In the past years, deep learning has seen an increase in usage in the domain of histopathological applications. However, while these approaches have shown great potential, in high-risk environments deep learning models need to be able to judge their uncertainty and be able to reject inputs when there is a significant chance of misclassification. In this work, we conduct a rigorous evaluation of the most commonly used uncertainty and robustness methods for the classification of Whole Slide Images, with a focus on the task of selective classification, where the model should reject the classification in situations in which it is uncertain. We conduct our experiments on tile-level under the aspects of domain shift and label noise, as well as on slide-level. In our experiments, we compare Deep Ensembles, Monte-Carlo Dropout, Stochastic Variational Inference, Test-Time Data Augmentation as well as ensembles of the latter approaches. We observe that ensembles of methods generally lead to better uncertainty estimates as well as an increased robustness towards domain shifts and label noise, while contrary to results from classical computer vision benchmarks no systematic gain of the other methods can be shown. Across methods, a rejection of the most uncertain samples reliably leads to a significant increase in classification accuracy on both in-distribution as well as out-of-distribution data. Furthermore, we conduct experiments comparing these methods under varying conditions of label noise. Lastly, we publish our code framework to facilitate further research on uncertainty estimation on histopathological data.
相關內容
While deep learning techniques have provided the state-of-the-art performance in various clinical tasks, explainability regarding their decision-making process can greatly enhance the credence of these methods for safer and quicker clinical adoption. With high flexibility, Gradient-weighted Class Activation Mapping (Grad-CAM) has been widely adopted to offer intuitive visual interpretation of various deep learning models' reasoning processes in computer-assisted diagnosis. However, despite the popularity of the technique, there is still a lack of systematic study on Grad-CAM's performance on different deep learning architectures. In this study, we investigate its robustness and effectiveness across different popular deep learning models, with a focus on the impact of the networks' depths and architecture types, by using a case study of automatic pneumothorax diagnosis in X-ray scans. Our results show that deeper neural networks do not necessarily contribute to a strong improvement of pneumothorax diagnosis accuracy, and the effectiveness of GradCAM also varies among different network architectures.
Although there have been remarkable advances in dialogue systems through the dialogue systems technology competition (DSTC), it remains one of the key challenges to building a robust task-oriented dialogue system with a speech interface. Most of the progress has been made for text-based dialogue systems since there are abundant datasets with written corpora while those with spoken dialogues are very scarce. However, as can be seen from voice assistant systems such as Siri and Alexa, it is of practical importance to transfer the success to spoken dialogues. In this paper, we describe our engineering effort in building a highly successful model that participated in the speech-aware dialogue systems technology challenge track in DSTC11. Our model consists of three major modules: (1) automatic speech recognition error correction to bridge the gap between the spoken and the text utterances, (2) text-based dialogue system (D3ST) for estimating the slots and values using slot descriptions, and (3) post-processing for recovering the error of the estimated slot value. Our experiments show that it is important to use an explicit automatic speech recognition error correction module, post-processing, and data augmentation to adapt a text-based dialogue state tracker for spoken dialogue corpora.
Machine learning (ML) is becoming a critical tool for interrogation of large complex data. Labeling, defined as the process of adding meaningful annotations, is a crucial step of supervised ML. However, labeling datasets is time consuming. Here we show that convolutional neural networks (CNNs), trained on crudely labeled astronomical videos, can be leveraged to improve the quality of data labeling and reduce the need for human intervention. We use videos of the solar magnetic field, crudely labeled into two classes: emergence or non-emergence of bipolar magnetic regions (BMRs), based on their first detection on the solar disk. We train CNNs using crude labels, manually verify, correct labeling vs. CNN disagreements, and repeat this process until convergence. Traditionally, flux emergence labelling is done manually. We find that a high-quality labeled dataset, derived through this iterative process, reduces the necessary manual verification by 50%. Furthermore, by gradually masking the videos and looking for maximum change in CNN inference, we locate BMR emergence time without retraining the CNN. This demonstrates the versatility of CNNs for simplifying the challenging task of labeling complex dynamic events.
