The developmental trajectory of object recognition robustness: children are like small adults but unlike big deep neural networks

Huber, Lukas S.; Geirhos, Robert; Wichmann, Felix A.

doi:10.48550/arxiv.2205.10144

Cited by 16 publications

(3 citation statements)

References 63 publications

(74 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Unlike humans, DNNs trained on natural images relied more on texture (e.g., classifying a catelephant image as an elephant; See Figure 7). Indeed, the CORnet-S model described as one of the best models of human vision largely classifies objects based on texture (Geirhos, Meding, & Wichmann, 2020), and this contrast between DNNs and humans extends to children and adults (Huber, Geirhos, & Wichmann 2022; but see Ritter et al, 2017, for the claim that DNN have a human-like shape-bias). 2022) compared how DNNs and humans learn to classify a set of novel stimuli defined by shape as well as one other non-shape diagnostic feature (including patch location and segment color as shown in Figure 8).…”

Section: Dnns Often Classify Images Based On Texture Rather Than Shapementioning

confidence: 99%

“…7). Indeed, the CORnet-S model described as one of the best models of human vision largely classifies objects based on texture (Geirhos et al, 2020b), and this contrast between DNNs and humans extends to children and adults (Huber, Geirhos, & Wichmann, 2022; but see Ritter, Barrett, Santoro, & Botvinick, 2017, for the claim that DNNs have a human-like shape-bias).…”

Section: Discrepanciesmentioning

confidence: 99%

See 1 more Smart Citation

Deep problems with neural network models of human vision

Bowers¹,

Dujmović²,

Montero³

et al. 2022

Behav Brain Sci

View full text Add to dashboard Cite

Deep neural networks (DNNs) have had extraordinary successes in classifying photographic images of objects and are often described as the best models of biological vision. This conclusion is largely based on three sets of findings: (1) DNNs are more accurate than any other model in classifying images taken from various datasets, (2) DNNs do the best job in predicting the pattern of human errors in classifying objects taken from various behavioral datasets, and (3) DNNs do the best job in predicting brain signals in response to images taken from various brain datasets (e.g., single cell responses or fMRI data). However, these behavioral and brain datasets do not test hypotheses regarding what features are contributing to good predictions and we show that the predictions may be mediated by DNNs that share little overlap with biological vision. More problematically, we show that DNNs account for almost no results from psychological research. This contradicts the common claim that DNNs are good, let alone the best, models of human object recognition. We argue that theorists interested in developing biologically plausible models of human vision need to direct their attention to explaining psychological findings. More generally, theorists need to build models that explain the results of experiments that manipulate independent variables designed to test hypotheses rather than compete on making the best predictions. We conclude by briefly summarizing various promising modelling approaches that focus on psychological data.

show abstract

Section: Dnns Often Classify Images Based On Texture Rather Than Shapementioning

confidence: 99%

Section: Discrepanciesmentioning

confidence: 99%

Deep problems with neural network models of human vision

Bowers¹,

Dujmović²,

Montero³

et al. 2022

Behav Brain Sci

View full text Add to dashboard Cite

show abstract

“…DNNs and humans extends to children and adults(Huber, Geirhos, & Wichmann 2022; but seeRitter et al, 2017, for the claim that DNN have a human-like shape-bias).…”

mentioning

confidence: 99%

Deep Problems with Neural Network Models of Human Vision

Bowers¹,

Dujmović²,

Montero³

et al. 2022

Preprint

View full text Add to dashboard Cite

Deep neural networks (DNNs) have had extraordinary successes in classifying photographic images of objects and are often described as the best models of biological vision. This conclusion is largely based on three sets of findings: (1) DNNs are more accurate than any other model in classifying images taken from various datasets, (2) DNNs do the best job in predicting the pattern of human errors in classifying objects taken from various behavioral benchmark datasets, and (3) DNNs do the best job in predicting brain signals in response to images taken from various brain benchmark datasets (e.g., single cell responses or fMRI data). However, most behavioral and brain benchmarks report the outcomes of observational experiments that do not manipulate any independent variables, and we show that the good prediction on these datasets may be mediated by DNNs that share little overlap with biological vision. More problematically, we show that DNNs account for almost no results from psychological research. This contradicts the common claim that DNNs are good, let alone the best, models of human object recognition. We argue that theorists interested in developing biologically plausible models of human vision need to direct their attention to explaining psychological findings. More generally, theorists need to build models that explain the results of experiments that manipulate independent variables designed to test hypotheses rather than compete on predicting observational data. We conclude by briefly summarizing various promising modelling approaches that focus on psychological data.

show abstract

Safer Than Perception: Increasing Resilience of Automated Vehicles Against Misperception

Fränzle,

Hein

2024

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Autonomous vehicles (AV) rely on environmental perception to take manoeuvre decisions. Safety assurance for AV thus hinges on achieving confidence in all percepts that are safe-guarding critical manoeuvres. As the safety targets for such critical manoeuvres are considerably higher than the statistical figures for the reliability of at least current learning-enabled classification algorithms within the environmental perception, we need means for assuring that the overall system is “safer than perception” in that the frequency of erratically adopting a critical manoeuvre is considerably lower than the frequency of individual misclassifications. We present a methodology for constructively generating reformulations of guard conditions that are more resilient to misperception than the original condition. The synthesized, rephrased guard conditions reconcile a given safety target, i.e. a given a societally accepted upper bound on erratically activating a critical manoeuvre due to a false positive in guard evaluation, with maximal availability, i.e. maximal true positive rate. By synthesizing a resilient rephrasing of the guard condition, the constructive setting presented herein complements the analytical setting from a previous companion paper [6], which merely analysed a given condition for its safety under uncertain perception.

show abstract

The developmental trajectory of object recognition robustness: children are like small adults but unlike big deep neural networks

Cited by 16 publications

References 63 publications

Deep problems with neural network models of human vision

Deep problems with neural network models of human vision

Deep Problems with Neural Network Models of Human Vision

Safer Than Perception: Increasing Resilience of Automated Vehicles Against Misperception

Contact Info

Product

Resources

About