[2603.25813] MAGNET: Autonomous Expert Model Generation via Decentralized Autoresearch and BitNet Training

Computer Science > Machine Learning arXiv:2603.25813 (cs) [Submitted on 26 Mar 2026] Title:MAGNET: Autonomous Expert Model Generation via Decentralized Autoresearch and BitNet Training Authors:Yongwan Kim, Sungchul Park View a PDF of the paper titled MAGNET: Autonomous Expert Model Generation via Decentralized Autoresearch and BitNet Training, by Yongwan Kim and 1 other authors View PDF HTML (experimental) Abstract:We present MAGNET (Model Autonomously Growing Network), a decentralized system for autonomous generation, training, and serving of domain-expert language models across commodity hardware. MAGNET integrates four components: (1) autoresearch, an autonomous ML research pipeline that automates dataset generation, hyperparameter exploration, evaluation, and error-driven iteration; (2) BitNet b1.58 ternary training, enabling CPU-native inference via this http URL without GPU hardware; (3) DiLoCo-based distributed merging for communication-efficient aggregation of domain specialists; and (4) on-chain contribution tracking on the HOOTi EVM chain. We validate autoresearch through three case studies: video safety classification (balanced accuracy 0.9287 to 0.9851), cryptocurrency directional prediction (41% to 54.9% hit rate), and BitNet hyperparameter optimization (10-phase sweep, -16.7% validation loss). Comments: Subjects: Machine Learning (cs.LG); Artificial Intelligence (cs.AI) Cite as: arXiv:2603.25813 [cs.LG]   (or arXiv:2603.25813v1 [cs.LG] for this version)   htt...