Filters








2,647 Hits in 6.6 sec

Real-Time Uncertainty Estimation in Computer Vision via Uncertainty-Aware Distribution Distillation [article]

Yichen Shen, Zhilu Zhang, Mert R. Sabuncu, Lin Sun
2020 arXiv   pre-print
We propose a simple, easy-to-optimize distillation method for learning the conditional predictive distribution of a pre-trained dropout model for fast, sample-free uncertainty estimation in computer vision  ...  uncertainty quantification, while achieving improved quality of both the uncertainty estimates and predictive performance over the regular dropout model.  ...  The running time of MC Dropout is optimized by caching results before the first dropout layer for a fair comparison.  ... 
arXiv:2007.15857v2 fatcat:j4gzazka35btlnv3u3zl7gjyni

STUN: Self-Teaching Uncertainty Estimation for Place Recognition [article]

Kaiwen Cai, Chris Xiaoxuan Lu, Xiaowei Huang
2022 arXiv   pre-print
During the online inference phase, we only use the student net to generate a place prediction in conjunction with the uncertainty.  ...  However, a place recognition in the wild often suffers from erroneous predictions due to image variations, e.g., changing viewpoints and street appearance.  ...  BTL [5] : We follow the parameters of the original paper without extra modification for a fair comparison.  ... 
arXiv:2203.01851v1 fatcat:7kwmog5f3fffrads64cvx6bf7a

Gaze Training by Modulated Dropout Improves Imitation Learning [article]

Yuying Chen, Congcong Liu, Lei Tai, Ming Liu, Bertram E. Shi
2019 arXiv   pre-print
Prediction error in steering commands is reduced by 23.5% compared to uniform dropout.  ...  Consistent with these results, the gaze-modulated dropout net shows lower model uncertainty.  ...  Gaze-Modulated Dropout Evaluation 1) Prediction error vs. drop probability: To make a fair comparison, we scan over dps from 0.1 to 0.8 with a step of 0.1.  ... 
arXiv:1904.08377v2 fatcat:yleqwqx2nfbjbab5ela26p67ne

Iterative Distillation for Better Uncertainty Estimates in Multitask Emotion Recognition [article]

Didan Deng, Liang Wu, Bertram E. Shi
2021 arXiv   pre-print
Our method generates single student models that provide accurate estimates of uncertainty for in-domain samples and a student ensemble that can detect out-of-domain samples.  ...  From a Bayesian perspective, we propose to use deep ensembles to capture uncertainty for multiple emotion descriptors, i.e., action units, discrete expression labels and continuous descriptors.  ...  For a fair comparison, we adopted the test-time cross-validation in [1] to compute the NLL in TS. The optimal temperature was optimized on a randomlysplit half of the validation set.  ... 
arXiv:2108.04228v2 fatcat:zlqous3inbdnnlvmvzhsvwtqsy

Semi-supervised Left Atrium Segmentation with Mutual Consistency Training [article]

Yicheng Wu, Minfeng Xu, Zongyuan Ge, Jianfei Cai, Lei Zhang
2021 arXiv   pre-print
We believe that these unlabeled regions may contain more crucial information to minimize the uncertainty prediction for the model and should be emphasized in the training process.  ...  Such mutual consistency encourages the two decoders to have consistent and low-entropy predictions and enables the model to gradually capture generalized features from these unlabeled challenging regions  ...  We also appreciate the efforts devoted to collect and share the LA database [16] and several available repositories [6, 7, 17] .  ... 
arXiv:2103.02911v2 fatcat:q5la4hz2hvcffeqju5e47pok5a

College Student Retention Risk Analysis From Educational Database using Multi-Task Multi-Modal Neural Fusion [article]

Mohammad Arif Ul Alam
2021 arXiv   pre-print
We develop a Multimodal Spatiotemporal Neural Fusion network for Multi-Task Learning (MSNF-MTCL) to predict 5 important students' retention risks: future dropout, next semester dropout, type of dropout  ...  , duration of dropout and cause of dropout.  ...  [32] proposed a fair student dropout prediction system from educational database.  ... 
arXiv:2109.05178v1 fatcat:4adtjrfj2ba6pceqv4am3kwuda

Assembly Quality Detection Based on Class-Imbalanced Semi-Supervised Learning

Zichen Lu, Jiabin Jiang, Pin Cao, Yongying Yang
2021 Applied Sciences  
Based on the mean teacher algorithm, the proposed algorithm uses certainty to select reliable teacher predictions for student learning dynamically, and loss functions are modified to improve the model's  ...  Due to the imperfect assembly process, the unqualified assembly of a missing gasket or lead seal will affect the product's performance and possibly cause safety accidents.  ...  Acknowledgments: This work was supported by the State Key Laboratory of Modern Optical Instrumentation of Zhejiang University and Zernike Optics Co., Ltd.  ... 
doi:10.3390/app112110373 fatcat:dhzewau3r5ffpml5h7vez5gtuu

Certainty Driven Consistency Loss on Multi-Teacher Networks for Semi-Supervised Learning [article]

Lu Liu, Robby T. Tan
2021 arXiv   pre-print
In this paper, we propose a novel Certainty-driven Consistency Loss (CCL) that exploits the predictive uncertainty in the consistency loss to let the student dynamically learn from reliable targets.  ...  Typically, a student model is trained to be consistent with teacher prediction for the inputs under different perturbations.  ...  In Filtering CCL, shown inFig. 1 (a), the teacher filters out uncertain predictions and gradually selects a subset of certain predictions (i.e. of low uncertainty), that are robust targets for the student  ... 
arXiv:1901.05657v7 fatcat:zovhmsmelfh3nah2t7bsz27gpu

