2021
DOI: 10.1016/j.patter.2021.100348
|View full text |Cite
|
Sign up to set email alerts
|

Grounding deep neural network predictions of human categorization behavior in understandable functional features: The case of face identity

Abstract: Highlights d DNNs modeled how humans rate the similarity of familiar faces to random face stimuli d A generative model controlled the shape and texture features of the face stimuli d The best DNN predicted human behavior because it used similar face-shape features d Explaining human behavior from causal features is difficult with naturalistic images

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
3
2

Citation Types

0
31
0

Year Published

2021
2021
2023
2023

Publication Types

Select...
6
1
1

Relationship

1
7

Authors

Journals

citations
Cited by 23 publications
(31 citation statements)
references
References 97 publications
0
31
0
Order By: Relevance
“…Obviously, it is too simplistic and wrong in the eyes of current-day chemists. In the research of artificial intelligence, it is a widely asked question if AI ‘thinks’ like humans or how to make AI behave like humans [ 33 ]. The observation that ChemTS behaved like a chemist in the 19 th century is an encouraging sign for us.…”
Section: Discussionmentioning
confidence: 99%
“…Obviously, it is too simplistic and wrong in the eyes of current-day chemists. In the research of artificial intelligence, it is a widely asked question if AI ‘thinks’ like humans or how to make AI behave like humans [ 33 ]. The observation that ChemTS behaved like a chemist in the 19 th century is an encouraging sign for us.…”
Section: Discussionmentioning
confidence: 99%
“…Figure S10 suggests a framework for explaining the difference between our results and the results from previous studies that compared DCNNs to brain responses. Because dynamic information plays a major role in the geometry of brain representations (Haxby et al, 2020b;Nastase et al, 2017;Russ and Leopold, 2015), static images could generate higher correlation values between brain responses and DCNNs that do not use motion information (Daube et al, 2021;Grossman et al, 2019;Tsantani et al, 2021). Similarly, studies that used stimuli spanning superordinate categories (e.g., with multiple visual categories (Konkle and Alvarez, 2022;Murty et al, 2021)) would bias representations towards categorical information, reducing the contribution of information that is needed for within-class individuation such as face identification.…”
Section: Discussionmentioning
confidence: 99%
“…The human system for face perception is serving all of these goals during naturalistic viewing, and processes for face identification, besides playing only a small part that is finished quickly at the onset, may also be integrated with other functions in such a way that identification cannot be simply dissociated as a modular process. Perhaps in the future, artificial neural networks trained with more ecological objective functions (Daube et al, 2021;Hasson et al, 2020;Ranjan et al, 2017;Zhuang et al, 2021), requiring not just face recognition, but extending to facial dynamics, attention, memory, social context, and social judgments, will create face representations that afford a more ecologically-valid model that better captures the face processing system in humans. Our findings show that current state-of-the-art DCNNs are early-stage models for the human face processing system, and gaps exist between the current face DCNNs and the goal of developing in silico artificial intelligence models that mimic human intelligence in real-world, naturalistic scenarios.…”
Section: Discussionmentioning
confidence: 99%
“…Other tools such as neural networks have also been considered. Daube et al [45] expounded on how a neural network resonates with the receptor the human body has. These neural connections can then be connected to calculate and recognize patterns which eventually leads to classifying affecting factors.…”
Section: Machine Learning Algorithm and Related Studiesmentioning
confidence: 99%
“…It was deduced that a deep learning neural network may contribute to a higher accuracy rate if the model considered is quite complex. Therefore, an ensemble of ANN and RFC was utilized in this study following several studies [21][22][23][42][43][44][45][46][47][48][49][50].…”
Section: Machine Learning Algorithm and Related Studiesmentioning
confidence: 99%