Graph-based semi-supervised node classification (GraphSSC) has wide applications, ranging from networking and security to data mining and machine learning, etc. However, existing centralized GraphSSC methods are impractical to solve many real-world graph-based problems, as collecting the entire graph and labeling a reasonable number of labels is time-consuming and costly, and data privacy may be also violated. Federated learning (FL) is an emerging learning paradigm that enables collaborative learning among multiple clients, which can mitigate the issue of label scarcity and protect data privacy as well. Therefore, performing GraphSSC under the FL setting is a promising solution to solve real-world graph-based problems. However, existing FL methods 1) perform poorly when data across clients are non-IID, 2) cannot handle data with new label domains, and 3) cannot leverage unlabeled data, while all these issues naturally happen in real-world graph-based problems. To address the above issues, we propose the first FL framework, namely GraphFL, for semi-supervised node classification on graphs. Our framework is motivated by meta-learning methods. Specifically, we propose two GraphFL methods to respectively address the non-IID issue in graph data and handle the tasks with new label domains. Furthermore, we design a self-training method to leverage unlabeled graph data. We adopt representative graph neural networks as GraphSSC methods and evaluate GraphFL on multiple graph datasets. Experimental results demonstrate that GraphFL significantly outperforms the compared FL baseline and GraphFL with self-training can obtain better performance.
This article explores search strategies for the design of parameterized quantum circuits. We propose several optimization approaches including random search plus survival of the fittest, reinforcement learning both with classical and hybrid quantum classical controllers and Bayesian optimization as decision makers to design a quantum circuit in an automated way for a specific task such as multi-labeled classification over a dataset. We introduce nontrivial circuit architectures that are arduous to be hand-designed and efficient in terms of trainability. In addition, we introduce reuploading of initial data into quantum circuits as an option to find more general designs. We numerically show that some of the suggested architectures for the Iris dataset accomplish better results compared to the established parameterized quantum circuit designs in the literature. In addition, we investigate the trainability of these structures on the unseen dataset Glass. We report meaningful advantages over the benchmarks for the classification of the Glass dataset which supports the fact that the suggested designs are inherently more trainable.
Land-cover classification using remote sensing imagery is an important Earth observation task. Recently, land cover classification has benefited from the development of fully connected neural networks for semantic segmentation. The benchmark datasets available for training deep segmentation models in remote sensing imagery tend to be small, however, often consisting of only a handful of images from a single location with a single scale. This limits the models' ability to generalize to other datasets. Domain adaptation has been proposed to improve the models' generalization but we find these approaches are not effective for dealing with the scale variation commonly found between remote sensing image collections. We therefore propose a scale aware adversarial learning framework to perform joint cross-location and cross-scale land-cover classification. The framework has a dual discriminator architecture with a standard feature discriminator as well as a novel scale discriminator. We also introduce a scale attention module which produces scale-enhanced features. Experimental results show that the proposed framework outperforms state-of-the-art domain adaptation methods by a large margin.
In recent years, deep learning has dominated progress in the field of medical image analysis. We find however, that the ability of current deep learning approaches to represent the complex geometric structures of many medical images is insufficient. One limitation is that deep learning models require a tremendous amount of data, and it is very difficult to obtain a sufficient amount with the necessary detail. A second limitation is that there are underlying features of these medical images that are well established, but the black-box nature of existing convolutional neural networks (CNNs) do not allow us to exploit them. In this paper, we revisit Gabor filters and introduce a deformable Gabor convolution (DGConv) to expand deep networks interpretability and enable complex spatial variations. The features are learned at deformable sampling locations with adaptive Gabor convolutions to improve representativeness and robustness to complex objects. The DGConv replaces standard convolutional layers and is easily trained end-to-end, resulting in deformable Gabor feature network (DGFN) with few additional parameters and minimal additional training cost. We introduce DGFN for addressing deep multi-instance multi-label classification on the INbreast dataset for mammograms and on the ChestX-ray14 dataset for pulmonary x-ray images.
