The Information Bottleneck (IB) objective uses information theory to formulate a task-performance versus robustness trade-off. It has been successfully applied in the standard discriminative classification setting. We pose the question whether the IB can also be used to train generative likelihood models such as normalizing flows. Since normalizing flows use invertible network architectures (INNs), they are information-preserving by construction. This seems contradictory to the idea of a bottleneck. In this work, firstly, we develop the theory and methodology of IB-INNs, a class of conditional normalizing flows where INNs are trained using the IB objective: Introducing a small amount of {\em controlled} information loss allows for an asymptotically exact formulation of the IB, while keeping the INN's generative capabilities intact. Secondly, we investigate the properties of these models experimentally, specifically used as generative classifiers. This model class offers advantages such as improved uncertainty quantification and out-of-distribution detection, but traditional generative classifier solutions suffer considerably in classification accuracy. We find the trade-off parameter in the IB controls a mix of generative capabilities and accuracy close to standard classifiers. Empirically, our uncertainty estimates in this mixed regime compare favourably to conventional generative and discriminative classifiers.
Compared to sequential learning models, graph-based neural networks exhibit some excellent properties, such as ability capturing global information. In this paper, we investigate graph-based neural networks for text classification problem. A new framework TensorGCN (tensor graph convolutional networks), is presented for this task. A text graph tensor is firstly constructed to describe semantic, syntactic, and sequential contextual information. Then, two kinds of propagation learning perform on the text graph tensor. The first is intra-graph propagation used for aggregating information from neighborhood nodes in a single graph. The second is inter-graph propagation used for harmonizing heterogeneous information between graphs. Extensive experiments are conducted on benchmark datasets, and the results illustrate the effectiveness of our proposed framework. Our proposed TensorGCN presents an effective way to harmonize and integrate heterogeneous information from different kinds of graphs.
Text contained in an image carries high-level semantics that can be exploited to achieve richer image understanding. In particular, the mere presence of text provides strong guiding content that should be employed to tackle a diversity of computer vision tasks such as image retrieval, fine-grained classification, and visual question answering. In this paper, we address the problem of fine-grained classification and image retrieval by leveraging textual information along with visual cues to comprehend the existing intrinsic relation between the two modalities. The novelty of the proposed model consists of the usage of a PHOC descriptor to construct a bag of textual words along with a Fisher Vector Encoding that captures the morphology of text. This approach provides a stronger multimodal representation for this task and as our experiments demonstrate, it achieves state-of-the-art results on two different tasks, fine-grained classification and image retrieval.
While deep learning has shown promise in the domain of disease classification from medical images, models based on state-of-the-art convolutional neural network architectures often exhibit performance loss due to dataset shift. Models trained using data from one hospital system achieve high predictive performance when tested on data from the same hospital, but perform significantly worse when they are tested in different hospital systems. Furthermore, even within a given hospital system, deep learning models have been shown to depend on hospital- and patient-level confounders rather than meaningful pathology to make classifications. In order for these models to be safely deployed, we would like to ensure that they do not use confounding variables to make their classification, and that they will work well even when tested on images from hospitals that were not included in the training data. We attempt to address this problem in the context of pneumonia classification from chest radiographs. We propose an approach based on adversarial optimization, which allows us to learn more robust models that do not depend on confounders. Specifically, we demonstrate improved out-of-hospital generalization performance of a pneumonia classifier by training a model that is invariant to the view position of chest radiographs (anterior-posterior vs. posterior-anterior). Our approach leads to better predictive performance on external hospital data than both a standard baseline and previously proposed methods to handle confounding, and also suggests a method for identifying models that may rely on confounders. Code available at https://github.com/suinleelab/cxr_adv.
