Ghost Attractor Networks offer efficient sequential generation with stable latent structures

Researchers have introduced Ghost Attractor Networks (GANs), a novel dynamical decoder designed to improve sequential generation efficiency and control in large-scale models. GANs utilize a learned potential with a basin-attractor structure to enable closed-loop control, such as phase-conditioned action generation and cross-step latent carry-over. Empirically, a GAN model demonstrated a significant reduction in parameters and latency compared to a Diffusion Transformer while achieving comparable or superior accuracy on robotic action decoding tasks and closed-loop benchmarks. AI

IMPACT Introduces a more efficient and controllable method for sequential generation, potentially impacting robotics and other generative AI applications.

RANK_REASON The cluster contains a research paper detailing a new model architecture. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• CVAE
• diffusion
• Diffusion Transformer
• Ghost Attractor Networks
• LIBERO-10
• multilayer perceptron
• Neural ODE
• transformer

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →