语义集中用于自我监督的密集表示学习学习语义集中
IEEE transactions on pattern analysis and machine intelligence
|September 23, 2025
概括
这项研究引入了密集的自我监督学习 (SSL) 的明确语义集中,克服了补丁过度分散. 像补丁对应蒸和对象意识过器这样的新方法可以改善密集表示学习.
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:
- 图像级自主监督学习 (SSL) 已经取得了显著的进步,但学习密集补丁表示仍然很困难.
- 现有的方法遭受了补丁过度分散,其中来自同一实例的补丁分散,降低了密集任务的性能.
- 图像级SSL隐含地解决了通过语义集中的过度分散,这与密集的SSL不直接适用.
研究的目的:
- 为密集的自我监督学习 (SSL) 开发明确的语义集中方法.
- 为了应对密集SSL中的空间敏感性和复杂场景的挑战.
- 改进自我监督学习中的补丁密集表示的学习.
主要方法:
- 建议进行补丁对应蒸,以打破严格的空间对齐并学习补丁关系.
- 引入了耐噪排名损失,扩展了强大的伪标签学习的平均精度 (AP) 损失.
- 开发了一个基于对象的过器,使用交叉注意来通过可学习的对象原型来表示补丁,从而实现基于对象的空间映射.
主要成果:
- 提出的方法有效地减轻了密集SSL中的过度分散现象.
- 补丁对应蒸和耐噪音排名损失改善了从杂的伪标签中学习.
- 对象感知过器通过创建基于对象的表示空间,成功地在复杂场景中区分了共享模式.
结论:
- 显式语义集中是密集的自我监督学习的可行和有效方法.
- 开发的技术显著提高了通过SSL学习的密集表示的质量.
- 跨各种任务的经验结果验证了拟议方法在改善下游性能方面的有效性.
相关概念视频
State Space Representation
534
The frequency-domain technique, commonly used in analyzing and designing feedback control systems, is effective for linear, time-invariant systems. However, it falls short when dealing with nonlinear, time-varying, and multiple-input multiple-output systems. The time-domain or state-space approach addresses these limitations by utilizing state variables to construct simultaneous, first-order differential equations, known as state equations, for an nth-order system.
Consider an RLC circuit, a...
Consider an RLC circuit, a...
534
Solution Concentration and Dilution
124.1K
The relative amount of a given solution component is known as its concentration. Often, though not always, a solution contains one component with a concentration that is significantly greater than that of all other components. This component is called the solvent and may be viewed as the medium in which the other components are dispersed or dissolved. Solutions in which water is the solvent are, of course, very common on our planet. A solution in which water is the solvent is called an aqueous...
124.1K
Self-Schemas
35.4K
In general, a schema is a mental construct consisting of a cluster or collection of related concepts (Bartlett, 1932). There are many different types of schemata, and they all have one thing in common: schemata are a method of organizing information that allows the brain to work more efficiently. When a schema is activated, the brain makes immediate assumptions about the person or object being observed.
35.4K
Concentration Cells
25.6K
A concentration cell is a type of a voltaic cell constructed by connecting two almost identical half-cells, both based on the same half-reaction and using the same electrode, differing only in the concentration of one redox species. A concentration cell's potential, therefore, is determined only by the concentration difference of the particular redox species.
Consider the following voltaic cell:
Consider the following voltaic cell:
25.6K
Expressing Solution Concentration
66.0K
A solute is a component of a solution that is typically present at a much lower concentration than the solvent. Solute concentrations are often described with qualitative terms such as dilute (of relatively low concentration) and concentrated (of relatively high concentration).
Concentrations may be quantitatively assessed using a wide variety of measurement units, each convenient for particular applications. Molarity (M) is a useful concentration unit for many applications in chemistry.
Concentrations may be quantitatively assessed using a wide variety of measurement units, each convenient for particular applications. Molarity (M) is a useful concentration unit for many applications in chemistry.
66.0K
Self-Presentation: Self-Monitoring and Self-Handicapping
43.5K
People can go to great lengths to protect their self-image and present themselves in ways that they want others to see them. Sociologist Erving Goffman presented the idea that a person is like an actor on a stage. Calling his theory dramaturgy, Goffman believed that we use “impression management” to present ourselves to others as we hope to be perceived. Each situation is a new scene, and individuals perform different roles depending on who is present (Goffman, 1959). Think about...
43.5K