We consider the problem of learning a neural network classifier. Under the information bottleneck (IB) principle, we associate with this classification problem a representation learning problem, which we call "IB learning". We show that IB learning is, in fact, equivalent to a special class of the quantization problem. The classical results in rate-distortion theory then suggest that IB learning can benefit from a "vector quantization" approach, namely, simultaneously learning the representations of multiple input objects. Such an approach assisted with some variational techniques, result in a novel learning framework, "Aggregated Learning", for classification with neural network models. In this framework, several objects are jointly classified by a single neural network. The effectiveness of this framework is verified through extensive experiments on standard image recognition and text classification tasks.
Graph neural networks (GNN) rely on graph operations that include neural network training for various graph related tasks. Recently, several attempts have been made to apply the GNNs to functional magnetic resonance image (fMRI) data. Despite recent progresses, a common limitation is its difficulty to explain the classification results in a neuroscientifically explainable way. Here, we develop a framework for analyzing the fMRI data using the Graph Isomorphism Network (GIN), which was recently proposed as a powerful GNN for graph classification. One of the important contributions of this paper is the observation that the GIN is a dual representation of convolutional neural network (CNN) in the graph space where the shift operation is defined using the adjacency matrix. This understanding enables us to exploit CNN-based saliency map techniques for the GNN, which we tailor to the proposed GIN with one-hot encoding, to visualize the important regions of the brain. We validate our proposed framework using large-scale resting-state fMRI (rs-fMRI) data for classifying the sex of the subject based on the graph structure of the brain. The experiment was consistent with our expectation such that the obtained saliency map show high correspondence with previous neuroimaging evidences related to sex differences.
In recent years, Deep Learning(DL) techniques have been extensively deployed for computer vision tasks, particularly visual classification problems, where new algorithms reported to achieve or even surpass the human performance. While many recent works demonstrated that DL models are vulnerable to adversarial examples. Fortunately, generating adversarial examples usually requires white-box access to the victim model, and real-world cloud-based image classification services are more complex than white-box classifier,the architecture and parameters of DL models on cloud platforms cannot be obtained by the attacker. The attacker can only access the APIs opened by cloud platforms. Thus, keeping models in the cloud can usually give a (false) sense of security. In this paper, we mainly focus on studying the security of real-world cloud-based image classification services. Specifically, (1) We propose a novel attack method, Fast Featuremap Loss PGD (FFL-PGD) attack based on Substitution model, which achieves a high bypass rate with a very limited number of queries. Instead of millions of queries in previous studies, our method finds the adversarial examples using only two queries per image; and (2) we make the first attempt to conduct an extensive empirical study of black-box attacks against real-world cloud-based classification services. Through evaluations on four popular cloud platforms including Amazon, Google, Microsoft, Clarifai, we demonstrate that FFL-PGD attack has a success rate over 90\% among different classification services. (3) We discuss the possible defenses to address these security challenges in cloud-based classification services. Our defense technology is mainly divided into model training stage and image preprocessing stage.
Dimensionality reduction is an important operation in information visualization, feature extraction, clustering, regression, and classification, especially for processing noisy high dimensional data. However, most existing approaches preserve either the global or the local structure of the data, but not both. Approaches that preserve only the global data structure, such as principal component analysis (PCA), are usually sensitive to outliers. Approaches that preserve only the local data structure, such as locality preserving projections, are usually unsupervised (and hence cannot use label information) and uses a fixed similarity graph. We propose a novel linear dimensionality reduction approach, supervised discriminative sparse PCA with adaptive neighbors (SDSPCAAN), to integrate neighborhood-free supervised discriminative sparse PCA and projected clustering with adaptive neighbors. As a result, both global and local data structures, as well as the label information, are used for better dimensionality reduction. Classification experiments on nine high-dimensional datasets validated the effectiveness and robustness of our proposed SDSPCAAN.
