深度学习中不确定性的研究及应用

        随着深度学习的发展与普及，深度学习模型的可靠性变得极其重要。通过对深度学习中的不确定性进行建模可以使模型在输出预测结果的同时给出相应的不确定性信息，从而避免因为错误的输出而造成严重的损失。目前，不确定性建模的方法主要有贝叶斯神经网络、蒙特卡洛随机失活方法（Monte Carlo Dropout, MC Dropout）、深度集成方法（Deep Ensemble）、证据神经网络（Evidence Neural Networks）和单向传播法（Single Forward Pass Uncertainty Methods）等五种。然而并没有通用的评价指标用以对这些方法进行评估。随着不确定性建模方法的发展，不确定性在医学图像分割中也得到了广泛的关注。在医学图像分割中常常会存在显著的数据不确定性。针对同一张医学图像，不同的医生往往会给出不同的标注结果，从而引入数据不确定性；但是现有的深度学习模型学习的映射关系均是一对一的，并没有考虑数据不确定性的问题。针对这两个挑战，本文展开以下研究:
        针对不确定性建模方法缺乏评价指标的问题，本文给出了两个基本假设：模型分类正确的样本的预测不确定性低于模型分类错误的样本的预测不确定性；模型对分布内（In Distribution, ID）数据的预测不确定性低于对分布外（Out of Distribution, OoD）数据的预测不确定性。基于这两条假设本文给出了两个基于一致性指数的评价标准，并在MNIST、CIFAR-10 和CIFAR-100 三个数据集上进行了实验验证。实验结果证实了给出的评价标准的有效性，并且进一步证实了深度集成方法的优异性能；实验结果还表明常用的不确定性建模方法在对OoD 数据给出较高不确定性的同时会增大在ID 数据集上的不确定性。
        针对医学图像分割中的多标注问题，本文给出了一个由共享编码器和多个独立解码器组成的网络结构，并将注意力机制引入该结构从而给出了OEMD-AG UNet模型。在OEMD-AG U-Net 模型的基础上，本文通过对每个解码器进行数据不确定性建模给出了SOEMD-AG U-Net 模型。在数据集QUBIQ 2020 和LIDC-IDRI 上进行的实验表明了这两个模型可以清楚地捕获到不同标注之间的一致性和差异性。

[1] KRIZHEVSKY A, SUTSKEVER I, HINTON G E. Imagenet Classification with Deep Convolutional Neural Networks[J]. Advances in Neural Information Processing Systems, 2012, 25.
[2] HINTON G, DENG L, YU D, et al. Deep Neural Networks for Acoustic Modeling in Speech Recognition: The Shared Views of Four Research Groups[J]. IEEE Signal processing magazine, 2012, 29(6): 82-97.
[3] MIKOLOV T, CHEN K, CORRADO G, et al. Efficient Estimation of Word Representations in Vector Space[C]//International Conference on Learning Representations. 2013.
[4] JUMPER J, EVANS R, PRITZEL A, et al. Highly Accurate Protein Structure Prediction with AlphaFold[J]. Nature, 2021, 596(7873): 583-589.
[5] DAVIES A, VELIČKOVIĆ P, BUESING L, et al. Advancing Mathematics by Guiding Human Intuition with AI[J]. Nature, 2021, 600(7887): 70-74.
[6] DEGRAVE J, FELICI F, BUCHLI J, et al. Magnetic Control of Tokamak Plasmas through Deep Reinforcement Learning[J]. Nature, 2022, 602(7897): 414-419.
[7] GAL Y, GHAHRAMANI Z. Dropout as a Bayesian Approximation: Representing Model Uncertainty in Deep Learning[C]//International Conference on Machine Learning. 2016: 1050-1059.
[8] HABIB K. ODI Resume Report on Investigation PE 16-007 Concerning Tesla Automatic Vehicle Control Systems[M]//Technical Report. NHTSA Office of Defects Investigation, 2016.
[9] LIU X, LI Y, WU C, et al. Adv-BNN: Improved Adversarial Defense through Robust Bayesian Neural Network[C]//International Conference on Learning Representations. 2018.
[10] HOULSBY N, HUSZÁR F, GHAHRAMANI Z, et al. Bayesian Active Learning for Classification and Preference Learning[J]. Stat, 2011, 1050: 24.
[11] KENDALL A, GAL Y. What Uncertainties do We Need in Bayesian Deep Learning for Computer Vision?[C]//Advances in Neural Information Processing Systems. 2017: 5574-5584.
[12] GAL Y. Uncertainty in Deep Learning[J]. University of Cambridge, 2016, 1(3).
