极端的图像转换对人类和机器的影响不同
Girik Malik1, Dakarai Crowder2, Ennio Mingolla2
1Northeastern University, Boston, MA, 02115, USA. malik.gi@northeastern.edu.
Biological cybernetics
|June 13, 2023
概括
人工神经网络 (ANN) 与人类不同,与抽象模式作斗争. 这项研究引入了新的图像转换,揭示了ANN在某些任务中表现优于人类,但在其他任务中落后,这表明AI视觉的改进.
科学领域:
- 计算神经科学是一种神经科学.
- 计算机视觉 计算机视觉
- 认知科学 认知科学
背景情况:
- 最近的人工神经网络 (ANN) 在对象识别中模仿灵长类神经和人类性能.
- 网络网络通常依赖于低级特征,使其易受分布外或对抗性输入的影响.
- 人类视觉处理在抽象模式识别方面表现出色,表现出对极端图像扭曲的弹性.
研究的目的:
- 引入由神经生理学发现启发的新型图像转换.
- 用这些转换来评估和比较人类和ANN的对象识别性能.
- 与人类视觉能力相比,识别挑战ANN的特定转换.
主要方法:
- 基于神经生理学原理开发了一套新的图像转换.
- 在使用这些转换的对象识别任务上对人类和ANN性能进行比较评估.
- 精度差异的量化和基于人类数据的变换的难度排名的创建.
主要成果:
- 在某些图像转换方面,ANN超越了人类.
- 人类在其他易于人类感知的转换方面显著优于ANN.
- 对人类视觉处理的转换难度建立了一个明确的排名.
结论:
- 人类视觉处理和当前的ANN在如何处理视觉信息方面存在根本的差异.
- 与人类视觉不同,ANN在概括抽象模式方面表现出局限性.
- 人类视觉处理的洞察力可以指导开发更强大,更适应性的ANN.
相关概念视频
01:26Transformation
36
Microbial communities are dynamic environments where cell lysis releases free DNA into the surroundings. Other cells can take up this extracellular DNA through a process known as transformation.When a cell incorporates this foreign DNA into its genome, resulting in genetic modification, the process is known as transformation. Cells capable of this process are termed competent. Competence can be natural, as observed in certain bacteria and archaea, or artificially induced in the...
36
01:17Properties of the z-Transform I
232
The z-transform is a fundamental tool in digital signal processing, enabling the analysis of discrete-time systems through its various properties. It is an invaluable tool for analyzing discrete-time systems, offering a range of properties that simplify complex signal manipulations. One fundamental property is linearity. For any two discrete-time signals, the z-transform of their linear combination equals the same linear combination of their individual z-transforms. This property is essential...
232