Understanding how deep convolutional neural networks classify data has been subject to extensive research. This paper proposes a technique to visualize and interpret intermediate layers of unsupervised deep convolutional networks by averaging over individual feature maps in each convolutional layer and inferring underlying distributions of words with non-linear regression techniques. A GAN-based architecture (ciwGAN arXiv:2006.02951) that includes a Generator, a Discriminator, and a classifier was trained on unlabeled sliced lexical items from TIMIT. The training process results in a deep convolutional network that learns to classify words into discrete classes only from the requirement of the Generator to output informative data. This classifier network has no access to the training data -- only to the generated data. We propose a technique to visualize individual convolutional layers in the classifier that yields highly informative time-series data for each convolutional layer and apply it to unobserved test data. Using non-linear regression, we infer underlying distributions for each word which allows us to analyze both absolute values and shapes of individual words at different convolutional layers, as well as perform hypothesis testing on their acoustic properties. The technique also allows us to test individual phone contrasts and how they are represented at each layer.
Network-valued data are encountered in a wide range of applications and pose challenges in learning due to their complex structure and absence of vertex correspondence. Typical examples of such problems include classification or grouping of protein structures and social networks. Various methods, ranging from graph kernels to graph neural networks, have been proposed that achieve some success in graph classification problems. However, most methods have limited theoretical justification, and their applicability beyond classification remains unexplored. In this work, we propose methods for clustering multiple graphs, without vertex correspondence, that are inspired by the recent literature on estimating graphons -- symmetric functions corresponding to infinite vertex limit of graphs. We propose a novel graph distance based on sorting-and-smoothing graphon estimators. Using the proposed graph distance, we present two clustering algorithms and show that they achieve state-of-the-art results. We prove the statistical consistency of both algorithms under Lipschitz assumptions on the graph degrees. We further study the applicability of the proposed distance for graph two-sample testing problems.
Contrastive representation learning encourages data representation to make semantically similar pairs closer than randomly drawn negative samples, which has been successful in various domains such as vision, language, and graphs. Recent theoretical studies have attempted to explain the benefit of the large negative sample size by upper-bounding the downstream classification loss with the contrastive loss. However, the previous surrogate bounds have two drawbacks: they are only legitimate for a limited range of negative sample sizes and prohibitively large even within that range. Due to these drawbacks, there still does not exist a consensus on how negative sample size theoretically correlates with downstream classification performance. Following the simplified setting where positive pairs are drawn from the true distribution (not generated by data augmentation; as supposed in previous studies), this study establishes surrogate upper and lower bounds for the downstream classification loss for all negative sample sizes that best explain the empirical observations on the negative sample size in the earlier studies. Our bounds suggest that the contrastive loss can be viewed as a surrogate objective of the downstream loss and larger negative sample sizes improve downstream classification because the surrogate gap between contrastive and supervised losses decays. We verify that our theory is consistent with experiments on synthetic, vision, and language datasets.
The recent ongoing coronavirus pandemic highlights the importance of hand hygiene practices in our daily lives, with governments and worldwide health authorities promoting good hand hygiene practices. More than one million cases of hospital-acquired infections occur in Europe annually. Hand hygiene compliance may reduce the risk of transmission by reducing the number of infections as well as healthcare expenditures. In this paper, the World Health Organization, hand hygiene gestures are recorded and analyzed with the construction of an aluminum frame, placed at the laboratory sink. The hand hygiene gestures are recorded for thirty participants after conducting a training session about hand hygiene gestures demonstration. The video recordings are converted into image files and are organized into six different hand hygiene classes. The Resnet50 framework selection for the classification of multiclass hand hygiene stages. The model is trained with the first set of classes; Fingers Interlaced, P2PFingers Interlaced, and Rotational Rub for 25 epochs. An accuracy of 44 percent for the first set of experiments with a loss score greater than 1.5 in the validation set is achieved. The training steps for the second set of classes; Rub hands palm to palm, Fingers Interlocked, Thumb Rub are 50 epochs. An accuracy of 72 percent is achieved for the second set with a loss score of less than 0.8 for the validation set. In this work, a preliminary analysis of a robust hand hygiene dataset with transfer learning takes place. The future aim for deploying a hand hygiene prediction system for healthcare workers in real-time.
