For machine learning perception problems, human-level classification performance is used as an estimate of top algorithm performance. Thus, it is important to understand as precisely as possible the factors that impact human-level performance. Knowing this 1) provides a benchmark for model performance, 2) tells a project manager what type of data to obtain for human labelers in order to get accurate labels, and 3) enables ground-truth analysis--largely conducted by humans--to be carried out smoothly. In this empirical study, we explored the relationship between resolution and human classification performance using the MNIST data set down-sampled to various resolutions. The quantitative heuristic we derived could prove useful for predicting machine model performance, predicting data storage requirements, and saving valuable resources in the deployment of machine learning projects. It also has the potential to be used in a wide variety of fields such as remote sensing, medical imaging, scientific imaging, and astronomy.
Cross-domain sentiment classification has drawn much attention in recent years. Most existing approaches focus on learning domain-invariant representations in both the source and target domains, while few of them pay attention to the domain-specific information. Despite the non-transferability of the domain-specific information, simultaneously learning domain-dependent representations can facilitate the learning of domain-invariant representations. In this paper, we focus on aspect-level cross-domain sentiment classification, and propose to distill the domain-invariant sentiment features with the help of an orthogonal domain-dependent task, i.e. aspect detection, which is built on the aspects varying widely in different domains. We conduct extensive experiments on three public datasets and the experimental results demonstrate the effectiveness of our method.
We propose a framework for constructing and analyzing multiclass and multioutput classification metrics, i.e., involving multiple, possibly correlated multiclass labels. Our analysis reveals novel insights on the geometry of feasible confusion tensors -- including necessary and sufficient conditions for the equivalence between optimizing an arbitrary non-decomposable metric and learning a weighted classifier. Further, we analyze averaging methodologies commonly used to compute multioutput metrics and characterize the corresponding Bayes optimal classifiers. We show that the plug-in estimator based on this characterization is consistent and is easily implemented as a post-processing rule. Empirical results on synthetic and benchmark datasets support the theoretical findings.
Various psychological factors affect how individuals express emotions. Yet, when we collect data intended for use in building emotion recognition systems, we often try to do so by creating paradigms that are designed just with a focus on eliciting emotional behavior. Algorithms trained with these types of data are unlikely to function outside of controlled environments because our emotions naturally change as a function of these other factors. In this work, we study how the multimodal expressions of emotion change when an individual is under varying levels of stress. We hypothesize that stress produces modulations that can hide the true underlying emotions of individuals and that we can make emotion recognition algorithms more generalizable by controlling for variations in stress. To this end, we use adversarial networks to decorrelate stress modulations from emotion representations. We study how stress alters acoustic and lexical emotional predictions, paying special attention to how modulations due to stress affect the transferability of learned emotion recognition models across domains. Our results show that stress is indeed encoded in trained emotion classifiers and that this encoding varies across levels of emotions and across the lexical and acoustic modalities. Our results also show that emotion recognition models that control for stress during training have better generalizability when applied to new domains, compared to models that do not control for stress during training. We conclude that is is necessary to consider the effect of extraneous psychological factors when building and testing emotion recognition models.
The goal of lossy data compression is to reduce the storage cost of a data set $X$ while retaining as much information as possible about something ($Y$) that you care about. For example, what aspects of an image $X$ contain the most information about whether it depicts a cat? Mathematically, this corresponds to finding a mapping $X\to Z\equiv f(X)$ that maximizes the mutual information $I(Z,Y)$ while the entropy $H(Z)$ is kept below some fixed threshold. We present a method for mapping out the Pareto frontier for classification tasks, reflecting the tradeoff between retained entropy and class information. We first show how a random variable $X$ (an image, say) drawn from a class $Y\in\{1,...,n\}$ can be distilled into a vector $W=f(X)\in \mathbb{R}^{n-1}$ losslessly, so that $I(W,Y)=I(X,Y)$; for example, for a binary classification task of cats and dogs, each image $X$ is mapped into a single real number $W$ retaining all information that helps distinguish cats from dogs. For the $n=2$ case of binary classification, we then show how $W$ can be further compressed into a discrete variable $Z=g_\beta(W)\in\{1,...,m_\beta\}$ by binning $W$ into $m_\beta$ bins, in such a way that varying the parameter $\beta$ sweeps out the full Pareto frontier, solving a generalization of the Discrete Information Bottleneck (DIB) problem. We argue that the most interesting points on this frontier are "corners" maximizing $I(Z,Y)$ for a fixed number of bins $m=2,3...$ which can be conveniently be found without multiobjective optimization. We apply this method to the CIFAR-10, MNIST and Fashion-MNIST datasets, illustrating how it can be interpreted as an information-theoretically optimal image clustering algorithm.