[13] MACKAY D J. A Practical Bayesian Framework for Backpropagation Networks[J]. Neural Computation, 1992, 4(3): 448-472.
[14] NEAL R M. Bayesian Learning for Neural Networks: volume 118[M]. Springer Science & Business Media, 2012.
[15] WELLING M, TEH Y W. Bayesian Learning via Stochastic Gradient Langevin Dynamics[C]//International Conference on Machine Learning. 2011: 681-688.
[16] CHEN T, FOX E, GUESTRIN C. Stochastic Gradient Hamiltonian Monte Carlo[C]//International Conference on Machine Learning. 2014: 1683-1691.
[17] YAO J, PAN W, GHOSH S, et al. Quality of Uncertainty Quantification for Bayesian Neural Network Inference[A]. 2019. arXiv: 1906.09686.
[18] GRAVES A. Practical Variational Inference for Neural Networks[C]//Advances in Neural Information Processing Systems. 2011: 2348-2356.
[19] BLUNDELL C, CORNEBISE J, KAVUKCUOGLU K, et al. Weight Uncertainty in Neural Network[C]//International Conference on Machine Learning. 2015: 1613-1622.
[20] LOUIZOS C, WELLING M. Structured and Efficient Variational Deep Learning with Matrix Gaussian Posteriors[C]//International Conference on Machine Learning. 2016: 1708-1716.
[21] LOUIZOS C, WELLING M. Multiplicative Normalizing Flows for Variational Bayesian Neural Networks[C]//International Conference on Machine Learning. 2017: 2218-2227.
[22] PAWLOWSKI N, BROCK A, LEE M C, et al. Implicit Weight Uncertainty in Neural Networks[A]. 2017. arXiv: 1711.01297.
[23] HERNÁNDEZ-LOBATO J M, ADAMS R. Probabilistic Backpropagation for Scalable Learning of Bayesian Neural Networks[C]//International Conference on Machine Learning. 2015:1861-1869.
[24] HERNANDEZ-LOBATO J, LI Y, ROWLAND M, et al. Black-Box Alpha Divergence Minimization[C]//International Conference on Machine Learning. 2016: 1511-1520.
[25] KINGMA D P, SALIMANS T, WELLING M. Variational Dropout and the Local Reparameterization Trick[J]. Advances in Neural Information Processing Systems, 2015, 28.
[26] WILLIAMS C K. Computing With Infinite Networks[C]//Advances in Neural Information Processing Systems. 1997: 295-301.
[27] LAKSHMINARAYANAN B, PRITZEL A, BLUNDELL C. Simple and Scalable Predictive Uncertainty Estimation Using Deep Ensembles[C]//Advances in Neural Information Processing Systems. 2017: 6402-6413.
[28] TAGASOVSKA N, LOPEZ-PAZ D. Frequentist Uncertainty Estimates for Deep Learning[A]. 2018. arXiv: 1811.00908.
[29] WENZEL F, SNOEK J, TRAN D, et al. Hyperparameter Ensembles for Robustness and Uncertainty Quantification[A]. 2020. arXiv: 2006.13570.
[30] HAVASI M, JENATTON R, FORT S, et al. Training Independent Subnetworks for Robust Prediction[C]//International Conference on Learning Representations. 2021.
[31] AMINI A, SCHWARTING W, SOLEIMANY A, et al. Deep Evidential Regression[J]. Advances in Neural Information Processing Systems, 2020, 33.
[32] SENSOY M, KAPLAN L, KANDEMIR M. Evidential Deep Learning to Quantify Classification Uncertainty[C]//Advances in Neural Information Processing Systems: volume 31. Curran Associates, Inc., 2018.
[33] VAN AMERSFOORT J, SMITH L, TEH Y W, et al. Uncertainty Estimation Using a Single Deep Deterministic Neural Network[C]//International Conference on Machine Learning. PMLR, 2020: 9690-9700.
[34] LIU J, LIN Z, PADHY S, et al. Simple and Principled Uncertainty Estimation with Deterministic Deep Learning via Distance Awareness[J]. Advances in Neural Information Processing Systems, 2020, 33: 7498-7512.
[35] VAN AMERSFOORT J, SMITH L, JESSON A, et al. On Feature Collapse and Deep Kernel Learning for Single Forward Pass Uncertainty[A]. 2021. arXiv: 2102.11409.
[36] MUKHOTI J, VAN AMERSFOORT J, TORR P H, et al. Deep Deterministic Uncertainty for Semantic Segmentation[A]. 2021. arXiv: 2111.00079.
[37] LITJENS G, KOOI T, BEJNORDI B E, et al. A Survey on Deep Learning in Medical Image Analysis[J]. Medical Image Analysis, 2017, 42: 60-88.
