AI helps you reading Science

AI generates interpretation videos

AI extracts and analyses the key points of the paper to generate videos automatically


pub
Go Generating

AI Traceability

AI parses the academic lineage of this thesis


Master Reading Tree
Generate MRT

AI Insight

AI extracts a summary of this paper


Weibo:
We show that concept bottleneck models trained in this manner can achieve task accuracies competitive with or even higher than standard models

Concept Bottleneck Models

ICML, pp.5338-5348, (2020)

Cited by: 12|Views243
EI
Full Text
Bibtex
Weibo

Abstract

We seek to learn models that we can interact with using high-level concepts: if the model did not think there was a bone spur in the x-ray, would it still predict severe arthritis? State-of-the-art models today do not typically support the manipulation of concepts like "the existence of bone spurs", as they are trained end-to-end to go ...More

Code:

Data:

0
Introduction
  • Suppose that a radiologist is collaborating with a machine learning model to grade the severity of knee osteoarthritis.
  • State-of-the-art models today do not typically support such queries: they are end-to-end models that go directly from raw input x to target y, and the authors cannot interact with them using the same high-level concepts that practitioners reason with, like “joint space narrowing” or “bone spurs”.
  • They take in an input x, predict concepts c, and use those concepts to predict the target y (Figure 1)
Highlights
  • Suppose that a radiologist is collaborating with a machine learning model to grade the severity of knee osteoarthritis
  • We propose a straightforward method for turning any end-to-end neural network into a concept bottleneck model, given concept annotations at training time: we resize one of the layers to match the number of concepts provided, and add an intermediate loss that encourages the neurons in that layer to align component-wise to the provided concepts
  • We show that concept bottleneck models trained in this manner can achieve task accuracies competitive with or even higher than standard models
  • We investigate if concept bottleneck models can be more robust than standard models to spurious correlations that hold in the training distribution but not the test distribution
  • Concept bottleneck models can compete on task accuracy while supporting intervention and interpretation, allowing practitioners to reason about these models in terms of highlevel concepts they are familiar with, and enabling more effective human-model collaboration through test-time intervention
  • We believe that these models can be promising in settings like medicine, where the high stakes incentivize human experts to collaborate with models at test time, and where the tasks are often normatively defined with respect to a set of standard concepts (e.g., “osteoarthritis is marked by the presence of bone spurs”)
Methods
  • Methods in Natural Language

    Processing (EMNLP), pp. 724–731, 2005.

    Chen, Z., Bei, Y., and Rudin, C.
  • Processing (EMNLP), pp.
  • Z., Bei, Y., and Rudin, C.
  • Concept whitening for interpretable image recognition.
  • ArXiv preprint arXiv:2002.01650, 2020.
  • Bernstein, M.
  • S. Flock: Hybrid CrowdMachine learning classifiers.
  • In Proceedings of the 18th ACM Conference on Computer Supported Cooperative Work & Social Computing, pp.
  • In Proceedings of the 18th ACM Conference on Computer Supported Cooperative Work & Social Computing, pp. 600–611, 2015
Results
  • Because the provided concepts are noisy, the authors denoise them by majority voting, e.g., if more than 50% of crows have black wings in the data, the authors set all crows to have black wings.
Conclusion
  • Concept bottleneck models can compete on task accuracy while supporting intervention and interpretation, allowing practitioners to reason about these models in terms of highlevel concepts they are familiar with, and enabling more effective human-model collaboration through test-time intervention.
  • Similar methods can be used to refine existing concepts and make them more discriminative (Duan et al, 2012)
Summary
  • Introduction:

    Suppose that a radiologist is collaborating with a machine learning model to grade the severity of knee osteoarthritis.
  • State-of-the-art models today do not typically support such queries: they are end-to-end models that go directly from raw input x to target y, and the authors cannot interact with them using the same high-level concepts that practitioners reason with, like “joint space narrowing” or “bone spurs”.
  • They take in an input x, predict concepts c, and use those concepts to predict the target y (Figure 1)
  • Objectives:

    The authors' goal is to characterize concept bottleneck models more fully: Is there a tradeoff between task accuracy and concept interpretability? Do.
  • Methods:

    Methods in Natural Language

    Processing (EMNLP), pp. 724–731, 2005.