Despite the remarkable success of deep learning in pattern recognition, deep network models face the problem of training a large number of parameters. In this paper, we propose and evaluate a novel multi-path wavelet neural network architecture for image classification with far less number of trainable parameters. The model architecture consists of a multi-path layout with several levels of wavelet decompositions performed in parallel followed by fully connected layers. These decomposition operations comprise wavelet neurons with learnable parameters, which are updated during the training phase using the back-propagation algorithm. We evaluate the performance of the introduced network using common image datasets without data augmentation except for SVHN and compare the results with influential deep learning models. Our findings support the possibility of reducing the number of parameters significantly in deep neural networks without compromising its accuracy.
We propose a simple yet effective method to reduce the redundancy of DenseNet by substantially decreasing the number of stacked modules by replacing the original bottleneck by our SMG module, which is augmented by local residual. Furthermore, SMG module is equipped with an efficient two-stage pipeline, which aims to DenseNet-like architectures that need to integrate all previous outputs, i.e., squeezing the incoming informative but redundant features gradually by hierarchical convolutions as a hourglass shape and then exciting it by multi-kernel depthwise convolutions, the output of which would be compact and hold more informative multi-scale features. We further develop a forget and an update gate by introducing the popular attention modules to implement the effective fusion instead of a simple addition between reused and new features. Due to the Hybrid Connectivity (nested combination of global dense and local residual) and Gated mechanisms, we called our network as the HCGNet. Experimental results on CIFAR and ImageNet datasets show that HCGNet is more prominently efficient than DenseNet, and can also significantly outperform state-of-the-art networks with less complexity. Moreover, HCGNet also shows the remarkable interpretability and robustness by network dissection and adversarial defense, respectively. On MS-COCO, HCGNet can consistently learn better features than popular backbones.
Text attributes, such as user and product information in product reviews, have been used to improve the performance of sentiment classification models. The de facto standard method is to incorporate them as additional biases in the attention mechanism, and more performance gains are achieved by extending the model architecture. In this paper, we show that the above method is the least effective way to represent and inject attributes. To demonstrate this hypothesis, unlike previous models with complicated architectures, we limit our base model to a simple BiLSTM with attention classifier, and instead focus on how and where the attributes should be incorporated in the model. We propose to represent attributes as chunk-wise importance weight matrices and consider four locations in the model (i.e., embedding, encoding, attention, classifier) to inject attributes. Experiments show that our proposed method achieves significant improvements over the standard approach and that attention mechanism is the worst location to inject attributes, contradicting prior work. We also outperform the state-of-the-art despite our use of a simple base model. Finally, we show that these representations transfer well to other tasks. Model implementation and datasets are released here: https://github.com/rktamplayo/CHIM.
In this paper, we propose a convolutional neural network with mapping layers (MCNN) for hyperspectral image (HSI) classification. The proposed mapping layers map the input patch into a low dimensional subspace by multilinear algebra. We use our mapping layers to reduce the spectral and spatial redundancy and maintain most energy of the input. The feature extracted by our mapping layers can also reduce the number of following convolutional layers for feature extraction. Our MCNN architecture avoids the declining accuracy with increasing layers phenomenon of deep learning models for HSI classification and also saves the training time for its effective mapping layers. Furthermore, we impose the 3-D convolutional kernel on convolutional layer to extract the spectral-spatial features for HSI. We tested our MCNN on three datasets of Indian Pines, University of Pavia and Salinas, and we achieved the classification accuracy of 98.3%, 99.5% and 99.3%, respectively. Experimental results demonstrate that the proposed MCNN can significantly improve the classification accuracy and save much time consumption.
