Tag
The Controlled Dynamics Attractor Transformer (CDAT) combines a mixture von Mises-Fisher attention energy with a Hopfield refinement energy and CANN-inspired excitation-inhibition modulation, providing topology-constrained dynamical systems for stable inference. It achieves state-of-the-art performance on graph anomaly detection and classification benchmarks.
This paper introduces HoReN, a parameter-preserving model editing method that uses normalized Hopfield retrieval to handle large-scale sequential updates to large language models. It addresses issues of knowledge accumulation and routing challenges, demonstrating stable performance on 50K sequential edits where prior methods degrade.