Although deep learning has achieved great success in image classification tasks, its performance is subject to the quantity and quality of training samples. For classification of polarimetric synthetic aperture radar (PolSAR) images, it is nearly impossible to annotate the images from visual interpretation. Therefore, it is urgent for remote sensing scientists to develop new techniques for PolSAR image classification under the condition of very few training samples. In this letter, we take the advantage of active learning and propose active ensemble deep learning (AEDL) for PolSAR image classification. We first show that only 35\% of the predicted labels of a deep learning model's snapshots near its convergence were exactly the same. The disagreement between snapshots is non-negligible. From the perspective of multiview learning, the snapshots together serve as a good committee to evaluate the importance of unlabeled instances. Using the snapshots committee to give out the informativeness of unlabeled data, the proposed AEDL achieved better performance on two real PolSAR images compared with standard active learning strategies. It achieved the same classification accuracy with only 86% and 55% of the training samples compared with breaking ties active learning and random selection for the Flevoland dataset.
Graph neural networks are promising architecture for learning and inference with graph-structured data. Yet difficulties in modelling the ``parts'' and their ``interactions'' still persist in terms of graph classification, where graph-level representations are usually obtained by squeezing the whole graph into a single vector through graph pooling. From complex systems point of view, mixing all the parts of a system together can affect both model interpretability and predictive performance, because properties of a complex system arise largely from the interaction among its components. We analyze the intrinsic difficulty in graph classification under the unified concept of ``resolution dilemmas'' with learning theoretic recovery guarantees, and propose ``SLIM'', an inductive neural network model for Structural Landmarking and Interaction Modelling. It turns out, that by solving the resolution dilemmas, and leveraging explicit interacting relation between component parts of a graph to explain its complexity, SLIM is more interpretable, accurate, and offers new insight in graph representation learning.
Recent years, the approaches based on neural networks have shown remarkable potential for sentence modeling. There are two main neural network structures: recurrent neural network (RNN) and convolution neural network (CNN). RNN can capture long term dependencies and store the semantics of the previous information in a fixed-sized vector. However, RNN is a biased model and its ability to extract global semantics is restricted by the fixed-sized vector. Alternatively, CNN is able to capture n-gram features of texts by utilizing convolutional filters. But the width of convolutional filters restricts its performance. In order to combine the strengths of the two kinds of networks and alleviate their shortcomings, this paper proposes Attention-based Multichannel Convolutional Neural Network (AMCNN) for text classification. AMCNN utilizes a bi-directional long short-term memory to encode the history and future information of words into high dimensional representations, so that the information of both the front and back of the sentence can be fully expressed. Then the scalar attention and vectorial attention are applied to obtain multichannel representations. The scalar attention can calculate the word-level importance and the vectorial attention can calculate the feature-level importance. In the classification task, AMCNN uses a CNN structure to cpture word relations on the representations generated by the scalar and vectorial attention mechanism instead of calculating the weighted sums. It can effectively extract the n-gram features of the text. The experimental results on the benchmark datasets demonstrate that AMCNN achieves better performance than state-of-the-art methods. In addition, the visualization results verify the semantic richness of multichannel representations.
This paper presents our approach for task 2 and task 3 of Social Media Mining for Health (SMM4H) 2020 shared tasks. In task 2, we have to differentiate adverse drug reaction (ADR) tweets from nonADR tweets and is treated as binary classification. Task3 involves extracting ADR mentions and then mapping them to MedDRA codes. Extracting ADR mentions is treated as sequence labeling and normalizing ADR mentions is treated as multi-class classification. Our system is based on pre-trained language model RoBERTa and it achieves a) F1-score of 58% in task2 which is 12% more than the average score b) relaxed F1-score of 70.1% in ADR extraction of task 3 which is 13.7% more than the average score and relaxed F1-score of 35% in ADR extraction + normalization of task3 which is 5.8% more than the average score. Overall, our models achieve promising results in both the tasks with significant improvements over average scores.
Recent works have empirically shown that there exist adversarial examples that can be hidden from neural network interpretability (namely, making network interpretation maps visually similar), or interpretability is itself susceptible to adversarial attacks. In this paper, we theoretically show that with a proper measurement of interpretation, it is actually difficult to prevent prediction-evasion adversarial attacks from causing interpretation discrepancy, as confirmed by experiments on MNIST, CIFAR-10 and Restricted ImageNet. Spurred by that, we develop an interpretability-aware defensive scheme built only on promoting robust interpretation (without the need for resorting to adversarial loss minimization). We show that our defense achieves both robust classification and robust interpretation, outperforming state-of-the-art adversarial training methods against attacks of large perturbation in particular.