Benefiting from the development of deep learning, text-to-speech (TTS) techniques using clean speech have achieved significant performance improvements. The data collected from real scenes often contain noise and generally needs to be denoised by speech enhancement models. Noise-robust TTS models are often trained using the enhanced speech, which thus suffer from speech distortion and background noise that affect the quality of the synthesized speech. Meanwhile, it was shown that self-supervised pre-trained models exhibit excellent noise robustness on many speech tasks, implying that the learned representation has a better tolerance for noise perturbations. In this work, we therefore explore pre-trained models to improve the noise robustness of TTS models. Based on HIFI-GAN we first propose a representation-to-waveform vocoder, which aims to learn to map the representation of pre-trained models to the waveform. We then propose a text-to-representation Fastspeech2 model, which aims to learn to map text to pre-trained model representations. Experimental results on the LJSpeech and LibriTTS datasets show that our method outperforms those using speech enhancement methods in both subjective and objective metrics. Audio samples are available at: //zqs01.github.io/rep2wav/.
Machine learning (ML) has made BigCloneBench popular for semantic clone detection tools. However, BigCloneBench only has a few Java semantic clones. In addition, due to the design principles of how the benchmark was created, imbalance issues have been identified, including the ambiguity in the definition of semantic clones. Thus, ML-based clone detection algorithms trained on BigCloneBench may overlook semantic clones or report incorrect results. The SemanticCloneBench features Stack Overflow clones of several languages. However, it lacks samples for ML-based clone detection. There is also a marked lack of cross-language clone benchmarks. The widely used CLCDSA dataset lacks reusable examples that can't be used in real-world software systems, making it inadequate for ML-based clone detection. The OpenAI GPT-3 model has shown outstanding text production, including code generation and summarization. In this paper, we used the GPT-3 model to generate a complete benchmark for both semantic and cross-language clones. Using SemanticCloneBench's genuine language clones, we tested several prompts to see which yielded better results using GPT-3 question formulation. Then, we used NiCad to filter Type-1 and Type-2 clones from GPT-3 output. We used a GUI-assisted Clone Validator tool to manually validate all clone pairings with nine judges. Functionality testing and CloneCognition verified our benchmark has no syntactic clones. Later, we validated SourcererCC, Oreo and CLCDSA tools on our benchmark. The poor performance of these tools suggests GPTCloneBench has no syntactic clone. From 77,207 Clone pairs of SemanticCloneBench/GPT-3 output, we created a benchmark with 37,149 genuine semantic clone pairs, 19,288 false semantic pairs, and 20,770 cross-language clones across four languages (Java, C, C#, and Python).
In recent years, deep learning has emerged as a powerful approach in remote sensing applications, particularly in segmentation and classification techniques that play a crucial role in extracting significant land features from satellite and aerial imagery. However, only a limited number of papers have discussed the use of deep learning for interactive segmentation in land cover classification tasks. In this study, we aim to bridge the gap between interactive segmentation and remote sensing image analysis by conducting a benchmark study on various deep learning-based interactive segmentation models. We assessed the performance of five state-of-the-art interactive segmentation methods (SimpleClick, FocalClick, Iterative Click Loss (ICL), Reviving Iterative Training with Mask Guidance for Interactive Segmentation (RITM), and Segment Anything (SAM)) on two high-resolution aerial imagery datasets. To enhance the segmentation results without requiring multiple models, we introduced the Cascade-Forward Refinement (CFR) approach, an innovative inference strategy for interactive segmentation. We evaluated these interactive segmentation methods on various land cover types, object sizes, and band combinations in remote sensing. Surprisingly, the popularly discussed method, SAM, proved to be ineffective for remote sensing images. Conversely, the point-based approach used in the SimpleClick models consistently outperformed the other methods in all experiments. Building upon these findings, we developed a dedicated online tool called RSISeg for interactive segmentation of remote sensing data. RSISeg incorporates a well-performing interactive model, fine-tuned with remote sensing data. Additionally, we integrated the SAM model into this tool. Compared to existing interactive segmentation tools, RSISeg offers strong interactivity, modifiability, and adaptability to remote sensing data.