    Chen, Z., Bei, Y., and Rudin, C.
  • Processing (EMNLP), pp.
  • Z., Bei, Y., and Rudin, C.
  • Concept whitening for interpretable image recognition.
  • ArXiv preprint arXiv:2002.01650, 2020.
  • Bernstein, M.
  • S. Flock: Hybrid CrowdMachine learning classifiers.
  • In Proceedings of the 18th ACM Conference on Computer Supported Cooperative Work & Social Computing, pp.
  • In Proceedings of the 18th ACM Conference on Computer Supported Cooperative Work & Social Computing, pp. 600–611, 2015
  • Results:

    Because the provided concepts are noisy, the authors denoise them by majority voting, e.g., if more than 50% of crows have black wings in the data, the authors set all crows to have black wings.
  • Conclusion:

    Concept bottleneck models can compete on task accuracy while supporting intervention and interpretation, allowing practitioners to reason about these models in terms of highlevel concepts they are familiar with, and enabling more effective human-model collaboration through test-time intervention.
  • Similar methods can be used to refine existing concepts and make them more discriminative (Duan et al, 2012)
Tables
  • Table1: Task errors with ±2SD over random seeds. Overall, con-
  • Table2: Average concept errors. Bottleneck models have lower cept bottleneck models are competitive with standard models
  • Table3: Task and concept error with background shifts. Bottleneck models have substantially lower task error than the standard model
Download tables as Excel
Related work
  • Concept bottleneck models. Models that bottleneck on human-specified concepts—where the model first predicts the concepts, then uses only those predicted concepts to make a final prediction—have been previously used for specific applications (Kumar et al, 2009; Lampert et al, 2009). Early versions did not use end-to-end neural networks, which soon overtook them in predictive accuracy. Consequently, bottleneck models have historically been more popular for few-shot learning settings, where shared concepts might allow generalization to unseen contexts, rather than the standard supervised setting we consider here.

