Open AccessPosted Content
Which Shortcut Cues Will DNNs Choose? A Study from the Parameter-Space Perspective.
Reads0
Chats0
TLDR
In this article, the authors design a training setup with several shortcut cues, named WCST-ML, where each cue is equally conducive to the visual recognition problem at hand, and observe that certain cues are preferred to others, solutions biased to the easy-to-learn cues tend to converge to relatively flat minima on the loss surface.Abstract:
Deep neural networks (DNNs) often rely on easy-to-learn discriminatory features, or cues, that are not necessarily essential to the problem at hand. For example, ducks in an image may be recognized based on their typical background scenery, such as lakes or streams. This phenomenon, also known as shortcut learning, is emerging as a key limitation of the current generation of machine learning models. In this work, we introduce a set of experiments to deepen our understanding of shortcut learning and its implications. We design a training setup with several shortcut cues, named WCST-ML, where each cue is equally conducive to the visual recognition problem at hand. Even under equal opportunities, we observe that (1) certain cues are preferred to others, (2) solutions biased to the easy-to-learn cues tend to converge to relatively flat minima on the loss surface, and (3) the solutions focusing on those preferred cues are far more abundant in the parameter space. We explain the abundance of certain cues via their Kolmogorov (descriptional) complexity: solutions corresponding to Kolmogorov-simple cues are abundant in the parameter space and are thus preferred by DNNs. Our studies are based on the synthetic dataset DSprites and the face dataset UTKFace. In our WCST-ML, we observe that the inborn bias of models leans toward simple cues, such as color and ethnicity. Our findings emphasize the importance of active human intervention to remove the inborn model biases that may cause negative societal impacts.read more
Citations
More filters
Posted Content
Learning Fair Classifiers with Partially Annotated Group Labels
TL;DR: The authors proposed a Confidence-based Group Label Assignment (CGL) strategy that is readily applicable to any fairness-aware learning method, which utilizes an auxiliary group classifier to assign pseudo group labels where random labels are assigned to low confident samples.
References
More filters
Journal ArticleDOI
Input-output maps are strongly biased towards simple outputs.
TL;DR: A practical bound is provided on the probability that a randomly generated computer program produces a given output of a given complexity and this upper bound is applied to RNA folding and financial trading algorithms.
Proceedings Article
Random deep neural networks are biased towards simple functions
TL;DR: In this paper, it was shown that the binary classifiers of bit strings generated by random wide deep neural networks with ReLU activation function are biased towards simple functions and that the average Hamming distance of the closest input bit string with a different classification is at least sqrt(n / (2π log n), where n is the length of the string.
Proceedings Article
What shapes feature representations? Exploring datasets, architectures, and training
TL;DR: The authors found that when two features redundantly predict the labels, the model preferentially represents one, and its preference reflects what was most linearly decodable from the untrained model.
Proceedings Article
The Pitfalls of Simplicity Bias in Neural Networks
TL;DR: In this paper, the authors attempt to reconcile Simplicity bias and the superior standard generalization of neural networks with the non-robustness observed in practice by designing datasets that incorporate a precise notion of simplicity, comprising multiple predictive features with varying levels of simplicity.
Journal ArticleDOI
Wisconsin Card Sorting Test scores and clinical and sociodemographic correlates in Schizophrenia: multiple logistic regression analysis
Masahiro Banno,Takayoshi Koide,Branko Aleksic,Takashi Okada,Tsutomu Kikuchi,Kunihiro Kohmura,Yasunori Adachi,Naoko Kawano,Tetsuya Iidaka,Norio Ozaki +9 more
TL;DR: Age, education years, PANSS negative scale score and duration of illness affected WCST factor scores in patients with schizophrenia, and using WC ST factor scores may reduce the possibility of type I errors due to multiple comparisons.
Related Papers (5)
Exploring the Role of Stress in Bayesian Word Segmentation using Adaptor Grammars
Relative cue encoding in the context of sophisticated models of categorization: Separating information from categorization.
Keith S. Apfelbaum,Bob McMurray +1 more