[38] NAIR T, PRECUP D, ARNOLD D L, et al. Exploring Uncertainty Measures in Deep Networks for Multiple Sclerosis Lesion Detection and Segmentation[C]//International Conference on Medical Image Computing and Computer-Assisted Intervention. Springer, 2018: 655-663.
[39] GHESU F C, GEORGESCU B, GIBSON E, et al. Quantifying and Leveraging Classification Uncertainty for Chest Radiograph Assessment[C]//International Conference on Medical Image Computing and Computer-Assisted Intervention. Springer, 2019: 676-684.
[40] KARIMI D, ZENG Q, MATHUR P, et al. Accurate and Robust Deep Learning-Based Segmentation of the Prostate Clinical Target Volume in Ultrasound Images[J]. Medical Image Analysis, 2019, 57: 186-196.
[41] KWON Y, WON J H, KIM B J, et al. Uncertainty Quantification Using Bayesian Neural Networks in Classification: Application to Biomedical Image Segmentation[J]. Computational Statistics & Data Analysis, 2020, 142: 106816.
[42] JOSKOWICZ L, COHEN D, CAPLAN N, et al. Inter-Observer Variability of Manual Contour Delineation of Structures in CT[J]. European Radiology, 2019, 29(3): 1391-1399.
[43] BECKER A S, CHAITANYA K, SCHAWKAT K, et al. Variability of Manual Segmentation of the Prostate in Axial T2-Weighted MRI: A Multi-Reader Study[J]. European Journal of Radiology, 2019, 121: 108716.
[44] WARFIELD S K, ZOU K H, WELLS W M. Simultaneous Truth and Performance Level Estimation (STAPLE): an Algorithm for the Validation of Image Segmentation[J]. IEEE Transactions on Medical Imaging, 2004, 23(7): 903-921.
[45] ASMAN A J, LANDMAN B A. Formulating Spatially Varying Performance in the Statistical Fusion Framework[J]. IEEE Transactions on Medical Imaging, 2012, 31(6): 1326-1336.
[46] CARDOSO M J, LEUNG K, MODAT M, et al. STEPS: Similarity and Truth Estimation for Propagated Segmentations and Its Application to Hippocampal Segmentation and Brain Parcelation[J]. Medical Image Analysis, 2013, 17(6): 671-684.
[47] AKHONDI-ASL A, HOYTE L, LOCKHART M E, et al. A Logarithmic Opinion Pool Based STAPLE Algorithm for the Fusion of Segmentations with Associated Reliability Weights[J]. IEEE Transactions on Medical Imaging, 2014, 33(10): 1997-2009.
[48] JOSKOWICZ L, COHEN D, CAPLAN N, et al. Automatic Segmentation Variability Estimation with Segmentation Priors[J]. Medical Image Analysis, 2018, 50: 54-64.
[49] KOHL S, ROMERA-PAREDES B, MEYER C, et al. A Probabilistic U-Net for Segmentation of Ambiguous Images[C]//Advances in Neural Information Processing Systems. 2018: 6965-6975.
[50] BAUMGARTNER C F, TEZCAN K C, CHAITANYA K, et al. Phiseg: Capturing Uncertainty in Medical Image Segmentation[C]//International Conference on Medical Image Computing and Computer-Assisted Intervention. Springer, 2019: 119-127.
[51] TANNO R, SAEEDI A, SANKARANARAYANAN S, et al. Learning From Noisy Labels by Regularized Estimation of Annotator Confusion[C]//IEEE Conference on Computer Vision and Pattern Recognition. 2019: 11244-11253.
[52] ZHANG L, TANNO R, XU M C, et al. Disentangling human error from ground truth in segmentation of medical images[J]. Advances in Neural Information Processing Systems, 2020, 33: 15750-15762.
[53] MONTEIRO M, LE FOLGOC L, COELHO DE CASTRO D, et al. Stochastic Segmentation Networks: Modelling Spatially Correlated Aleatoric Uncertainty[C]//Advances in Neural Information Processing Systems: volume 33. Curran Associates, Inc., 2020: 12756-12767.
[54] HU S, WORRALL D, KNEGT S, et al. Supervised Uncertainty Quantification for Segmentation with Multiple Annotations[C]//International Conference on Medical Image Computing and Computer-Assisted Intervention. Springer, 2019: 137-145.
[55] ROSENBLATT F. Principles of Neurodynamics. Perceptrons and the Theory of Brain Mechanisms[R]. Cornell Aeronautical Lab Inc Buffalo NY, 1961.
