Abstract
Due to the imbalanced and limited data, semi-supervised medical image segmentation methods often fail to produce superior performance for some specific tailed classes. Inadequate training for those particular classes could introduce more noise to the generated pseudo labels, affecting overall learning. To alleviate this shortcoming and identify the under-performing classes, we propose maintaining a confidence array that records class-wise performance during training. A fuzzy fusion of these confidence scores is proposed to adaptively prioritize individual confidence metrics in every sample rather than traditional ensemble approaches, where a set of predefined fixed weights are assigned for all the test cases. Further, we introduce a robust class-wise sampling method and dynamic stabilization for a better training strategy. Our proposed method considers all the under-performing classes with dynamic weighting and tries to remove most of the noises during training. Upon evaluation on two cardiac MRI datasets, ACDC and MMWHS, our proposed method shows effectiveness and generalizability and outperforms several state-of-the-art methods found in the literature.
Identifying Out-of-Distribution Samples in Real-Time for Safety-Critical 2D Object Detection with Margin Entropy Loss
Authors: Yannik Blei, Nicolas Jourdan, Nils Gählert
Subjects: Computer Vision and Pattern Recognition (cs.CV)
Abstract
Convolutional Neural Networks (CNNs) are nowadays often employed in vision-based perception stacks for safetycritical applications such as autonomous driving or Unmanned Aerial Vehicles (UAVs). Due to the safety requirements in those use cases, it is important to know the limitations of the CNN and, thus, to detect Out-of-Distribution (OOD) samples. In this work, we present an approach to enable OOD detection for 2D object detection by employing the margin entropy (ME) loss. The proposed method is easy to implement and can be applied to most existing object detection architectures. In addition, we introduce Separability as a metric for detecting OOD samples in object detection. We show that a CNN trained with the ME loss significantly outperforms OOD detection using standard confidence scores. At the same time, the runtime of the underlying object detection framework remains constant rendering the ME loss a powerful tool to enable OOD detection.
Seq-UPS: Sequential Uncertainty-aware Pseudo-label Selection for Semi-Supervised Text Recognition
Authors: Gaurav Patel, Jan Allebach, Qiang Qiu
Subjects: Computer Vision and Pattern Recognition (cs.CV)
Abstract
This paper looks at semi-supervised learning (SSL) for image-based text recognition. One of the most popular SSL approaches is pseudo-labeling (PL). PL approaches assign labels to unlabeled data before re-training the model with a combination of labeled and pseudo-labeled data. However, PL methods are severely degraded by noise and are prone to over-fitting to noisy labels, due to the inclusion of erroneous high confidence pseudo-labels generated from poorly calibrated models, thus, rendering threshold-based selection ineffective. Moreover, the combinatorial complexity of the hypothesis space and the error accumulation due to multiple incorrect autoregressive steps posit pseudo-labeling challenging for sequence models. To this end, we propose a pseudo-label generation and an uncertainty-based data selection framework for semi-supervised text recognition. We first use Beam-Search inference to yield highly probable hypotheses to assign pseudo-labels to the unlabelled examples. Then we adopt an ensemble of models, sampled by applying dropout, to obtain a robust estimate of the uncertainty associated with the prediction, considering both the character-level and word-level predictive distribution to select good quality pseudo-labels. Extensive experiments on several benchmark handwriting and scene-text datasets show that our method outperforms the baseline approaches and the previous state-of-the-art semi-supervised text-recognition methods.
Keyword: scaling
Efficient Methods for Natural Language Processing: A Survey
Authors: Marcos Treviso, Tianchu Ji, Ji-Ung Lee, Betty van Aken, Qingqing Cao, Manuel R. Ciosici, Michael Hassid, Kenneth Heafield, Sara Hooker, Pedro H. Martins, André F. T. Martins, Peter Milder, Colin Raffel, Edwin Simpson, Noam Slonim, Niranjan Balasubramanian, Leon Derczynski, Roy Schwartz
Abstract
Getting the most out of limited resources allows advances in natural language processing (NLP) research and practice while being conservative with resources. Those resources may be data, time, storage, or energy. Recent work in NLP has yielded interesting results from scaling; however, using only scale to improve results means that resource consumption also scales. That relationship motivates research into efficient methods that require less resources to achieve similar results. This survey relates and synthesises methods and findings in those efficiencies in NLP, aiming to guide new researchers in the field and inspire the development of new methods.