Spiking Neural Networks (SNNs) are brain-inspired, event-driven machine learning algorithms that have been widely recognized in producing ultra-high-energy-efficient hardware. Among existing SNNs, unsupervised SNNs based on synaptic plasticity, especially Spike-Timing-Dependent Plasticity (STDP), are considered to have great potential in imitating the learning process of the biological brain. Nevertheless, the existing STDP-based SNNs have limitations in constrained learning capability and/or slow learning speed. Most STDP-based SNNs adopted a slow-learning Fully-Connected (FC) architecture and used a sub-optimal vote-based scheme for spike decoding. In this paper, we overcome these limitations with: 1) a design of high-parallelism network architecture, inspired by the Inception module in Artificial Neural Networks (ANNs); 2) use of a Vote-for-All (VFA) decoding layer as a replacement to the standard vote-based spike decoding scheme, to reduce the information loss in spike decoding and, 3) a proposed adaptive repolarization (resetting) mechanism that accelerates SNNs' learning by enhancing spiking activities. Our experimental results on two established benchmark datasets (MNIST/EMNIST) show that our network architecture resulted in superior performance compared to the widely used FC architecture and a more advanced Locally-Connected (LC) architecture, and that our SNN achieved competitive results with state-of-the-art unsupervised SNNs (95.64%/80.11% accuracy on the MNIST/EMNISE dataset) while having superior learning efficiency and robustness against hardware damage. Our SNN achieved great classification accuracy with only hundreds of training iterations, and random destruction of large numbers of synapses or neurons only led to negligible performance degradation.
In real-world scenarios, data tends to exhibit a long-tailed distribution, which increases the difficulty of training deep networks. In this paper, we propose a novel self-paced knowledge distillation framework, termed Learning From Multiple Experts (LFME). Our method is inspired by the observation that networks trained on less imbalanced subsets of the distribution often yield better performances than their jointly-trained counterparts. We refer to these models as 'Experts', and the proposed LFME framework aggregates the knowledge from multiple 'Experts' to learn a unified student model. Specifically, the proposed framework involves two levels of adaptive learning schedules: Self-paced Expert Selection and Curriculum Instance Selection, so that the knowledge is adaptively transferred to the 'Student'. We conduct extensive experiments and demonstrate that our method is able to achieve superior performances compared to state-of-the-art methods. We also show that our method can be easily plugged into state-of-the-art long-tailed classification algorithms for further improvements.
In next-generation communications, massive machine-type communications (mMTC) induce severe burden on base stations. To address such an issue, automatic modulation classification (AMC) can help to reduce signaling overhead by blindly recognizing the modulation types without handshaking. Thus, it plays an important role in future intelligent modems. The emerging deep learning (DL) technique stores intelligence in the network, resulting in superior performance over traditional approaches. However, conventional DL-based approaches suffer from heavy training overhead, memory overhead, and computational complexity, which severely hinder practical applications for resource-limited scenarios, such as Vehicle-to-Everything (V2X) applications. Furthermore, the overhead of online retraining under time-varying fading channels has not been studied in the prior arts. In this work, an accumulated polar feature-based DL with a channel compensation mechanism is proposed to cope with the aforementioned issues. Firstly, the simulation results show that learning features from the polar domain with historical data information can approach near-optimal performance while reducing training overhead by 99.8 times. Secondly, the proposed neural network-based channel estimator (NN-CE) can learn the channel response and compensate for the distorted channel with 13% improvement. Moreover, in applying this lightweight NN-CE in a time-varying fading channel, two efficient mechanisms of online retraining are proposed, which can reduce transmission overhead and retraining overhead by 90% and 76%, respectively. Finally, the performance of the proposed approach is evaluated and compared with prior arts on a public dataset to demonstrate its great efficiency and lightness.
Nowadays, an abundance of short text is being generated that uses nonstandard writing styles influenced by regional languages. Such informal and code-switched content are under-resourced in terms of labeled datasets and language models even for popular tasks like sentiment classification. In this work, we (1) present a labeled dataset called MultiSenti for sentiment classification of code-switched informal short text, (2) explore the feasibility of adapting resources from a resource-rich language for an informal one, and (3) propose a deep learning-based model for sentiment classification of code-switched informal short text. We aim to achieve this without any lexical normalization, language translation, or code-switching indication. The performance of the proposed models is compared with three existing multilingual sentiment classification models. The results show that the proposed model performs better in general and adapting character-based embeddings yield equivalent performance while being computationally more efficient than training word-based domain-specific embeddings.