    More recently, deep neural networks with concept bottlenecks have re-emerged as targeted tools for solving particular tasks, e.g., Fauw et al (2018) for retinal disease diagnosis, Yi et al (2018) for visual question-answering, and Bucher et al (2018) for content-based image retrieval. Losch et al (2019) and Chen et al (2020) also explore learning concept-based models via auxiliary datasets.
Funding
  • PWK was supported by a Facebook PhD Fellowship
  • SM was supported by an NSF Graduate Fellowship
  • EP was supported by a Hertz Fellowship
  • Other funding came from the PECASE Award
Reference
  • Bau, D., Zhou, B., Khosla, A., Oliva, A., and Torralba, A. Network dissection: Quantifying interpretability of deep visual representations. In Computer Vision and Pattern Recognition (CVPR), pp. 6541–6549, 2017.
    Google ScholarLocate open access versionFindings
  • Bengio, Y., Courville, A., and Vincent, P. Representation learning: A review and new perspectives. IEEE Transactions on Pattern Analysis and Machine Intelligence, 35 (8):1798–1828, 2013.
    Google ScholarLocate open access versionFindings
  • Bucher, M., Herbin, S., and Jurie, F. Semantic bottleneck for computer vision tasks. In Asian Conference on Computer Vision, pp. 695–712, 2018.
    Google ScholarLocate open access versionFindings
  • Methods in Natural Language Processing (EMNLP), pp. 724–731, 2005.
    Google ScholarLocate open access versionFindings
  • Chen, Z., Bei, Y., and Rudin, C. Concept whitening for interpretable image recognition. arXiv preprint arXiv:2002.01650, 2020.
    Findings
  • Cheng, J. and Bernstein, M. S. Flock: Hybrid CrowdMachine learning classifiers. In Proceedings of the 18th ACM Conference on Computer Supported Cooperative Work & Social Computing, pp. 600–611, 2015.
    Google ScholarLocate open access versionFindings
  • Clough, J. R., Oksuz, I., Puyol-Anton, E., Ruijsink, B., King, A. P., and Schnabel, J. A. Global and local interpretability for cardiac MRI classification. In International Conference on Medical Image Computing and ComputerAssisted Intervention, pp. 656–664, 2019.
    Google ScholarLocate open access versionFindings
  • Cui, Y., Song, Y., Sun, C., Howard, A., and Belongie, S. Large scale fine-grained categorization and domainspecific transfer learning. In Computer Vision and Pattern Recognition (CVPR), pp. 4109–4118, 2018.
    Google ScholarLocate open access versionFindings
  • Duan, K., Parikh, D., Crandall, D., and Grauman, K. Discovering localized attributes for fine-grained recognition. In Computer Vision and Pattern Recognition (CVPR), pp. 3474–3481, 2012.
    Google ScholarLocate open access versionFindings
  • Fauw, J. D., Ledsam, J. R., Romera-Paredes, B., Nikolov, S., Tomasev, N., Blackwell, S., Askham, H., Glorot, X., O’Donoghue, B., Visentin, D., et al. Clinically applicable deep learning for diagnosis and referral in retinal disease. Nature Medicine, 24(9):1342–1350, 2018.
    Google ScholarLocate open access versionFindings
  • Fong, R. C. and Vedaldi, A. Interpretable explanations of black boxes by meaningful perturbation. In International Conference on Computer Vision (ICCV), pp. 3429–3437, 2017.
    Google ScholarLocate open access versionFindings
  • Goyal, Y., Shalit, U., and Kim, B. Explaining classifiers with causal concept effect (CaCE). arXiv preprint arXiv:1907.07165, 2019.
    Findings
  • Graziani, M., Andrearczyk, V., and Muller, H. Regression concept vectors for bidirectional explanations in histopathology. In Understanding and Interpreting Machine Learning in Medical Image Computing Applications, pp. 124–132, 2018.
    Google ScholarLocate open access versionFindings
  • He, K., Zhang, X., Ren, S., and Sun, J. Deep residual learning for image recognition. In Computer Vision and Pattern Recognition (CVPR), 2016.
    Google ScholarLocate open access versionFindings
  • Huang, S., Xu, Z., Tao, D., and Zhang, Y. Part-stacked CNN for fine-grained visual categorization. In Computer Vision and Pattern Recognition (CVPR), pp. 1173–1182, 2016.
    Google ScholarLocate open access versionFindings
  • Kellgren, J. and Lawrence, J. Radiological assessment of osteo-arthrosis. Annals of the Rheumatic Diseases, 16(4), 1957.
    Google ScholarLocate open access versionFindings
  • Kim, B., Wattenberg, M., Gilmer, J., Cai, C., Wexler, J., Viegas, F., et al. Interpretability beyond feature attribution: Quantitative testing with concept activation vectors (tcav). In International Conference on Machine Learning (ICML), pp. 2668–2677, 2018.