We hypothesize that due to the greedy nature of learning in multi-modal deep neural networks, these models tend to rely on just one modality while under-fitting the other modalities. Such behavior is counter-intuitive and hurts the models' generalization, as we observe empirically. To estimate the model's dependence on each modality, we compute the gain on the accuracy when the model has access to it in addition to another modality. We refer to this gain as the conditional utilization rate. In the experiments, we consistently observe an imbalance in conditional utilization rates between modalities, across multiple tasks and architectures. Since conditional utilization rate cannot be computed efficiently during training, we introduce a proxy for it based on the pace at which the model learns from each modality, which we refer to as the conditional learning speed. We propose an algorithm to balance the conditional learning speeds between modalities during training and demonstrate that it indeed addresses the issue of greedy learning. The proposed algorithm improves the model's generalization on three datasets: Colored MNIST, Princeton ModelNet40, and NVIDIA Dynamic Hand Gesture.
The remarkable practical success of deep learning has revealed some major surprises from a theoretical perspective. In particular, simple gradient methods easily find near-optimal solutions to non-convex optimization problems, and despite giving a near-perfect fit to training data without any explicit effort to control model complexity, these methods exhibit excellent predictive accuracy. We conjecture that specific principles underlie these phenomena: that overparametrization allows gradient methods to find interpolating solutions, that these methods implicitly impose regularization, and that overparametrization leads to benign overfitting. We survey recent theoretical progress that provides examples illustrating these principles in simpler settings. We first review classical uniform convergence results and why they fall short of explaining aspects of the behavior of deep learning methods. We give examples of implicit regularization in simple settings, where gradient methods lead to minimal norm functions that perfectly fit the training data. Then we review prediction methods that exhibit benign overfitting, focusing on regression problems with quadratic loss. For these methods, we can decompose the prediction rule into a simple component that is useful for prediction and a spiky component that is useful for overfitting but, in a favorable setting, does not harm prediction accuracy. We focus specifically on the linear regime for neural networks, where the network can be approximated by a linear model. In this regime, we demonstrate the success of gradient flow, and we consider benign overfitting with two-layer networks, giving an exact asymptotic analysis that precisely demonstrates the impact of overparametrization. We conclude by highlighting the key challenges that arise in extending these insights to realistic deep learning settings.
A key requirement for the success of supervised deep learning is a large labeled dataset - a condition that is difficult to meet in medical image analysis. Self-supervised learning (SSL) can help in this regard by providing a strategy to pre-train a neural network with unlabeled data, followed by fine-tuning for a downstream task with limited annotations. Contrastive learning, a particular variant of SSL, is a powerful technique for learning image-level representations. In this work, we propose strategies for extending the contrastive learning framework for segmentation of volumetric medical images in the semi-supervised setting with limited annotations, by leveraging domain-specific and problem-specific cues. Specifically, we propose (1) novel contrasting strategies that leverage structural similarity across volumetric medical images (domain-specific cue) and (2) a local version of the contrastive loss to learn distinctive representations of local regions that are useful for per-pixel segmentation (problem-specific cue). We carry out an extensive evaluation on three Magnetic Resonance Imaging (MRI) datasets. In the limited annotation setting, the proposed method yields substantial improvements compared to other self-supervision and semi-supervised learning techniques. When combined with a simple data augmentation technique, the proposed method reaches within 8% of benchmark performance using only two labeled MRI volumes for training, corresponding to only 4% (for ACDC) of the training data used to train the benchmark.
Graph representation learning for hypergraphs can be used to extract patterns among higher-order interactions that are critically important in many real world problems. Current approaches designed for hypergraphs, however, are unable to handle different types of hypergraphs and are typically not generic for various learning tasks. Indeed, models that can predict variable-sized heterogeneous hyperedges have not been available. Here we develop a new self-attention based graph neural network called Hyper-SAGNN applicable to homogeneous and heterogeneous hypergraphs with variable hyperedge sizes. We perform extensive evaluations on multiple datasets, including four benchmark network datasets and two single-cell Hi-C datasets in genomics. We demonstrate that Hyper-SAGNN significantly outperforms the state-of-the-art methods on traditional tasks while also achieving great performance on a new task called outsider identification. Hyper-SAGNN will be useful for graph representation learning to uncover complex higher-order interactions in different applications.
小貼士
登錄享
相關主題
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191