A combination of cloud-based deep learning (DL) algorithms with portable/wearable (P/W) devices has been developed as a smart heath care system to support automatic cardiac arrhythmias (CAs) classification using electrocardiography (ECG). However, long-term and continuous ECG monitoring is challenging because of limitations of batteries and transmission bandwidth of P/W devices while incorporated with consumer electronics (CE). A feasible approach to address this challenge is to decrease sampling rates. However, low sampling rates lead to low-resolution signals that hinder the CAs classification performance. In this study, we propose a DL-based ECG signal super-resolution framework (called SRECG) to enhance low-resolution ECG signals by jointly considering the accuracies when applied to the DL-based high-resolution multiclass classifier (HMC) of CAs. In our experiments, we downsampled the ECG signals from the CPSC2018 dataset and evaluated their HMC accuracies with and without the SRECG. Experimental results show that SRECG can well improve the HMC accuracies as compared to traditional interpolation methods. Moreover, approximately half of the CAs classification accuracies of HMC were maintained within the enhanced ECG signals by SRECG. The promising results confirm that SRECG can be suitably used to enhance low-resolution ECG signals from P/W devices with CE to improve their cloud-based HMC performances.
Increasing use of ML technologies in privacy-sensitive domains such as medical diagnoses, lifestyle predictions, and business decisions highlights the need to better understand if these ML technologies are introducing leakages of sensitive and proprietary training data. In this paper, we focus on one kind of model inversion attacks, where the adversary knows non-sensitive attributes about instances in the training data and aims to infer the value of a sensitive attribute unknown to the adversary, using oracle access to the target classification model. We devise two novel model inversion attribute inference attacks -- confidence modeling-based attack and confidence score-based attack, and also extend our attack to the case where some of the other (non-sensitive) attributes are unknown to the adversary. Furthermore, while previous work uses accuracy as the metric to evaluate the effectiveness of attribute inference attacks, we find that accuracy is not informative when the sensitive attribute distribution is unbalanced. We identify two metrics that are better for evaluating attribute inference attacks, namely G-mean and Matthews correlation coefficient (MCC). We evaluate our attacks on two types of machine learning models, decision tree and deep neural network, trained with two real datasets. Experimental results show that our newly proposed attacks significantly outperform the state-of-the-art attacks. Moreover, we empirically show that specific groups in the training dataset (grouped by attributes, e.g., gender, race) could be more vulnerable to model inversion attacks. We also demonstrate that our attacks' performances are not impacted significantly when some of the other (non-sensitive) attributes are also unknown to the adversary.
Decoupled learning is a branch of model parallelism which parallelizes the training of a network by splitting it depth-wise into multiple modules. Techniques from decoupled learning usually lead to stale gradient effect because of their asynchronous implementation, thereby causing performance degradation. In this paper, we propose an accumulated decoupled learning (ADL) which incorporates the gradient accumulation technique to mitigate the stale gradient effect. We give both theoretical and empirical evidences regarding how the gradient staleness can be reduced. We prove that the proposed method can converge to critical points, i.e., the gradients converge to 0, in spite of its asynchronous nature. Empirical validation is provided by training deep convolutional neural networks to perform classification tasks on CIFAR-10 and ImageNet datasets. The ADL is shown to outperform several state-of-the-arts in the classification tasks, and is the fastest among the compared methods.
Dealing with land cover classification of the new image sources has also turned to be a complex problem requiring large amount of memory and processing time. In order to cope with these problems, statistical learning has greatly helped in the last years to develop statistical retrieval and classification models that can ingest large amounts of Earth observation data. Kernel methods constitute a family of powerful machine learning algorithms, which have found wide use in remote sensing and geosciences. However, kernel methods are still not widely adopted because of the high computational cost when dealing with large scale problems, such as the inversion of radiative transfer models or the classification of high spatial-spectral-temporal resolution data. This paper introduces an efficient kernel method for fast statistical retrieval of bio-geo-physical parameters and image classification problems. The method allows to approximate a kernel matrix with a set of projections on random bases sampled from the Fourier domain. The method is simple, computationally very efficient in both memory and processing costs, and easily parallelizable. We show that kernel regression and classification is now possible for datasets with millions of examples and high dimensionality. Examples on atmospheric parameter retrieval from hyperspectral infrared sounders like IASI/Metop; large scale emulation and inversion of the familiar PROSAIL radiative transfer model on Sentinel-2 data; and the identification of clouds over landmarks in time series of MSG/Seviri images show the efficiency and effectiveness of the proposed technique.
