Adaptive DAG-Based BFT Consensus Using Machine Learning

Overview

Today's large-scale distributed systems operate in untrustworthy environments where mutually distrusting entities communicate over unreliable infrastructure. In order to maintain operational guarantees, these systems deploy Byzantine fault-tolerant (BFT) consensus protocols to ensure that non-faulty nodes execute and commit requests from clients in the same order despite the presence of Byzantine, or malicious, nodes. However, these protocols often rely on static configuration parameters set in ideal states which fail to capture the optimal setting under real-world changing network conditions, compromising vital throughput and latency in systems where timely operations are essential. This research introduces a reinforcement learning (RL) system employed to dynamically adjust intra-protocol parameters in real-time to optimize performance. The focus is on applying this adaptive approach to emerging Directed Acyclic Graph (DAG)-based BFT protocols in recent years, which enable parallel transaction processing but require precise parameter tuning due to larger and more complex designs. Using the Autobahn DAG-based BFT consensus protocol as a testbed, this project demonstrates that there is no single optimal configuration that exists across varying network conditions and that the parameter space reaches a size unrealistic for heuristic or rule-based approaches, requiring a learning-based approach to optimization. Through this project, previously unexplored consensus protocol parameter management, adaptivity, and interdependency are established.

Competition history

• AJAS 2026 Category not listed

Related projects

Closest projects by meaning, across every fair and year in the corpus.

Browse more like this

• AJAS None projects
• Every AJAS 2026 project