Efficiently implementing remote sensing image classification with high spatial resolution imagery can provide a significant value in Land Use and Land Cover (LULC) classification. The new advances in remote sensing and deep learning technologies have facilitated the extraction of spatiotemporal information for LULC classification. Moreover, the diverse disciplines of science, including remote sensing, have utilised tremendous improvements in image classification by Convolutional Neural Networks (CNNs) with transfer learning. In this study, instead of training CNNs from scratch, the transfer learning is applied to fine-tune pre-trained networks Visual Geometry Group (VGG16) and Wide Residual Networks (WRNs), by replacing the final layer with additional layers, for LULC classification using the red-green-blue version of the EuroSAT dataset. Moreover, the performance and computational time are compared and optimised with techniques, such as early stopping, gradient clipping, adaptive learning rates, and data augmentation. The proposed approaches have addressed the limited-data problem, and very good accuracies are achieved. The results show that the proposed method based on the WRNs performs better than the previous best-stated results in terms of the computational efficiency and accuracy from 98.57% to 99.17%.
Globally, cataract is a common eye disease and one of the leading causes of blindness and vision impairment. The traditional process of detecting cataracts involves eye examination using a slit-lamp microscope or ophthalmoscope by an ophthalmologist, who checks for clouding of the normally clear lens of the eye. The lack of resources and unavailability of a sufficient number of experts pose a burden to the healthcare system throughout the world, and researchers are exploring the use of AI solutions for assisting the experts. Inspired by the progress in iris recognition, in this research, we present a novel algorithm for cataract detection using near-infrared eye images. The NIR cameras, which are popularly used in iris recognition, are of relatively low cost and easy to operate compared to ophthalmoscope setup for data capture. However, such NIR images have not been explored for cataract detection. We present deep learning-based eye segmentation and multitask network classification networks for cataract detection using NIR images as input. The proposed segmentation algorithm efficiently and effectively detects non-ideal eye boundaries and is cost-effective, and the classification network yields very high classification performance on the cataract dataset.
In this paper, we propose a balancing training method to address problems in imbalanced data learning. To this end, we derive a new loss used in the balancing training phase that alleviates the influence of samples that cause an overfitted decision boundary. The proposed loss efficiently improves the performance of any type of imbalance learning methods. In experiments on multiple benchmark data sets, we demonstrate the validity of our method and reveal that the proposed loss outperforms the state-of-the-art cost-sensitive loss methods. Furthermore, since our loss is not restricted to a specific task, model, or training method, it can be easily used in combination with other recent re-sampling, meta-learning, and cost-sensitive learning methods for class-imbalance problems.
Reflective and textureless surfaces such as windows, mirrors, and walls can be a challenge for object and scene reconstruction. These surfaces are often poorly reconstructed and filled with depth discontinuities and holes, making it difficult to cohesively reconstruct scenes that contain these planar discontinuities. We propose Echoreconstruction, an audio-visual method that uses the reflections of sound to aid in geometry and audio reconstruction for virtual conferencing, teleimmersion, and other AR/VR experience. The mobile phone prototype emits pulsed audio, while recording video for RGB-based 3D reconstruction and audio-visual classification. Reflected sound and images from the video are input into our audio (EchoCNN-A) and audio-visual (EchoCNN-AV) convolutional neural networks for surface and sound source detection, depth estimation, and material classification. The inferences from these classifications enhance scene 3D reconstructions containing open spaces and reflective surfaces by depth filtering, inpainting, and placement of unmixed sound sources in the scene. Our prototype, VR demo, and experimental results from real-world and virtual scenes with challenging surfaces and sound indicate high success rates on classification of material, depth estimation, and closed/open surfaces, leading to considerable visual and audio improvement in 3D scenes (see Figure 1).
