agent portfolio management in chaotic environments via Directed Acyclic Graphs.
Deterministic Credit Assignment for Multi-Agent Portfolio Management via a Provenance DAG
This project showcases advanced research in Multi-Agent Reinforcement Learning (MARL) applied to chaotic financial environments. The core focus is establishing a foundational framework that solves the credit assignment problem in cooperative multi-agent swarms.
Core Paradigm:
In chaotic market environments, assigning accurate, deterministic rewards or credit to individual reinforcement learning agents is highly complex due to transaction noise, overlapping action spaces, and joint-payoff functions.
To overcome this, this framework maps Directed Acyclic Graphs (DAGs) to track data provenance and explicitly assign deterministic credit back to specific agents.
Key Milestones:
- Academic Track: Fully engineered as part of the academic track 21CSP302L.
- Paper Submission: Submitted to an IEEE conference on April 26, 2026.
- Systems Design: Implemented data tracking mechanisms ensuring transaction provenance logging before execution actions are committed.