Brain hierarchy score: Which deep neural networks are hierarchically brain-like?
Brain hierarchy score: Which deep neural networks are hierarchically brain-like?
复制标题
DOI：
10.1016/j.isci.2021.103013
复制
发表时间：
2021-09-24
期刊：
影响因子：
5.8
通讯作者：
Kamitani Y
中科院分区：
文献类型：
--
作者：
Nonaka S;Majima K;Aoki SC;Kamitani Y
通过深度神经网络（DNN）实现人类水平的图像识别，激发了人们对DNN是否以及如何与大脑相似的兴趣。DNN和视觉皮层都执行分层处理，并且在表示视觉特征时，分层视觉区域和DNN层之间已经显示出对应关系。在这里，我们提出了大脑层次（BH）分数作为一个度量，以量化基于神经解码和编码分析的层次对应程度，其中DNN单元激活和人脑活动是相互预测的。我们发现，29个具有各种架构的预训练DNN的BH分数与图像识别性能呈负相关，因此表明最近开发的高性能DNN不一定是大脑。DNN模型的实验操作表明，具有广泛空间整合的单路径顺序前馈架构对类脑层次结构至关重要。我们的方法可以提供新的方法来设计DNN在其代表性的同源性的大脑。提出了一种类脑层次结构的度量方法，用于描述DNN的特征。利用人类功能磁共振成像（fMRI）编码/解码量化了层次结构的对应关系。在典型的DNN模型中，高性能模型不是类脑模型。
Achievement of human-level image recognition by deep neural networks (DNNs) has spurred interest in whether and how DNNs are brain-like. Both DNNs and the visual cortex perform hierarchical processing, and correspondence has been shown between hierarchical visual areas and DNN layers in representing visual features. Here, we propose the brain hierarchy (BH) score as a metric to quantify the degree of hierarchical correspondence based on neural decoding and encoding analyses where DNN unit activations and human brain activity are predicted from each other. We find that BH scores for 29 pre-trained DNNs with various architectures are negatively correlated with image recognition performance, thus indicating that recently developed high-performance DNNs are not necessarily brain-like. Experimental manipulations of DNN models suggest that single-path sequential feedforward architecture with broad spatial integration is critical to brain-like hierarchy. Our method may provide new ways to design DNNs in light of their representational homology to the brain. A measure for brain-like hierarchy is proposed to characterize DNNs Encoding/decoding with human fMRI quantifies the hierarchical correspondence Among representative DNN models, high-performance models are not brain-like Critical factors for brain-like hierarchy are explored Neuroscience; Neural networks; Human-centered computing
登录
查看更多内容
影响因子：
64.8
作者：
Epstein, R;Kanwisher, N
通讯作者：
Kanwisher, N
影响因子：
3.6
作者：
Carlson, Thomas;Hogendoorn, Hinze;Verstraten, Frans A. J.
通讯作者：
Verstraten, Frans A. J.
影响因子：
5.7
作者：
Eickenberg, Michael;Gramfort, Alexandre;Thirion, Bertrand
通讯作者：
Thirion, Bertrand
影响因子：
56.9
作者：
Bashivan, Pouya;Kar, Kohitij;DiCarlo, James J.
通讯作者：
DiCarlo, James J.
影响因子：
3.7
作者：
Felleman, Daniel J.;Van Essen, David C.
通讯作者：
Van Essen, David C.