轻量级可修复的完整神经网络用于移动应用程序
IEEE transactions on neural networks and learning systems
|October 27, 2025
概括
我们为移动设备引入了轻量级可重定位的集成神经网络 (RINNs). 这些RINN有效地处理连续集成层,在ImageNet.Net上实现高精度与低延迟.
科学领域:
- 计算机科学 计算机科学
- 人工智能的人工智能
- 机器学习 机器学习
背景情况:
- 标准集成神经网络 (INN) 由于其复杂的结构,在移动设备上面临部署挑战.
- 连续集成层中的重组参数化问题阻碍了有效的推理.
研究的目的:
- 为了解决INN在资源有限的移动设备上部署的限制.
- 开发一种新型的轻量级和高效的神经网络.
主要方法:
- 提出了一种持续的重构策略,将培训时间整合层转换为推断时间前结构.
- 扩展了MetaFormer (类似Vision Transformer) 的架构,用于连续集成层.
- 引入了一个过度参数化的整体分支,以增强表示能力.
主要成果:
- 开发了轻量级可重定位INN (RINNs),在移动设备上具有强大的性能.
- 在ImageNet.Net上实现了超过79.1%的top-1精度,延迟时间为0.87ms.
- 与离散模型相比,在结构修剪方面表现出优越的强度.
结论:
- 对于在移动平台上部署先进的神经网络,RINNs提供了一个有前途的解决方案.
- 提出的方法使得在资源有限的环境中能够高效,准确地进行深度学习推断.
相关概念视频
Neural Regulation
43.1K
Digestion begins with a cephalic phase that prepares the digestive system to receive food. When our brain processes visual or olfactory information about food, it triggers impulses in the cranial nerves innervating the salivary glands and stomach to prepare for food.
43.1K
PI Controller: Design
1.2K
Proportional Integral (PI) controllers are a fundamental component in modern control systems, widely used to enhance performance and mitigate steady-state errors. They are particularly effective in applications such as automatic brightness adjustment on smartphones, where they excel at mitigating steady-state errors for step-function inputs. Unlike PD controllers, which require time-varying errors to function optimally, PI controllers leverage their integral component to address residual...
1.2K
Approximate Integration
4
In many practical and theoretical contexts, the exact value of a definite integral may be inaccessible. This limitation typically arises when the antiderivative of a function is either unknown or cannot be expressed in a closed mathematical form. Alternatively, it can occur when a function is defined not by a formula but by a finite set of empirical data points, such as those collected during experiments. In these cases, approximate integration techniques provide a valuable solution.One of the...
4
Applications of Integration to Probability Density Functions
4
Continuous probability distributions are used to model random variables that can take on any real value within a specified range. These variables do not take on isolated or countable values but rather exist on a continuum. For example, the height of an individual can be measured with increasing precision—such as 163.5 or 165.25 centimeters—demonstrating that height is a continuous random variable.The behavior of such variables is described using a probability density function (PDF),...
4