Skin cancer is among the most common cancer types. Dermoscopic image analysis improves the diagnostic accuracy for detection of malignant melanoma and other pigmented skin lesions when compared to unaided visual inspection. Hence, computer-based methods to support medical experts in the diagnostic procedure are of great interest. Fine-tuning pre-trained convolutional neural networks (CNNs) has been shown to work well for skin lesion classification. Pre-trained CNNs are usually trained with natural images of a fixed image size which is typically significantly smaller than captured skin lesion images and consequently dermoscopic images are downsampled for fine-tuning. However, useful medical information may be lost during this transformation. In this paper, we explore the effect of input image size on skin lesion classification performance of fine-tuned CNNs. For this, we resize dermoscopic images to different resolutions, ranging from 64x64 to 768x768 pixels and investigate the resulting classification performance of three well-established CNNs, namely DenseNet-121, ResNet-18, and ResNet-50. Our results show that using very small images (of size 64x64 pixels) degrades the classification performance, while images of size 128x128 pixels and above support good performance with larger image sizes leading to slightly improved classification. We further propose a novel fusion approach based on a three-level ensemble strategy that exploits multiple fine-tuned networks trained with dermoscopic images at various sizes. When applied on the ISIC 2017 skin lesion classification challenge, our fusion approach yields an area under the receiver operating characteristic curve of 89.2% and 96.6% for melanoma classification and seborrheic keratosis classification, respectively, outperforming state-of-the-art algorithms.
Data augmentation is a key practice in machine learning for improving generalization performance. However, finding the best data augmentation hyperparameters requires domain knowledge or a computationally demanding search. We address this issue by proposing an efficient approach to automatically train a network that learns an effective distribution of transformations to improve its generalization. Using bilevel optimization, we directly optimize the data augmentation parameters using a validation set. This framework can be used as a general solution to learn the optimal data augmentation jointly with an end task model like a classifier. Results show that our joint training method produces an image classification accuracy that is comparable to or better than carefully hand-crafted data augmentation. Yet, it does not need an expensive external validation loop on the data augmentation hyperparameters.
Recent works have shown that deep metric learning algorithms can benefit from weak supervision from another input modality. This additional modality can be incorporated directly into the popular triplet-based loss function as distances. Also recently, classification loss and proxy-based metric learning have been observed to lead to faster convergence as well as better retrieval results, all the while without requiring complex and costly sampling strategies. In this paper we propose an extension to the existing adaptive margin for classification-based deep metric learning. Our extension introduces a separate margin for each negative proxy per sample. These margins are computed during training from precomputed distances of the classes in the other modality. Our results set a new state-of-the-art on both on the Amazon fashion retrieval dataset as well as on the public DeepFashion dataset. This was observed with both fastText- and BERT-based embeddings for the additional textual modality. Our results were achieved with faster convergence and lower code complexity than the prior state-of-the-art.
Static gesture recognition is an effective non-verbal communication channel between a user and their devices; however many modern methods are sensitive to the relative pose of the user's hands with respect to the capture device, as parts of the gesture can become occluded. We present two methodologies for gesture recognition via synchronized recording from two depth cameras to alleviate this occlusion problem. One is a more classic approach using iterative closest point registration to accurately fuse point clouds and a single PointNet architecture for classification, and the other is a dual Point-Net architecture for classification without registration. On a manually collected data-set of 20,100 point clouds we show a 39.2% reduction in misclassification for the fused point cloud method, and 53.4% for the dual PointNet, when compared to a standard single camera pipeline.
Entomologists, ecologists and others struggle to rapidly and accurately identify the species of bumble bees they encounter in their field work and research. The current process requires the bees to be mounted, then physically shipped to a taxonomic expert for proper categorization. We investigated whether an image classification system derived from transfer learning can do this task. We used Google Inception, Oxford VGG16 and VGG19 and Microsoft ResNet 50. We found Inception and VGG classifiers were able to make some progress at identifying bumble bee species from the available data, whereas ResNet was not. Individual classifiers achieved accuracies of up to 23% for single species identification and 44% top-3 labels, where a composite model performed better, 27% and 50%. We feel the performance was most hampered by our limited data set of 5,000-plus labeled images of 29 species, with individual species represented by 59 -315 images.