Diabetes is a raising problem that affects many people globally. Diabetic patients are at risk of developing foot ulcer that usually leads to limb amputation, causing significant morbidity, and psychological distress. In order to develop a self monitoring mobile application, it is necessary to be able to classify such ulcers into either of the following classes: Infection, Ischaemia, None, or Both. In this work, we compare the performance of a classical transfer-learning-based method, with the performance of a hybrid classical-quantum Classifier on diabetic foot ulcer classification task. As such, we merge the pre-trained Xception network with a multi-class variational classifier. Thus, after modifying and re-training the Xception network, we extract the output of a mid-layer and employ it as deep-features presenters of the given images. Finally, we use those deep-features to train multi-class variational classifier, where each classifier is implemented on an individual variational circuit. The method is then evaluated on the blind test set DFUC2021. The results proves that our proposed hybrid classical-quantum Classifier leads to considerable improvement compared to solely relying on transfer learning concept through training the modified version of Xception network.
Plant breeding programs extensively monitor the evolution of seed kernels for seed certification, wherein lies the need to appropriately label the seed kernels by type and quality. However, the breeding environments are large where the monitoring of seed kernels can be challenging due to the minuscule size of seed kernels. The use of unmanned aerial vehicles aids in seed monitoring and labeling since they can capture images at low altitudes whilst being able to access even the remotest areas in the environment. A key bottleneck in the labeling of seeds using UAV imagery is drone altitude i.e. the classification accuracy decreases as the altitude increases due to lower image detail. Convolutional neural networks are a great tool for multi-class image classification when there is a training dataset that closely represents the different scenarios that the network might encounter during evaluation. The article addresses the challenge of training data creation using Domain Randomization wherein synthetic image datasets are generated from a meager sample of seeds captured by the bottom camera of an autonomously driven Parrot AR Drone 2.0. Besides, the article proposes a seed classification framework as a proof-of-concept using the convolutional neural networks of Microsoft's ResNet-100, Oxford's VGG-16, and VGG-19. To enhance the classification accuracy of the framework, an ensemble model is developed resulting in an overall accuracy of 94.6%.
Learning accurate classifiers for novel categories from very few examples, known as few-shot image classification, is a challenging task in statistical machine learning and computer vision. The performance in few-shot classification suffers from the bias in the estimation of classifier parameters; however, an effective underlying bias reduction technique that could alleviate this issue in training few-shot classifiers has been overlooked. In this work, we demonstrate the effectiveness of Firth bias reduction in few-shot classification. Theoretically, Firth bias reduction removes the $O(N^{-1})$ first order term from the small-sample bias of the Maximum Likelihood Estimator. Here we show that the general Firth bias reduction technique simplifies to encouraging uniform class assignment probabilities for multinomial logistic classification, and almost has the same effect in cosine classifiers. We derive an easy-to-implement optimization objective for Firth penalized multinomial logistic and cosine classifiers, which is equivalent to penalizing the cross-entropy loss with a KL-divergence between the uniform label distribution and the predictions. Then, we empirically evaluate that it is consistently effective across the board for few-shot image classification, regardless of (1) the feature representations from different backbones, (2) the number of samples per class, and (3) the number of classes. Finally, we show the robustness of Firth bias reduction, in the case of imbalanced data distribution. Our implementation is available at https://github.com/ehsansaleh/firth_bias_reduction
Institutions are increasingly relying on machine learning models to identify and alert on abnormal events, such as fraud, cyber attacks and system failures. These alerts often need to be manually investigated by specialists. Given the operational cost of manual inspections, the suspicious events are selected by alerting systems with carefully designed thresholds. In this paper, we consider an imbalanced binary classification problem, where events arrive sequentially and only a limited number of suspicious events can be inspected. We model the event arrivals as a non-homogeneous Poisson process, and compare various suspicious event selection methods including those based on static and adaptive thresholds. For each method, we analytically characterize the tradeoff between the minority-class detection rate and the inspection capacity as a function of the data class imbalance and the classifier confidence score densities. We implement the selection methods on a real public fraud detection dataset and compare the empirical results with analytical bounds. Finally, we investigate how class imbalance and the choice of classifier impact the tradeoff.
