arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2602.03024cs.LGcs.AI
收藏

一致性深度均衡模型

Consistency Deep Equilibrium Models

School of Electronic Information and Communications, Huazhong University of Science and Technology(华中科技大学电子信息学院);School of Electronic Information(电子信息学院);Communications, Huazhong University of Science(华中科技大学通信学院);School of Science, Wuhan University of Technology(武汉理工大学理学院)

Junchao Lin, Zenan Ling, Jingwen Xu, Robert C. Qiu

AI总结:

提出一致性深度均衡模型(C-DEQ),通过一致性蒸馏将DEQ迭代推理过程视为沿ODE轨迹演化,训练模型将中间状态直接映射到不动点,实现少步推理并保持性能,同时支持多步评估以灵活权衡计算与性能,实验表明在相同少步推理预算下精度提升2-20倍。

摘要:

深度均衡模型(DEQ)已成为深度学习中的一种强大范式,能够以恒定的内存使用量建模无限深度网络。然而,由于不动点求解器的迭代性质,DEQ会带来显著的推理延迟。在这项工作中,我们引入了一致性深度均衡模型(C-DEQ),这是一种利用一致性蒸馏来加速DEQ推理的新框架。我们将DEQ迭代推理过程视为沿固定ODE轨迹向均衡演化。沿着这条轨迹,我们训练C-DEQ将中间状态一致地直接映射到不动点,从而在保持教师DEQ性能的同时实现少步推理。同时,它支持多步评估,以灵活地权衡计算与性能提升。跨多个领域任务的广泛实验表明,在相同的少步推理预算下,C-DEQ相比隐式DEQ实现了2-20倍的精度提升。我们的代码可在https://github.com/landrarwolf/CDEQ获取。

英文摘要:

Deep Equilibrium Models (DEQs) have emerged as a powerful paradigm in deep learning, offering the ability to model infinite-depth networks with constant memory usage. However, DEQs incur significant inference latency due to the iterative nature of fixed-point solvers. In this work, we introduce the Consistency Deep Equilibrium Model (C-DEQ), a novel framework that leverages consistency distillation to accelerate DEQ inference. We cast the DEQ iterative inference process as evolution along a fixed ODE trajectory toward the equilibrium. Along this trajectory, we train C-DEQs to consistently map intermediate states directly to the fixed point, enabling few-step inference while preserving the performance of the teacher DEQ. At the same time, it facilitates multi-step evaluation to flexibly trade computation for performance gains. Extensive experiments across various domain tasks demonstrate that C-DEQs achieve consistent 2-20$\times$ accuracy improvements over implicit DEQs under the same few-step inference budget. Our code is available at https://github.com/landrarwolf/CDEQ.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