Spherical Graph Drawing by Multi-dimensional Scaling
Authors: Jacob Miller, Vahan Huroyan, Stephen Kobourov
Abstract
We describe an efficient and scalable spherical graph embedding method. The method uses a generalization of the Euclidean stress function for Multi-Dimensional Scaling adapted to spherical space, where geodesic pairwise distances are employed instead of Euclidean distances. The resulting spherical stress function is optimized by means of stochastic gradient descent. Quantitative and qualitative evaluations demonstrate the scalability and effectiveness of the proposed method. We also show that some graph families can be embedded with lower distortion on the sphere, than in Euclidean and hyperbolic spaces.
Efficient preconditioners for solving dynamical optimal transport via interior point methods
Authors: Enrico Facca, Gabriele Todeschi, Andrea Natale, Michele Benzi
Subjects: Numerical Analysis (math.NA); Optimization and Control (math.OC)
Abstract
In this paper we address the numerical solution of the quadratic optimal transport problem in its dynamical form, the so-called Benamou-Brenier formulation. When solved using interior point methods, the main computational bottleneck is the solution of large saddle point linear systems arising from the associated Newton-Raphson scheme. The main purpose of this paper is to design efficient preconditioners to solve these linear systems via iterative methods. Among the proposed preconditioners, we introduce one based on the partial commutation of the operators that compose the dual Schur complement of these saddle point linear systems, which we refer as BB-preconditioner. A series of numerical tests show that the BB-preconditioner is the most efficient among those presented, with a CPU-time scaling only slightly more than linearly with respect to the number of unknowns used to discretize the problem.
Holomorphic Equilibrium Propagation Computes Exact Gradients Through Finite Size Oscillations
Abstract
Equilibrium propagation (EP) is an alternative to backpropagation (BP) that allows the training of deep neural networks with local learning rules. It thus provides a compelling framework for training neuromorphic systems and understanding learning in neurobiology. However, EP requires infinitesimal teaching signals, thereby limiting its applicability in noisy physical systems. Moreover, the algorithm requires separate temporal phases and has not been applied to large-scale problems. Here we address these issues by extending EP to holomorphic networks. We show analytically that this extension naturally leads to exact gradients even for finite-amplitude teaching signals. Importantly, the gradient can be computed as the first Fourier coefficient from finite neuronal activity oscillations in continuous time without requiring separate phases. Further, we demonstrate in numerical simulations that our approach permits robust estimation of gradients in the presence of noise and that deeper models benefit from the finite teaching signals. Finally, we establish the first benchmark for EP on the ImageNet 32x32 dataset and show that it matches the performance of an equivalent network trained with BP. Our work provides analytical insights that enable scaling EP to large-scale problems and establishes a formal framework for how oscillations could support learning in biological and neuromorphic systems.
Keyword: calibration
ProCo: Prototype-aware Contrastive Learning for Long-tailed Medical Image Classification
Abstract
Medical image classification has been widely adopted in medical image analysis. However, due to the difficulty of collecting and labeling data in the medical area, medical image datasets are usually highly-imbalanced. To address this problem, previous works utilized class samples as prior for re-weighting or re-sampling but the feature representation is usually still not discriminative enough. In this paper, we adopt the contrastive learning to tackle the long-tailed medical imbalance problem. Specifically, we first propose the category prototype and adversarial proto-instance to generate representative contrastive pairs. Then, the prototype recalibration strategy is proposed to address the highly imbalanced data distribution. Finally, a unified proto-loss is designed to train our framework. The overall framework, namely as Prototype-aware Contrastive learning (ProCo), is unified as a single-stage pipeline in an end-to-end manner to alleviate the imbalanced problem in medical image classification, which is also a distinct progress than existing works as they follow the traditional two-stage pipeline. Extensive experiments on two highly-imbalanced medical image classification datasets demonstrate that our method outperforms the existing state-of-the-art methods by a large margin.
Keyword: out of distribution detection
There is no result
Keyword: out-of-distribution detection
There is no result
Keyword: expected calibration error
There is no result
Keyword: overconfident
There is no result
Keyword: overconfidence
There is no result
Keyword: confidence
Addressing Class Imbalance in Semi-supervised Image Segmentation: A Study on Cardiac MRI
Identifying Out-of-Distribution Samples in Real-Time for Safety-Critical 2D Object Detection with Margin Entropy Loss
Seq-UPS: Sequential Uncertainty-aware Pseudo-label Selection for Semi-Supervised Text Recognition
Keyword: scaling
Efficient Methods for Natural Language Processing: A Survey
Spherical Graph Drawing by Multi-dimensional Scaling
Efficient preconditioners for solving dynamical optimal transport via interior point methods
Holomorphic Equilibrium Propagation Computes Exact Gradients Through Finite Size Oscillations
Keyword: calibration
ProCo: Prototype-aware Contrastive Learning for Long-tailed Medical Image Classification