The capability of the self-attention mechanism to model the long-range dependencies has catapulted its deployment in vision models. Unlike convolution operators, self-attention offers infinite receptive field and enables compute-efficient modeling of global dependencies. However, the existing state-of-the-art attention mechanisms incur high compute and/or parameter overheads, and hence unfit for compact convolutional neural networks (CNNs). In this work, we propose a simple yet effective "Ultra-Lightweight Subspace Attention Mechanism" (ULSAM), which infers different attention maps for each feature map subspace. We argue that leaning separate attention maps for each feature subspace enables multi-scale and multi-frequency feature representation, which is more desirable for fine-grained image classification. Our method of subspace attention is orthogonal and complementary to the existing state-of-the-arts attention mechanisms used in vision models. ULSAM is end-to-end trainable and can be deployed as a plug-and-play module in the pre-existing compact CNNs. Notably, our work is the first attempt that uses a subspace attention mechanism to increase the efficiency of compact CNNs. To show the efficacy of ULSAM, we perform experiments with MobileNet-V1 and MobileNet-V2 as backbone architectures on ImageNet-1K and three fine-grained image classification datasets. We achieve $\approx$13% and $\approx$25% reduction in both the FLOPs and parameter counts of MobileNet-V2 with a 0.27% and more than 1% improvement in top-1 accuracy on the ImageNet-1K and fine-grained image classification datasets (respectively). Code and trained models are available at https://github.com/Nandan91/ULSAM.
Deep kernel map networks have shown excellent performances in various classification problems including image annotation. Their general recipe consists in aggregating several layers of singular value decompositions (SVDs) -- that map data from input spaces into high dimensional spaces -- while preserving the similarity of the underlying kernels. However, the potential of these deep map networks has not been fully explored as the original setting of these networks focuses mainly on the approximation quality of their kernels and ignores their discrimination power. In this paper, we introduce a novel "end-to-end" design for deep kernel map learning that balances the approximation quality of kernels and their discrimination power. Our method proceeds in two steps; first, layerwise SVD is applied in order to build initial deep kernel map approximations and then an "end-to-end" supervised learning is employed to further enhance their discrimination power while maintaining their efficiency. Extensive experiments, conducted on the challenging ImageCLEF annotation benchmark, show the high efficiency and the out-performance of this two-step process with respect to different related methods.
This study proposes a novel biologically plausible mechanism for generating low-dimensional spike-based text representation. First, we demonstrate how to transform documents into series of spikes spike trains which are subsequently used as input in the training process of a spiking neural network (SNN). The network is composed of biologically plausible elements, and trained according to the unsupervised Hebbian learning rule, Spike-Timing-Dependent Plasticity (STDP). After training, the SNN can be used to generate low-dimensional spike-based text representation suitable for text/document classification. Empirical results demonstrate that the generated text representation may be effectively used in text classification leading to an accuracy of $80.19\%$ on the bydate version of the 20 newsgroups data set, which is a leading result amongst approaches that rely on low-dimensional text representations.
Deep Neural Networks are often brittle on image classification tasks and known to misclassify inputs. While these misclassifications may be inevitable, all failure modes cannot be considered equal. Certain misclassifications (eg. classifying the image of a dog to an airplane) can perplex humans and result in the loss of human trust in the system. Even worse, these errors (eg. a person misclassified as a primate) can have odious societal impacts. Thus, in this work, we aim to reduce inexplicable errors. To address this challenge, we first discuss methods to obtain the class-level semantics that capture the human's expectation ($M^h$) regarding which classes are semantically close {\em vs.} ones that are far away. We show that for popular image benchmarks (like CIFAR-10, CIFAR-100, ImageNet), class-level semantics can be readily obtained by leveraging either human subject studies or publicly available human-curated knowledge bases. Second, we propose the use of Weighted Loss Functions (WLFs) to penalize misclassifications by the weight of their inexplicability. Finally, we show that training (or fine-tuning) existing classifiers with the proposed methods lead to Deep Neural Networks that have (1) comparable top-1 accuracy, (2) more explicable failure modes on both in-distribution and out-of-distribution (OOD) test data, and (3) incur significantly less cost in the gathering of additional human labels compared to existing works.