Despite the proliferation of numerous deep learning methods proposed for generic ECG classification and arrhythmia detection, compact systems with the real-time ability and high accuracy for classifying patient-specific ECG are still few. Particularly, the scarcity of patient-specific data poses an ultimate challenge to any classifier. Recently, compact 1D Convolutional Neural Networks (CNNs) have achieved the state-of-the-art performance level for the accurate classification of ventricular and supraventricular ectopic beats. However, several studies have demonstrated the fact that the learning performance of the conventional CNNs is limited because they are homogenous networks with a basic (linear) neuron model. In order to address this deficiency and further boost the patient-specific ECG classification performance, in this study, we propose 1D Self-organized Operational Neural Networks (1D Self-ONNs). Due to its self-organization capability, Self-ONNs have the utmost advantage and superiority over conventional ONNs where the prior operator search within the operator set library to find the best possible set of operators is entirely avoided. As the first study where 1D Self-ONNs are ever proposed for a classification task, our results over the MIT-BIH arrhythmia benchmark database demonstrate that 1D Self-ONNs can surpass 1D CNNs with a significant margin while having a similar computational complexity. Under AAMI recommendations and with minimal common training data used, over the entire MIT-BIH dataset 1D Self-ONNs have achieved 98% and 99.04% average accuracies, 76.6% and 93.7% average F1 scores on supra-ventricular and ventricular ectopic beat (VEB) classifications, respectively, which is the highest performance level ever reported.
A new recursive construction of $N$-ary error-correcting output code (ECOC) matrices for ensemble classification methods is presented, generalizing the classic doubling construction for binary Hadamard matrices. Given any prime integer $N$, this deterministic construction generates base-$N$ symmetric square matrices $M$ of prime-power dimension having optimal minimum Hamming distance between any two of its rows and columns. Experimental results for six datasets demonstrate that using these deterministic coding matrices for $N$-ary ECOC classification yields comparable and in many cases higher accuracy compared to using randomly generated coding matrices. This is particular true when $N$ is adaptively chosen so that the dimension of $M$ matches closely with the number of classes in a dataset, which reduces the loss in minimum Hamming distance when $M$ is truncated to fit the dataset. This is verified through a distance formula for $M$ which shows that these adaptive matrices have significantly higher minimum Hamming distance in comparison to randomly generated ones.
We study the classification problem for high-dimensional data with $n$ observations on $p$ features where the $p \times p$ covariance matrix $\Sigma$ exhibits a spiked eigenvalues structure and the vector $\zeta$, given by the difference between the whitened mean vectors, is sparse with sparsity at most $s$. We propose an adaptive classifier (adaptive with respect to the sparsity $s$) that first performs dimension reduction on the feature vectors prior to classification in the dimensionally reduced space, i.e., the classifier whitened the data, then screen the features by keeping only those corresponding to the $s$ largest coordinates of $\zeta$ and finally apply Fisher linear discriminant on the selected features. Leveraging recent results on entrywise matrix perturbation bounds for covariance matrices, we show that the resulting classifier is Bayes optimal whenever $n \rightarrow \infty$ and $s \sqrt{n^{-1} \ln p} \rightarrow 0$. Experimental results on real and synthetic data sets indicate that the proposed classifier is competitive with existing state-of-the-art methods while also selecting a smaller number of features.