This paper has proposed a new baseline deep learning model of more benefits for image classification. Different from the convolutional neural network(CNN) practice where filters are trained by back propagation to represent different patterns of an image, we are inspired by a method called "PCANet" in "PCANet: A Simple Deep Learning Baseline for Image Classification?" to choose filter vectors from basis vectors in frequency domain like Fourier coefficients or wavelets without back propagation. Researchers have demonstrated that those basis in frequency domain can usually provide physical insights, which adds to the interpretability of the model by analyzing the frequencies selected. Besides, the training process will also be more time efficient, mathematically clear and interpretable compared with the "black-box" training process of CNN.
We propose a novel method for automatic reasoning on knowledge graphs based on debate dynamics. The main idea is to frame the task of triple classification as a debate game between two reinforcement learning agents which extract arguments -- paths in the knowledge graph -- with the goal to promote the fact being true (thesis) or the fact being false (antithesis), respectively. Based on these arguments, a binary classifier, called the judge, decides whether the fact is true or false. The two agents can be considered as sparse, adversarial feature generators that present interpretable evidence for either the thesis or the antithesis. In contrast to other black-box methods, the arguments allow users to get an understanding of the decision of the judge. Since the focus of this work is to create an explainable method that maintains a competitive predictive accuracy, we benchmark our method on the triple classification and link prediction task. Thereby, we find that our method outperforms several baselines on the benchmark datasets FB15k-237, WN18RR, and Hetionet. We also conduct a survey and find that the extracted arguments are informative for users.
In this paper, we propose Stacked DeBERT, short for Stacked Denoising Bidirectional Encoder Representations from Transformers. This novel model improves robustness in incomplete data, when compared to existing systems, by designing a novel encoding scheme in BERT, a powerful language representation model solely based on attention mechanisms. Incomplete data in natural language processing refer to text with missing or incorrect words, and its presence can hinder the performance of current models that were not implemented to withstand such noises, but must still perform well even under duress. This is due to the fact that current approaches are built for and trained with clean and complete data, and thus are not able to extract features that can adequately represent incomplete data. Our proposed approach consists of obtaining intermediate input representations by applying an embedding layer to the input tokens followed by vanilla transformers. These intermediate features are given as input to novel denoising transformers which are responsible for obtaining richer input representations. The proposed approach takes advantage of stacks of multilayer perceptrons for the reconstruction of missing words' embeddings by extracting more abstract and meaningful hidden feature vectors, and bidirectional transformers for improved embedding representation. We consider two datasets for training and evaluation: the Chatbot Natural Language Understanding Evaluation Corpus and Kaggle's Twitter Sentiment Corpus. Our model shows improved F1-scores and better robustness in informal/incorrect texts present in tweets and in texts with Speech-to-Text error in the sentiment and intent classification tasks.
Nowadays, deep learning methods, especially the convolutional neural networks (CNNs), have shown impressive performance on extracting abstract and high-level features from the hyperspectral image. However, general training process of CNNs mainly considers the pixel-wise information or the samples' correlation to formulate the penalization while ignores the statistical properties especially the spectral variability of each class in the hyperspectral image. These samples-based penalizations would lead to the uncertainty of the training process due to the imbalanced and limited number of training samples. To overcome this problem, this work characterizes each class from the hyperspectral image as a statistical distribution and further develops a novel statistical loss with the distributions, not directly with samples for deep learning. Based on the Fisher discrimination criterion, the loss penalizes the sample variance of each class distribution to decrease the intra-class variance of the training samples. Moreover, an additional diversity-promoting condition is added to enlarge the inter-class variance between different class distributions and this could better discriminate samples from different classes in hyperspectral image. Finally, the statistical estimation form of the statistical loss is developed with the training samples through multi-variant statistical analysis. Experiments over the real-world hyperspectral images show the effectiveness of the developed statistical loss for deep learning.