Despite the popularity of guitar effects, there is very little existing research on classification and parameter estimation of specific plugins or effect units from guitar recordings. In this paper, convolutional neural networks were used for classification and parameter estimation for 13 overdrive, distortion and fuzz guitar effects. A novel dataset of processed electric guitar samples was assembled, with four sub-datasets consisting of monophonic or polyphonic samples and discrete or continuous settings values, for a total of about 250 hours of processed samples. Results were compared for networks trained and tested on the same or on a different sub-dataset. We found that discrete datasets could lead to equally high performance as continuous ones, whilst being easier to design, analyse and modify. Classification accuracy was above 80\%, with confusion matrices reflecting similarities in the effects timbre and circuits design. With parameter values between 0.0 and 1.0, the mean absolute error is in most cases below 0.05, while the root mean square error is below 0.1 in all cases but one.
Breast cancer is the second leading cause of cancer-related death after lung cancer in women. Early detection of breast cancer in X-ray mammography is believed to have effectively reduced the mortality rate. However, a relatively high false positive rate and a low specificity in mammography technology still exist. In this work, a computer-aided automatic mammogram analysis system is proposed to process the mammogram images and automatically discriminate them as either normal or cancerous, consisting of three consecutive image processing, feature selection, and image classification stages. In designing the system, the discrete wavelet transforms (Daubechies 2, Daubechies 4, and Biorthogonal 6.8) and the Fourier cosine transform were first used to parse the mammogram images and extract statistical features. Then, an entropy-based feature selection method was implemented to reduce the number of features. Finally, different pattern recognition methods (including the Back-propagation Network, the Linear Discriminant Analysis, and the Naive Bayes Classifier) and a voting classification scheme were employed. The performance of each classification strategy was evaluated for sensitivity, specificity, and accuracy and for general performance using the Receiver Operating Curve. Our method is validated on the dataset from the Eastern Health in Newfoundland and Labrador of Canada. The experimental results demonstrated that the proposed automatic mammogram analysis system could effectively improve the classification performances.
Contrastive learning (CL) has been successful as a powerful representation learning method. In this paper, we propose a contrastive learning framework for cross-domain sentiment classification. We aim to induce domain invariant optimal classifiers rather than distribution matching. To this end, we introduce in-domain contrastive learning and entropy minimization. Also, we find through ablation studies that these two techniques behaviour differently in case of large label distribution shift and conclude that the best practice is to choose one of them adaptively according to label distribution shift. The new state-of-the-art results our model achieves on standard benchmarks show the efficacy of the proposed method.
It is argued in [1] that [2] was able to classify EEG responses to visual stimuli solely because of the temporal correlation that exists in all EEG data and the use of a block design. We here show that the main claim in [1] is drastically overstated and their other analyses are seriously flawed by wrong methodological choices. To validate our counter-claims, we evaluate the performance of state-of-the-art methods on the dataset in [2] reaching about 50% classification accuracy over 40 classes, lower than in [2], but still significant. We then investigate the influence of EEG temporal correlation on classification accuracy by testing the same models in two additional experimental settings: one that replicates [1]'s rapid-design experiment, and another one that examines the data between blocks while subjects are shown a blank screen. In both cases, classification accuracy is at or near chance, in contrast to what [1] reports, indicating a negligible contribution of temporal correlation to classification accuracy. We, instead, are able to replicate the results in [1] only when intentionally contaminating our data by inducing a temporal correlation. This suggests that what Li et al. [1] demonstrate is that their data are strongly contaminated by temporal correlation and low signal-to-noise ratio. We argue that the reason why Li et al. [1] observe such high correlation in EEG data is their unconventional experimental design and settings that violate the basic cognitive neuroscience design recommendations, first and foremost the one of limiting the experiments' duration, as instead done in [2]. Our analyses in this paper refute the claims of the "perils and pitfalls of block-design" in [1]. Finally, we conclude the paper by examining a number of other oversimplistic statements, inconsistencies, misinterpretation of machine learning concepts, speculations and misleading claims in [1].
