VisAlign: Dataset for Measuring the Degree of Alignment between AI and Humans in Visual Perception
Not on the map. Why?
Abstract (arXiv)
AI alignment refers to models acting towards human-intended goals, preferences, or ethical principles. Given that most large-scale deep learning models act as black boxes and cannot be manually controlled, analyzing the similarity between models and humans can be a proxy measure for ensuring AI safety. In this paper, we focus on the models' visual perception alignment with humans, further referred to as AI-human visual alignment. Specifically, we propose a new dataset for measuring AI-human visual alignment in terms of image classification, a fundamental task in machine perception. In order to evaluate AI-human visual alignment, a dataset should encompass samples with various scenarios that may arise in the real world and have gold human perception labels. Our dataset consists of three groups of samples, namely Must-Act (i.e., Must-Classify), Must-Abstain, and Uncertain, based on the quantity and clarity of visual information in an image and further divided into eight categories. All samples have a gold human perception label; even Uncertain (severely blurry) sample labels were obtained via crowd-sourcing. The validity of our dataset is verified by sampling theory, statistical theories related to survey design, and experts in the related fields. Using our dataset, we analyze the visual alignment and reliability of five popular visual perception models and seven abstention methods. Our code and data is available at https://github.com/jiyounglee-0523/VisAlign.
Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Builds on 58 of its 100 references that are in the map. First posted on arXiv 3 Aug 2023.
Builds on
58 of 100 references.
References · 58
- 2017Attention Is All You Needcited by 60,361
- 2020An Image is Worth 16x16 Words: Transformers for Image Recognition at Scalecited by 30,186
- 2021High-Resolution Image Synthesis with Latent Diffusion Modelscited by 18,863
- 2021Swin Transformer: Hierarchical Vision Transformer using Shifted Windowscited by 12,139
- 2020A Simple Framework for Contrastive Learning of Visual Representationscited by 12,049
- 2018The Unreasonable Effectiveness of Deep Features as a Perceptual Metriccited by 11,177
- 2014ImageNet Large Scale Visual Recognition Challengecited by 8,498
- 2014Explaining and Harnessing Adversarial Examplescited by 7,364
- 2016Densely Connected Convolutional Networkscited by 6,720
- 2022Hierarchical Text-Conditional Image Generation with CLIP Latentscited by 6,585
- 2022Photorealistic Text-to-Image Diffusion Models with Deep Language Understandingcited by 6,355
- 2016"Why Should I Trust You?": Explaining the Predictions of Any Classifiercited by 5,070
38 more
- 2020Bootstrap your own latent: A new approach to self-supervised Learningcited by 4,750
- 2015Dropout as a Bayesian Approximation: Representing Model Uncertainty in Deep Learningcited by 4,448
- 2021Zero-Shot Text-to-Image Generationcited by 4,288
- 2016Simple and Scalable Predictive Uncertainty Estimation using Deep Ensemblescited by 3,926
- 2023Let's Verify Step by Stepcited by 3,789
- 2022A ConvNet for the 2020scited by 3,579
- 2019Benchmarking Neural Network Robustness to Common Corruptions and Perturbationscited by 2,877
- 2013Describing Textures in the Wildcited by 2,125
- 2020The Many Faces of Robustness: A Critical Analysis of Out-of-Distribution Generalizationcited by 1,850
- 2018A Simple Unified Framework for Detecting Out-of-Distribution Samples and Adversarial Attackscited by 1,485
- 2021MLP-Mixer: An all-MLP Architecture for Visioncited by 1,446
- 2018ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustnesscited by 1,432
- 2019Natural Adversarial Examplescited by 1,408
- 2018Moment Matching for Multi-Source Domain Adaptationcited by 1,331
- 2016Adversarial Machine Learning at Scalecited by 936
- 2014Deep Neural Networks are Easily Fooled: High Confidence Predictions for Unrecognizable Imagescited by 776
- 2021ImageNet-21K Pretraining for the Massescited by 613
- 2022Out-of-Distribution Detection with Deep Nearest Neighborscited by 600
- 2015Towards Open Set Deep Networkscited by 538
- 2019Using Self-Supervised Learning Can Improve Model Robustness and Uncertaintycited by 514
- 2020Generalized ODIN: Detecting Out-of-distribution Image without Learning from Out-of-distribution Datacited by 408
- 2020Noise or Signal: The Role of Image Backgrounds in Object Recognitioncited by 343
- 2022The Effects of Reward Misspecification: Mapping and Mitigating Misaligned Modelscited by 328
- 2022OpenOOD: Benchmarking Generalized Out-of-Distribution Detectioncited by 268
- 2022The Alignment Problem from a Deep Learning Perspectivecited by 258
- 2021Understanding Robustness of Transformers for Image Classificationcited by 249
- 2019Human uncertainty makes classification more robustcited by 235
- 2021Partial success in closing the gap between human and machine visioncited by 185
- 2020A Self-supervised Approach for Adversarial Robustnesscited by 180
- 2019MNIST-C: A Robustness Benchmark for Computer Visioncited by 138
- 2019Fairness in Deep Learning: A Computational Perspectivecited by 123
- 2022Plex: Towards Reliability using Pretrained Large Model Extensionscited by 102
- 2022The Fallacy of AI Functionalitycited by 88
- 2021Consequences of Misaligned AIcited by 80
- 2020Beyond accuracy: quantifying trial-by-trial behaviour of CNNs and humans by measuring error consistencycited by 78
- 2021The Evolution of Out-of-Distribution Robustness Throughout Fine-Tuningcited by 75
- 2022MagicMix: Semantic Mixing with Diffusion Modelscited by 63
- 2017VQS: Linking Segmentations to Questions and Answers for Supervised Attention in VQA and Question-Focused Semantic Segmentationcited by 50
50 most cited of 58 All 58 (JSON)
Built on by
Citing papers per year: 2024: 1.
1 paper.