In the context of the EU Horizon 2020 GRACIOUS project, we proposed a quantitative Weight of Evidence (WoE) approach for hazard classification of nanomaterials (NMs). This approach is based on the requirements of the European Regulation on Classification, Labelling and Packaging of Substances and Mixtures (the CLP Regulation), which implements the United Nations' Globally Harmonized System of Classification and Labelling of Chemicals (UN GHS) in the European Union. The goal of this WoE methodology is to facilitate classification of NMs according to CLP criteria, following the decision trees defined in ECHA's CLP regulatory guidance. The proposed methodology involves the following stages: (1) collection of data for different NMs related to the endpoint of interest: each study related to each NM is referred as a Line of Evidence (LoE); (2) computation of weighted scores for each LoE: each LoE is weighted by a score calculated based on agreed data quality and completeness criteria defined in the GRACIOUS project; (3) comparison and integration of the weighed LoEs for each NM: A Monte Carlo resampling approach is adopted to quantitatively and probabilistically integrate the weighted evidence; and (4) assignment of each NM to a hazard class: according to the results, each NM is assigned to one of the classes defined by the CLP regulation. Furthermore, to facilitate the integration and the classification of the weighted LoEs, an R tool was developed. Finally, the approach was tested against an endpoint relevant to CLP (Aquatic Toxicity) using data retrieved from the eNanoMapper database, results obtained were consistent to results in ECHA registration dossiers and in recent literature
We consider the problem of linear classification under general loss functions in the limited-data setting. Overfitting is a common problem here. The standard approaches to prevent overfitting are dimensionality reduction and regularization. But dimensionality reduction loses information, while regularization requires the user to choose a norm, or a prior, or a distance metric. We propose an algorithm called RoLin that needs no user choice and applies to a large class of loss functions. RoLin combines reliable information from the top principal components with a robust optimization to extract any useful information from unreliable subspaces. It also includes a new robust cross-validation that is better than existing cross-validation methods in the limited-data setting. Experiments on $25$ real-world datasets and three standard loss functions show that RoLin broadly outperforms both dimensionality reduction and regularization. Dimensionality reduction has $14\%-40\%$ worse test loss on average as compared to RoLin. Against $L_1$ and $L_2$ regularization, RoLin can be up to 3x better for logistic loss and 12x better for squared hinge loss. The differences are greatest for small sample sizes, where RoLin achieves the best loss on 2x to 3x more datasets than any competing method. For some datasets, RoLin with $15$ training samples is better than the best norm-based regularization with $1500$ samples.
Text classification is a primary task in natural language processing (NLP). Recently, graph neural networks (GNNs) have developed rapidly and been applied to text classification tasks. As a special kind of graph data, the tree has a simpler data structure and can provide rich hierarchical information for text classification. Inspired by the structural entropy, we construct the coding tree of the graph by minimizing the structural entropy and propose HINT, which aims to make full use of the hierarchical information contained in the text for the task of text classification. Specifically, we first establish a dependency parsing graph for each text. Then we designed a structural entropy minimization algorithm to decode the key information in the graph and convert each graph to its corresponding coding tree. Based on the hierarchical structure of the coding tree, the representation of the entire graph is obtained by updating the representation of non-leaf nodes in the coding tree layer by layer. Finally, we present the effectiveness of hierarchical information in text classification. Experimental results show that HINT outperforms the state-of-the-art methods on popular benchmarks while having a simple structure and few parameters.
In this paper, we propose a general framework for mitigating the disparities of the predicted classes with respect to secondary attributes within the data (e.g., race, gender etc.). Our proposed method involves learning a multi-objective function that in addition to learning the primary objective of predicting the primary class labels from the data, also employs a clustering-based heuristic to minimize the disparities of the class label distribution with respect to the cluster memberships, with the assumption that each cluster should ideally map to a distinct combination of attribute values. Experiments demonstrate effective mitigation of cognitive biases on a benchmark dataset without the use of annotations of secondary attribute values (the zero-shot case) or with the use of a small number of attribute value annotations (the few-shot case).
When compared to the image classification models, black-box adversarial attacks against video classification models have been largely understudied. This could be possible because, with video, the temporal dimension poses significant additional challenges in gradient estimation. Query-efficient black-box attacks rely on effectively estimated gradients towards maximizing the probability of misclassifying the target video. In this work, we demonstrate that such effective gradients can be searched for by parameterizing the temporal structure of the search space with geometric transformations. Specifically, we design a novel iterative algorithm Geometric TRAnsformed Perturbations (GEO-TRAP), for attacking video classification models. GEO-TRAP employs standard geometric transformation operations to reduce the search space for effective gradients into searching for a small group of parameters that define these operations. This group of parameters describes the geometric progression of gradients, resulting in a reduced and structured search space. Our algorithm inherently leads to successful perturbations with surprisingly few queries. For example, adversarial examples generated from GEO-TRAP have better attack success rates with ~73.55% fewer queries compared to the state-of-the-art method for video adversarial attacks on the widely used Jester dataset. Overall, our algorithm exposes vulnerabilities of diverse video classification models and achieves new state-of-the-art results under black-box settings on two large datasets. Code is available here: https://github.com/sli057/Geo-TRAP