Set classification aims to classify a set of observations as a whole, as opposed to classifying individual observations separately. To formally understand the unfamiliar concept of binary set classification, we first investigate the optimal decision rule under the normal distribution, which utilizes the empirical covariance of the set to be classified. We show that the number of observations in the set plays a critical role in bounding the Bayes risk. Under this framework, we further propose new methods of set classification. For the case where only a few parameters of the model drive the difference between two classes, we propose a computationally-efficient approach to parameter estimation using linear programming, leading to the Covariance-engaged LInear Programming Set (CLIPS) classifier. Its theoretical properties are investigated for both independent case and various (short-range and long-range dependent) time series structures among observations within each set. The convergence rates of estimation errors and risk of the CLIPS classifier are established to show that having multiple observations in a set leads to faster convergence rates, compared to the standard classification situation in which there is only one observation in the set. The applicable domains in which the CLIPS performs better than competitors are highlighted in a comprehensive simulation study. Finally, we illustrate the usefulness of the proposed methods in classification of real image data in histopathology.
This research proposes and evaluates scoring and assessment methods for Virtual Reality (VR) training simulators. VR simulators capture detailed n-dimensional human motion data which is useful for performance analysis. Custom made medical haptic VR training simulators were developed and used to record data from 271 trainees of multiple clinical experience levels. DTW Multivariate Prototyping (DTW-MP) is proposed. VR data was classified as Novice, Intermediate or Expert. Accuracy of algorithms applied for time-series classification were: dynamic time warping 1-nearest neighbor (DTW-1NN) 60%, nearest centroid SoftDTW classification 77.5%, Deep Learning: ResNet 85%, FCN 75%, CNN 72.5% and MCDCNN 28.5%. Expert VR data recordings can be used for guidance of novices. Assessment feedback can help trainees to improve skills and consistency. Motion analysis can identify different techniques used by individuals. Mistakes can be detected dynamically in real-time, raising alarms to prevent injuries.
Quadratic discriminant analysis (QDA) is a widely used classification technique that generalizes the linear discriminant analysis (LDA) classifier to the case of distinct covariance matrices among classes. For the QDA classifier to yield high classification performance, an accurate estimation of the covariance matrices is required. Such a task becomes all the more challenging in high dimensional settings, wherein the number of observations is comparable with the feature dimension. A popular way to enhance the performance of QDA classifier under these circumstances is to regularize the covariance matrix, giving the name regularized QDA (R-QDA) to the corresponding classifier. In this work, we consider the case in which the population covariance matrix has a spiked covariance structure, a model that is often assumed in several applications. Building on the classical QDA, we propose a novel quadratic classification technique, the parameters of which are chosen such that the fisher-discriminant ratio is maximized. Numerical simulations show that the proposed classifier not only outperforms the classical R-QDA for both synthetic and real data but also requires lower computational complexity, making it suitable to high dimensional settings.
Sequence classification is the supervised learning task of building models that predict class labels of unseen sequences of symbols. Although accuracy is paramount, in certain scenarios interpretability is a must. Unfortunately, such trade-off is often hard to achieve since we lack human-independent interpretability metrics. We introduce a novel sequence learning algorithm, that combines (i) linear classifiers - which are known to strike a good balance between predictive power and interpretability, and (ii) background knowledge embeddings. We extend the classic subsequence feature space with groups of symbols which are generated by background knowledge injected via word or graph embeddings, and use this new feature space to learn a linear classifier. We also present a new measure to evaluate the interpretability of a set of symbolic features based on the symbol embeddings. Experiments on human activity recognition from wearables and amino acid sequence classification show that our classification approach preserves predictive power, while delivering more interpretable models.
In this work we present a method for lung nodules segmentation, their texture classification and subsequent follow-up recommendation from the CT image of lung. Our method consists of neural network model based on popular U-Net architecture family but modified for the joint nodule segmentation and its texture classification tasks and an ensemble-based model for the follow-up recommendation. This solution was evaluated within the LNDb medical imaging challenge and produced the best nodule segmentation result on the final leaderboard.
Decision trees are a widely used method for classification, both by themselves and as the building blocks of multiple different ensemble learning methods. The Max-Cut decision tree involves novel modifications to a standard, baseline model of classification decision tree construction, precisely CART Gini. One modification involves an alternative splitting metric, maximum cut, based on maximizing the distance between all pairs of observations belonging to separate classes and separate sides of the threshold value. The other modification is to select the decision feature from a linear combination of the input features constructed using Principal Component Analysis (PCA) locally at each node. Our experiments show that this node-based localized PCA with the novel splitting modification can dramatically improve classification, while also significantly decreasing computational time compared to the baseline decision tree. Moreover, our results are most significant when evaluated on data sets with higher dimensions, or more classes; which, for the example data set CIFAR-100, enable a 49% improvement in accuracy while reducing CPU time by 94%. These introduced modifications dramatically advance the capabilities of decision trees for difficult classification tasks.