The pressing need to reduce the capacity of deep neural networks has stimulated the development of network dilution methods and their analysis. While the ability of $L_1$ and $L_0$ regularization to encourage sparsity is often mentioned, $L_2$ regularization is seldom discussed in this context. We present a novel framework for weight pruning by sampling from a probability function that favors the zeroing of smaller weights. In addition, we examine the contribution of $L_1$ and $L_2$ regularization to the dynamics of node pruning while optimizing for weight pruning. We then demonstrate the effectiveness of the proposed stochastic framework when used together with a weight decay regularizer on popular classification models in removing 50% of the nodes in an MLP for MNIST classification, 60% of the filters in VGG-16 for CIFAR10 classification, and on medical image models in removing 60% of the channels in a U-Net for instance segmentation and 50% of the channels in CNN model for COVID-19 detection. For these node-pruned networks, we also present competitive weight pruning results that are only slightly less accurate than the original, dense networks.
The quantification of emotional states is an important step to understanding wellbeing. Time series data from multiple modalities such as physiological and motion sensor data have proven to be integral for measuring and quantifying emotions. Monitoring emotional trajectories over long periods of time inherits some critical limitations in relation to the size of the training data. This shortcoming may hinder the development of reliable and accurate machine learning models. To address this problem, this paper proposes a framework to tackle the limitation in performing emotional state recognition on multiple multimodal datasets: 1) encoding multivariate time series data into coloured images; 2) leveraging pre-trained object recognition models to apply a Transfer Learning (TL) approach using the images from step 1; 3) utilising a 1D Convolutional Neural Network (CNN) to perform emotion classification from physiological data; 4) concatenating the pre-trained TL model with the 1D CNN. Furthermore, the possibility of performing TL to infer stress from physiological data is explored by initially training a 1D CNN using a large physical activity dataset and then applying the learned knowledge to the target dataset. We demonstrate that model performance when inferring real-world wellbeing rated on a 5-point Likert scale can be enhanced using our framework, resulting in up to 98.5% accuracy, outperforming a conventional CNN by 4.5%. Subject-independent models using the same approach resulted in an average of 72.3% accuracy (SD 0.038). The proposed CNN-TL-based methodology may overcome problems with small training datasets, thus improving on the performance of conventional deep learning methods.
The study of strategic or adversarial manipulation of testing data to fool a classifier has attracted much recent attention. Most previous works have focused on two extreme situations where any testing data point either is completely adversarial or always equally prefers the positive label. In this paper, we generalize both of these through a unified framework for strategic classification, and introduce the notion of strategic VC-dimension (SVC) to capture the PAC-learnability in our general strategic setup. SVC provably generalizes the recent concept of adversarial VC-dimension (AVC) introduced by Cullina et al. arXiv:1806.01471. We instantiate our framework for the fundamental strategic linear classification problem. We fully characterize: (1) the statistical learnability of linear classifiers by pinning down its SVC; (2) its computational tractability by pinning down the complexity of the empirical risk minimization problem. Interestingly, the SVC of linear classifiers is always upper bounded by its standard VC-dimension. This characterization also strictly generalizes the AVC bound for linear classifiers in arXiv:1806.01471.
Deep AUC Maximization (DAM) is a new paradigm for learning a deep neural network by maximizing the AUC score of the model on a dataset. Most previous works of AUC maximization focus on the perspective of optimization by designing efficient stochastic algorithms, and studies on generalization performance of large-scale DAM on difficult tasks are missing. In this work, we aim to make DAM more practical for interesting real-world applications (e.g., medical image classification). First, we propose a new margin-based min-max surrogate loss function for the AUC score (named as AUC min-max-margin loss or simply AUC margin loss for short). It is more robust than the commonly used AUC square loss, while enjoying the same advantage in terms of large-scale stochastic optimization. Second, we conduct extensive empirical studies of our DAM method on four difficult medical image classification tasks, namely (i) classification of chest x-ray images for identifying many threatening diseases, (ii) classification of images of skin lesions for identifying melanoma, (iii) classification of mammogram for breast cancer screening, and (iv) classification of microscopic images for identifying tumor tissue. Our studies demonstrate that the proposed DAM method improves the performance of optimizing cross-entropy loss by a large margin, and also achieves better performance than optimizing the existing AUC square loss on these medical image classification tasks. Specifically, our DAM method has achieved the 1st place on Stanford CheXpert competition on Aug. 31, 2020. To the best of our knowledge, this is the first work that makes DAM succeed on large-scale medical image datasets. We also conduct extensive ablation studies to demonstrate the advantages of the new AUC margin loss over the AUC square loss on benchmark datasets. The proposed method is implemented in our open-sourced library LibAUC (www.libauc.org).