Automatic bill classification is an attractive task with many potential applications such as automated detection and counting in images or videos. To address this purpose we present a Deep Learning Model to classify Chilean Banknotes, because of its successful results in image processing applications. For optimal performance of the proposed model, data augmentation techniques are introduced due to the limited number of image samples. Positive results were achieved in this work, verifying that it could be a stating point to be extended to more complex applications.
A white noise analysis of modern deep neural networks is presented to unveil their biases at the whole network level or the single neuron level. Our analysis is based on two popular and related methods in psychophysics and neurophysiology namely classification images and spike triggered analysis. These methods have been widely used to understand the underlying mechanisms of sensory systems in humans and monkeys. We leverage them to investigate the inherent biases of deep neural networks and to obtain a first-order approximation of their functionality. We emphasize on CNNs since they are currently the state of the art methods in computer vision and are a decent model of human visual processing. In addition, we study multi-layer perceptrons, logistic regression, and recurrent neural networks. Experiments over four classic datasets, MNIST, Fashion-MNIST, CIFAR-10, and ImageNet, show that the computed bias maps resemble the target classes and when used for classification lead to an over twofold performance than the chance level. Further, we show that classification images can be used to attack a black-box classifier and to detect adversarial patch attacks. Finally, we utilize spike triggered averaging to derive the filters of CNNs and explore how the behavior of a network changes when neurons in different layers are modulated. Our effort illustrates a successful example of borrowing from neurosciences to study ANNs and highlights the importance of cross-fertilization and synergy across machine learning, deep learning, and computational neuroscience.
The development of deep convolutional neural network architecture is critical to the improvement of image classification task performance. Many image classification studies use deep convolutional neural network and focus on modifying the network structure to improve image classification performance. Conversely, our study focuses on loss function design. Cross-entropy Loss (CEL) has been widely used for training deep convolutional neural network for the task of multi-class classification. Although CEL has been successfully implemented in several image classification tasks, it only focuses on the posterior probability of the correct class. For this reason, a negative log likelihood ratio loss (NLLR) was proposed to better differentiate between the correct class and the competing incorrect ones. However, during the training of the deep convolutional neural network, the value of NLLR is not always positive or negative, which severely affects the convergence of NLLR. Our proposed competing ratio loss (CRL) calculates the posterior probability ratio between the correct class and the competing incorrect classes to further enlarge the probability difference between the correct and incorrect classes. We added hyperparameters to CRL, thereby ensuring its value to be positive and that the update size of backpropagation is suitable for the CRL's fast convergence. To demonstrate the performance of CRL, we conducted experiments on general image classification tasks (CIFAR10/100, SVHN, ImageNet), the fine-grained image classification tasks (CUB200-2011 and Stanford Car), and the challenging face age estimation task (using Adience). Experimental results show the effectiveness and robustness of the proposed loss function on different deep convolutional neural network architectures and different image classification tasks.
Person re-identification has attracted many researchers' attention for its wide application, but it is still a very challenging task because only part of the image information can be used for personnel matching. Most of current methods uses CNN to learn to embeddings that can capture semantic similarity information among data points. Many of the state-of-the-arts methods use complex network structures with multiple branches that fuse multiple features while training or testing, using classification loss, Triplet loss or a combination of the two as loss function. However, the method that using Triplet loss as loss function converges slowly, and the method in which pull features of the same class as close as possible in features space leads to poor feature stability. This paper will combine the ranking motivated structured loss, proposed a new metric learning loss function that make the features of the same class are sparsely distributed into the range of small hyperspheres and the features of different classes are uniformly distributed at a clearly angle. And adopted a new single-branch network structure that only using global feature can also get great performance. The validity of our method is verified on the Market1501 and DukeMTMC-ReID person re-identification datasets. Finally acquires 90.9% rank-1 accuracy and 80.8% mAP on DukeMTMC-reID, 95.3% rank-1 accuracy and 88.7% mAP on Market1501. Codes and models are available in Github.https://github.com/Qidian213/Ranked_Person_ReID.