    Google ScholarLocate open access versionFindings
  • Kumar, N., Berg, A. C., Belhumeur, P. N., and Nayar, S. K. Attribute and simile classifiers for face verification. In International Conference on Computer Vision (ICCV), pp. 365–372, 2009.
    Google ScholarLocate open access versionFindings
  • Lampert, C. H., Nickisch, H., and Harmeling, S. Learning to detect unseen object classes by between-class attribute transfer. In Computer Vision and Pattern Recognition (CVPR), pp. 951–958, 2009.
    Google ScholarLocate open access versionFindings
  • Losch, M., Fritz, M., and Schiele, B. Interpretability beyond classification output: Semantic bottleneck networks. arXiv preprint arXiv:1907.10882, 2019.
    Findings
  • Melis, D. A. and Jaakkola, T. Towards robust interpretability with self-explaining neural networks. In Advances in Neural Information Processing Systems (NeurIPS), pp. 7775–7784, 2018.
    Google ScholarLocate open access versionFindings
  • Nevitt, M., Felson, D. T., and Lester, G. The Osteoarthritis Initiative. Cohort study protocol, 2006.
    Google ScholarFindings
  • Pierson, E., Cutler, D., Leskovec, J., Mullainathan, S., and Obermeyer, Z. Using machine learning to understand racial and socioeconomic differences in knee pain. NBER Machine Learning and Healthcare Conference, 2019.
    Google ScholarLocate open access versionFindings
  • Shrikumar, A., Greenside, P., and Kundaje, A. Learning important features through propagating activation differences. In International Conference on Machine Learning (ICML), 2017.
    Google ScholarLocate open access versionFindings
  • Simonyan, K., Vedaldi, A., and Zisserman, A. Deep inside convolutional networks: Visualising image classification models and saliency maps. arXiv preprint arXiv:1312.6034, 2013.
    Findings
  • Sprague, C., Wendoloski, E. B., and Guch, I. Interpretable AI for deep learning-based meteorological applications. In 99th American Meteorological Society Annual Meeting, 2019.
    Google ScholarLocate open access versionFindings
  • Sundararajan, M., Taly, A., and Yan, Q. Axiomatic attribution for deep networks. In International Conference on Machine Learning (ICML), pp. 3319–3328, 2017.
    Google ScholarLocate open access versionFindings
  • Sutton, C. and McCallum, A. Joint parsing and semantic role labeling. In Computational Natural Language Learning (CoNLL), 2005.
    Google ScholarLocate open access versionFindings
  • Szegedy, C., Vanhoucke, V., Ioffe, S., Shlens, J., and Wojna, Z. Rethinking the Inception architecture for computer vision. In Computer Vision and Pattern Recognition (CVPR), pp. 2818–2826, 2016.
    Google ScholarLocate open access versionFindings
  • Tiulpin, A., Thevenot, J., Rahtu, E., Lehenkari, P., and Saarakkala, S. Automatic knee osteoarthritis diagnosis from plain radiographs: A deep learning-based approach. Scientific Reports, 8(1):1–10, 2018.
    Google ScholarLocate open access versionFindings
  • Wah, C., Branson, S., Welinder, P., Perona, P., and Belongie, S. The Caltech-UCSD Birds-200-2011 dataset. Technical report, California Institute of Technology, 2011.
    Google ScholarFindings
  • Yi, K., Wu, J., Gan, C., Torralba, A., Kohli, P., and Tenenbaum, J. Neural-symbolic vqa: Disentangling reasoning from vision and language understanding. In Advances in Neural Information Processing Systems (NeurIPS), pp. 1031–1042, 2018.
    Google ScholarLocate open access versionFindings
  • Zelenko, D., Aone, C., and Richardella, A. Kernel methods for relation extraction. Journal of Machine Learning Research, 3(0):1083–1106, 2003.
    Google ScholarLocate open access versionFindings
  • Zhou, B., Lapedriza, A., Khosla, A., Oliva, A., and Torralba, A. Places: A 10 million image database for scene recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence, 40(6):1452–1464, 2017.
    Google ScholarLocate open access versionFindings
  • Zhou, B., Sun, Y., Bau, D., and Torralba, A. Interpretable basis decomposition for visual explanation. In European Conference on Computer Vision (ECCV), pp. 119–134, 2018.
    Google ScholarLocate open access versionFindings
Your rating :
0

 

Tags
Comments
数据免责声明
页面数据均来自互联网公开来源、合作出版商和通过AI技术自动分析结果,我们不对页面数据的有效性、准确性、正确性、可靠性、完整性和及时性做出任何承诺和保证。若有疑问,可以通过电子邮件方式联系我们:report@aminer.cn
小科