[56] LECUN Y, BOTTOU L, BENGIO Y, et al. Gradient-Based Learning Applied to Document Recognition[J]. Proceedings of the IEEE, 1998, 86(11): 2278-2324.
[57] VASWANI A, SHAZEER N, PARMAR N, et al. Attention is All You Need[J]. Advances in Neural Information Processing Systems, 2017, 30.
[58] KINGMA D P, WELLING M. Auto-Encoding Variational Bayes[C]//International Conference on Learning Representations. 2013.
[59] SRIVASTAVA N, HINTON G, KRIZHEVSKY A, et al. Dropout: A Simple Way to Prevent Neural Networks from Overfitting[J]. The Journal of Machine Learning Research, 2014, 15(1): 1929-1958.
[60] GULRAJANI I, AHMED F, ARJOVSKY M, et al. Improved Training of Wasserstein GANs[J]. Advances in Neural Information Processing Systems, 2017, 30.
[61] MIYATO T, KATAOKA T, KOYAMA M, et al. Spectral Normalization for Generative Adversarial Networks[C]//International Conference on Learning Representations. 2018.
[62] GOUK H, FRANK E, PFAHRINGER B, et al. Regularisation of Neural Networks by Enforcing Lipschitz Continuity[J]. Machine Learning, 2021, 110(2): 393-416.
[63] HE K, ZHANG X, REN S, et al. Deep Residual Learning for Image Recognition[C]//IEEE Conference on Computer Vision and Pattern Recognition. 2016: 770-778.
[64] JACOBSEN J H, SMEULDERS A, OYALLON E. I-RevNet: Deep Invertible Networks[C]//International Conference on Learning Representations. 2018.
[65] BEHRMANN J, GRATHWOHL W, CHEN R T, et al. Invertible Residual Networks[C]//International Conference on Machine Learning. PMLR, 2019: 573-582.
[66] TITSIAS M. Variational Learning of Inducing Variables in Sparse Gaussian Processes[C]//Artificial Intelligence and Statistics. PMLR, 2009: 567-574.
[67] FRIEDMAN J, HASTIE T, TIBSHIRANI R. The Elements of Statistical Learning, Volume 1[M]. Berlin, 2001.
[68] BRIER G W, et al. Verification of Forecasts Expressed in Terms of Probability[J]. Monthly Weather Review, 1950, 78(1): 1-3.
[69] LECUN Y, BENGIO Y, HINTON G. Deep Learning[J]. nature, 2015, 521(7553): 436-444.
[70] NAEINI M P, COOPER G, HAUSKRECHT M. Obtaining Well Calibrated Probabilities Using Bayesian Binning[C]//AAAI Conference on Artificial Intelligence. 2015.
[71] HARRELL F E, CALIFF R M, PRYOR D B, et al. Evaluating the Yield of Medical Tests[J]. JAMA, 1982, 247(18): 2543-2546.
[72] KRIZHEVSKY A, HINTON G, et al. Learning Multiple Layers of Features from Tiny Images[M]. Citeseer, 2009.
[73] XIAO H, RASUL K, VOLLGRAF R. Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms[EB/OL]. (2017-08-28).
[74] NETZER Y, WANG T, COATES A, et al. Reading Digits in Natural Images with Unsupervised Feature Learning[Z]. 2011.
[75] ZAGORUYKO S, KOMODAKIS N. Wide Residual Networks[C]//British Machine Vision Conference (BMVC). BMVA Press, 2016: 87.1-87.12.
[76] SHRIDHAR K, LAUMANN F, LIWICKI M. Uncertainty Estimations by Softplus Normalization in Bayesian Convolutional Neural Networks with Variational Inference[A]. 2018. arXiv: 1806.05978.
[77] SHRIDHAR K, LAUMANN F, LIWICKI M. A Comprehensive Guide to Bayesian Convolutional Neural Network with Variational Inference[A]. 2019. arXiv: 1901.02731.
[78] RONNEBERGER O, FISCHER P, BROX T. U-Net: Convolutional Networks for Biomedical Image Segmentation[C]//International Conference on Medical Image Computing and Computer-Assisted Intervention. Springer, 2015: 234-241.
[79] OKTAY O, SCHLEMPER J, FOLGOC L L, et al. Attention U-Net: Learning Where to Look for the Pancreas[A]. 2018. arXiv: 1804.03999.
[80] ARMATO III S G, MCLENNAN G, BIDAUT L, et al. The Lung Image Database Consortium (LIDC) and Image Database Resource Initiative (IDRI): a Completed Reference Database of Lung Nodules on CT Scans[J]. Medical Physics, 2011, 38(2): 915-931.