规范深度神经网络作为人类对称处理模型
Yoram S Bonneh1,2, Christopher W Tyler3,4
1School of Optometry and Vision Science, Faculty of Life Science, Bar-Ilan University, Ramat-Gan 5290002, Israel.
iScience
|January 15, 2025
概括
在自然图像上训练的深度神经网络 (DNN) 显示了与人类大脑相当的对称性检测能力. 这些网络在早期层中识别对称性,在类似于人类视觉处理区域的层中达到峰值性能.
科学领域:
- 认知神经科学 认知神经科学
- 计算机视觉 计算机视觉
- 人工智能的人工智能
背景情况:
- 物体对称性是一个普遍的环境特征,但其抽象性质对传统的计算机视觉算法构成了挑战.
- 了解生物和人工系统如何感知对称性对于推进这两个领域至关重要.
研究的目的:
- 调查在自然图像上训练的深度神经网络 (DNN) 是否会发展出类似于人类视觉系统的对称性检测能力.
- 识别DNN中负责处理视觉对称性的特定层.
主要方法:
- 在自然环境图像的数据集上训练了一个DNN.
- 随后,使用具有不同程度对称度 (1,2,4轴) 的无物体随机点图像对训练后的DNN进行了测试.
- 对称性歧视性在DNN的不同层面上进行了分析.
主要成果:
- 在DNN的初始层中,对称性编码是最小的.
- 在FC6层中观察到最显著的对称性歧视,这是一个完全连接的层.
- 这一层的性能特征与人体横向尾综合体 (LOC) 保持一致,LOC是参与视觉处理的大脑区域.
结论:
- 在自然图像上训练的Feedforward DNN表现出一种视觉处理形式,与人类大脑的扩展视觉层次结构相同.
- 这些发现表明,DNN可以作为理解人类视觉感知的有价值模型,特别是在对称性检测领域.
相关概念视频
Symmetry in Maxwell's Equations
2.9K
Once the fields have been calculated using Maxwell's four equations, the Lorentz force equation gives the force that the fields exert on a charged particle moving with a certain velocity. The Lorentz force equation combines the force of the electric field and of the magnetic field on the moving charge. Maxwell's equations and the Lorentz force law together encompass all the laws of electricity and magnetism. The symmetry that Maxwell introduced into his mathematical framework may not be...
2.9K
Neural Circuits
3.0K
Neural circuits and neuronal pools are two of the main structures found in the nervous system. Neural circuits are networks of neurons that work together to carry out a specific task or process. They consist of interconnected neurons and glial cells, which provide structural and metabolic support.
Neuronal pools are collections of nerve cells with similar functions and interact through chemical and electrical signals. These pools include both interneurons (the central neural circuit nodes that...
Neuronal pools are collections of nerve cells with similar functions and interact through chemical and electrical signals. These pools include both interneurons (the central neural circuit nodes that...
3.0K
Symmetry
436
The equation of an ellipse centered at the origin defines all points whose distances from the center maintain a constant ratio between the horizontal and vertical axes. This equation results in a smooth, closed curve that extends further along the x-axis than the y-axis, giving it a horizontal orientation. Such an ellipse demonstrates three kinds of symmetry: across the x-axis, across the y-axis, and about the origin. These symmetries are essential in understanding the graph's structure and...
436


