在抽象层面上对准机器和人类的视觉表示
Lukas Muttenthaler1,2,3,4, Klaus Greff5, Frieda Born6,7,8
1Google DeepMind, Berlin, Germany. lukas.muttenthaler@tu-berlin.de.
Nature
|November 12, 2025
概括
深度神经网络 (DNN) 往往无法像人类一般化,因为它们的表征缺乏等级结构. 这项研究将DNN与人类知识相结合,改善它们与人类认知的协调性,并提高机器学习的性能.
科学领域:
- 人工智能
- 认知科学
- 计算机视觉
背景情况:
- 深度神经网络 (DNN) 越来越多地被用作人类行为和神经表征的模型.
- 然而,DNN培训与人类学习之间存在显著差异,导致模型的概括性较差.
- 目前的视觉模型往往无法捕捉到人类概念知识的层次组织.
研究的目的:
- 识别和解决人类概念知识和DNN表示之间的不一致.
- 开发更符合人类的视觉模型,表现出更好的概括性和稳定性.
- 研究将人类知识注入人工智能 (AI) 系统的方法.
主要方法:
- 培养一个模范教师模仿人类对概念相似性的判断.
- 从教师模型转移到最先进的视觉基础模型.
- 在多个语义抽象层面上使用人类判断对相似性任务的评估模型.
主要成果:
- 人类对齐的模型对人类行为和不确定性进行了更准确的估计.
- 在各种机器学习任务中提高了性能,显示了增强的概括性和分布之外的稳定性.
- 模型成功地捕获了人类认知中存在的等级语义抽象.
结论:
- 将人类的知识整合到DNN中,
- 这种方法导致更强大的,可解释的,与人类结合的AI系统.
- 将人工智能与人类知识相结合, 提供了通往更有能力,更值得信赖的人工智能的途径.
相关概念视频
Depth Perception and Spatial Vision
1.8K
Depth perception is the ability to perceive objects three-dimensionally. It relies on two types of cues: binocular and monocular. Binocular cues depend on the combination of images from both eyes and how the eyes work together. Since the eyes are in slightly different positions, each eye captures a slightly different image. This disparity between images, known as binocular disparity, helps the brain interpret depth. When the brain compares these images, it determines the distance to an object.
1.8K
High-Level and Low-Level Awareness
618
Controlled processes in human consciousness represent high-alert mental states where individuals deliberately focus their attention on achieving specific goals. Controlled processes can be seen in situations like mastering new technology, where a person might become so absorbed that they ignore surrounding distractions. Such processes involve selective attention, requiring one to concentrate on particular elements of experience while disregarding others. These are governed by executive...
618
Machines: Problem Solving II
632
Machines are complex structures consisting of movable, pin-connected multi-force members that work together to transmit forces. Consider a lifting tong carrying a 100 kg load. It comprises movable sections DAF and CBG linked together with member AB.
632
Machines: Problem Solving I
670
A toggle clamp is a mechanical device commonly used for holding and clamping objects in various applications, such as woodworking, metalworking, and assembly operations. Consider a toggle clamp subjected to a force of 200 N at the handle. The vertical clamping force can be calculated, provided the dimensions of the toggle clamp are known.
The toggle clamp system is a machine structure consisting of movable, pin-connected multi-force members that form a stabilized system to transmit forces. The...
The toggle clamp system is a machine structure consisting of movable, pin-connected multi-force members that form a stabilized system to transmit forces. The...
670
Modeling and Similitude
598
Scaled modeling is a fundamental technique in engineering, enabling the study of large and complex systems by creating smaller, manageable replicas that recreate critical characteristics of the original. In hydrology and civil infrastructure, for example, scaled models of dams help analyze water flow, turbulence, and pressure. This method allows for accurate predictions of real-world behavior within a controlled environment, significantly reducing the cost and time involved in full-scale...
598
Gestalt Principles of Perception
1.0K
Gestalt principles provide a framework for understanding how humans perceive objects as unified wholes within their context. These principles are essential in explaining the cognitive processes that make sense of complex visual stimuli by organizing them into coherent groups. One fundamental principle is proximity, which posits that objects located close to each other are perceived as a collective group. For instance, when dots are positioned near one another, the visual system interprets them...
1.0K