Uncertainty-aware Mean Teacher for Source-free Unsupervised Domain Adaptive 3D Object Detection [article]

Deepti Hegde, Vishwanath Sindagi, Velat Kilic, A. Brinton Cooper, Mark Foster, Vishal Patel
2021 arXiv   pre-print
Effectively, we perform automatic soft-sampling of pseudo-labeled data while aligning predictions from the student and teacher networks.  ...  In order to avoid reinforcing errors caused by label noise, we propose an uncertainty-aware mean teacher framework which implicitly filters incorrect pseudo-labels during training.  ...  -Initialize student model with φ s for 1 ≤ epoch ≤ num epochs do -Train student net with target data annotated with pseudo-labels {Y pt i,J } M i=1 i=1 To ensure a fair comparison across all  ... 
arXiv:2109.14651v1 fatcat:5wo4dotrb5ejtnqcgqneagnkyi

Inconsistency-aware Uncertainty Estimation for Semi-supervised Medical Image Segmentation [article]

Yinghuan Shi, Jian Zhang, Tong Ling, Jiwen Lu, Yefeng Zheng, Qian Yu, Lei Qi, Yang Gao
2021 arXiv   pre-print
In this paper, we investigate a novel method of estimating uncertainty.  ...  Therefore, we present a new semi-supervised segmentation model, namely, conservative-radical network (CoraNet in short) based on our uncertainty estimation and separate self-training strategy.  ...  In this paper, we proposed a novel method of estimating uncertainty by capturing the inconsistent prediction between multiple cost-sensitive settings.  ... 
arXiv:2110.08762v1 fatcat:adu3h2pqgba4fh2efryh6tofqi

Leveraging Labeling Representations in Uncertainty-based Semi-supervised Segmentation [article]

Sukesh Adiga V, Jose Dolz, Herve Lombaert
2022 arXiv   pre-print
The predictions of unlabeled data are not reliable, therefore, uncertainty-aware methods have been proposed to gradually learn from meaningful and reliable predictions.  ...  A prominent way to utilize the unlabeled data is by consistency training which commonly uses a teacher-student network, where a teacher guides a student segmentation.  ...  Acknowledgments: This research work was partly funded by the Canada Research Chair on Shape Analysis in Medical Imaging, the Natural Sciences and Engineering Research Council of Canada (NSERC), and the  ... 
arXiv:2203.05682v1 fatcat:fcfcg7salvfx3b64shxwu7e4di

Uncertainty-Guided Mutual Consistency Learning for Semi-Supervised Medical Image Segmentation [article]

Yichi Zhang, Qingcheng Liao, Rushi Jiao, Jicong Zhang
2021 arXiv   pre-print
Medical image segmentation is a fundamental and critical step in many clinical approaches.  ...  In this paper, we propose a novel uncertainty-guided mutual consistency learning framework to effectively exploit unlabeled data by integrating intra-task consistency learning from up-to-date predictions  ...  fair comparison.  ... 
arXiv:2112.02508v1 fatcat:ofgv42dygvhyxphgh2wbcgdvoy

Predicting Math Student Success in the Initial Phase of College With Sparse Information Using Approaches From Statistical Learning

Pascal Kilian, Frank Loose, Augustin Kelava
2020 Frontiers in Education  
We investigate the completion of a first semester course as a dropout indicator and thereby provide not only good predictions, but also generate interpretable and practicable results together with easy-to-understand  ...  In math teacher education, dropout research relies mostly on frameworks which carry out extensive variable collections leading to a lack of practical applicability.  ...  This means the subject of research in this paper is not dropouts (from university or the study program) but dropouts and success in this lecture in the sense of a non-completion rate.  ... 
doi:10.3389/feduc.2020.502698 fatcat:4frsjsgli5e3pjsbzo7kpook64

A Review of Uncertainty Quantification in Deep Learning: Techniques, Applications and Challenges [article]

Moloud Abdar, Farhad Pourpanah, Sadiq Hussain, Dana Rezazadegan, Li Liu, Mohammad Ghavamzadeh, Paul Fieguth, Xiaochun Cao, Abbas Khosravi, U Rajendra Acharya, Vladimir Makarenkov, Saeid Nahavandi
2021 arXiv   pre-print
Uncertainty quantification (UQ) plays a pivotal role in reduction of uncertainties during both optimization and decision making processes.  ...  It can be applied to solve a variety of real-world applications in science and engineering.  ...  A schematic comparison of the three different uncertainty models [9] (MC dropout, Boostrap model and GMM is provided in Fig. 2 .  ... 
arXiv:2011.06225v4 fatcat:wwnl7duqwbcqbavat225jkns5u

Adversarial Distillation of Bayesian Neural Network Posteriors [article]

Kuan-Chieh Wang, Paul Vicol, James Lucas, Li Gu, Roger Grosse, Richard Zemel
2018 arXiv   pre-print
Bayesian neural networks (BNNs) allow us to reason about uncertainty in a principled way.  ...  However, SGLD and its extensions require storage of many copies of the model parameters, a potentially prohibitive cost, especially for large neural networks.  ...  We did not use momentum, for fair comparison with vanilla SGLD, which did not use momentum.  ... 
arXiv:1806.10317v1 fatcat:lzsdxuaxbffjxao3jzzpus4htm
« Previous Showing results 1 — 15 out of 2,647 results