The ability to interpret decisions taken by Machine Learning (ML) models is fundamental to encourage trust and reliability in different practical applications. Recent interpretation strategies focus on human understanding of the underlying decision mechanisms of the complex ML models. However, these strategies are restricted by the subjective biases of humans. To dissociate from such human biases, we propose an interpretation-by-distillation formulation that is defined relative to other ML models. We generalize the distillation technique for quantifying interpretability, using an information-theoretic perspective, removing the role of ground-truth from the definition of interpretability. Our work defines the entropy of supervised classification models, providing bounds on the entropy of Piece-Wise Linear Neural Networks (PWLNs), along with the first theoretical bounds on the interpretability of PWLNs. We evaluate our proposed framework on the MNIST, Fashion-MNIST and Stanford40 datasets and demonstrate the applicability of the proposed theoretical framework in different supervised classification scenarios.
Multi-task learning is widely used in computer vision. Currently, object detection models utilize shared feature map to complete classification and localization tasks simultaneously. By comparing the performance between the original Faster R-CNN and that with partially separated feature maps, we show that: (1) Sharing high-level features for the classification and localization tasks is sub-optimal; (2) Large stride is beneficial for classification but harmful for localization; (3) Global context information could improve the performance of classification. Based on these findings, we proposed a paradigm called Gap-optimized region based convolutional network (G-RCN), which aims to separating these two tasks and optimizing the gap between them. The paradigm was firstly applied to correct the current ResNet protocol by simply reducing the stride and moving the Conv5 block from the head to the feature extraction network, which brings 3.6 improvement of AP70 on the PASCAL VOC dataset and 1.5 improvement of AP on the COCO dataset for ResNet50. Next, the new method is applied on the Faster R-CNN with backbone of VGG16,ResNet50 and ResNet101, which brings above 2.0 improvement of AP70 on the PASCAL VOC dataset and above 1.9 improvement of AP on the COCO dataset. Noticeably, the implementation of G-RCN only involves a few structural modifications, with no extra module added.
Deep learning-based methods have achieved promising performance in early detection and classification of lung nodules, most of which discard unsure nodules and simply deal with a binary classification -- malignant vs benign. Recently, an unsure data model (UDM) was proposed to incorporate those unsure nodules by formulating this problem as an ordinal regression, showing better performance over traditional binary classification. To further explore the ordinal relationship for lung nodule classification, this paper proposes a meta ordinal regression forest (MORF), which improves upon the state-of-the-art ordinal regression method, deep ordinal regression forest (DORF), in three major ways. First, MORF can alleviate the biases of the predictions by making full use of deep features while DORF needs to fix the composition of decision trees before training. Second, MORF has a novel grouped feature selection (GFS) module to re-sample the split nodes of decision trees. Last, combined with GFS, MORF is equipped with a meta learning-based weighting scheme to map the features selected by GFS to tree-wise weights while DORF assigns equal weights for all trees. Experimental results on the LIDC-IDRI dataset demonstrate superior performance over existing methods, including the state-of-the-art DORF.
State-of-the-art video action classifiers often suffer from overfitting. They tend to be biased towards specific objects and scene cues, rather than the foreground action content, leading to sub-optimal generalization performances. Recent data augmentation strategies have been reported to address the overfitting problems in static image classifiers. Despite the effectiveness on the static image classifiers, data augmentation has rarely been studied for videos. For the first time in the field, we systematically analyze the efficacy of various data augmentation strategies on the video classification task. We then propose a powerful augmentation strategy VideoMix. VideoMix creates a new training video by inserting a video cuboid into another video. The ground truth labels are mixed proportionally to the number of voxels from each video. We show that VideoMix lets a model learn beyond the object and scene biases and extract more robust cues for action recognition. VideoMix consistently outperforms other augmentation baselines on Kinetics and the challenging Something-Something-V2 benchmarks. It also improves the weakly-supervised action localization performance on THUMOS'14. VideoMix pretrained models exhibit improved accuracies on the video detection task (AVA).