We perform the classification of ancient Roman Republican coins via recognizing their reverse motifs where various objects, faces, scenes, animals, and buildings are minted along with legends. Most of these coins are eroded due to their age and varying degrees of preservation, thereby affecting their informative attributes for visual recognition. Changes in the positions of principal symbols on the reverse motifs also cause huge variations among the coin types. Lastly, in-plane orientations, uneven illumination, and a moderate background clutter further make the classification task non-trivial and challenging. To this end, we present a novel network model, CoinNet, that employs compact bilinear pooling, residual groups, and feature attention layers. Furthermore, we gathered the largest and most diverse image dataset of the Roman Republican coins that contains more than 18,000 images belonging to 228 different reverse motifs. On this dataset, our model achieves a classification accuracy of more than \textbf{98\%} and outperforms the conventional bag-of-visual-words based approaches and more recent state-of-the-art deep learning methods. We also provide a detailed ablation study of our network and its generalization capability. Models and Datasets available at https://github.com/saeed-anwar/CoinNet
Multiclass decomposition splits a multiclass classification problem into a series of independent binary learners and recomposes them by combining their outputs to reconstruct the multiclass classification results. Three widely-used realizations of such decomposition methods are One-Versus-All (OVA), One-Versus-One (OVO), and Error-Correcting-Output-Code (ECOC). While OVA and OVO are quite simple, both of them assume all classes are orthogonal which neglect the latent correlation between classes in real-world. Error-Correcting-Output-Code (ECOC) based decomposition methods, on the other hand, are more preferable due to its integration of the correlation among classes. However, the performance of existing ECOC-based methods highly depends on the design of coding matrix and decoding strategy. Unfortunately, it is quite uncertain and time-consuming to discover an effective coding matrix with appropriate decoding strategy. To address this problem, we propose LightMC, an efficient dynamic multiclass decomposition algorithm. Instead of using fixed coding matrix and decoding strategy, LightMC uses a differentiable decoding strategy, which enables it to dynamically optimize the coding matrix and decoding strategy, toward increasing the overall accuracy of multiclass classification, via back propagation jointly with the training of base learners in an iterative way. Empirical experimental results on several public large-scale multiclass classification datasets have demonstrated the effectiveness of LightMC in terms of both good accuracy and high efficiency.
In this work, we introduce a recently developed early classification mechanism to satellite-based agricultural monitoring. It augments existing classification models by an additional stopping probability based on the previously seen information. This mechanism is end-to-end trainable and derives its stopping decision solely from the observed satellite data. We show results on field parcels in central Europe where sufficient ground truth data is available for an empiric evaluation of the results with local phenological information obtained from authorities. We observe that the recurrent neural network outfitted with this early classification mechanism was able to distinguish the many of the crop types before the end of the vegetative period. Further, we associated these stopping times with evaluated ground truth information and saw that the times of classification were related to characteristic events of the observed plants' phenology.
Large-scale datasets play a fundamental role in training deep learning models. However, dataset collection is difficult in domains that involve sensitive information. Collaborative learning techniques provide a privacy-preserving solution, by enabling training over a number of private datasets that are not shared by their owners. However, recently, it has been shown that the existing collaborative learning frameworks are vulnerable to an active adversary that runs a generative adversarial network (GAN) attack. In this work, we propose a novel classification model that is resilient against such attacks by design. More specifically, we introduce a key-based classification model and a principled training scheme that protects class scores by using class-specific private keys, which effectively hide the information necessary for a GAN attack. We additionally show how to utilize high dimensional keys to improve the robustness against attacks without increasing the model complexity. Our detailed experiments demonstrate the effectiveness of the proposed technique. Source code is available at https://github.com/mbsariyildiz/key-protected-classification.
In this paper, we propose a simple while effective unsupervised deep feature transfer algorithm for low resolution image classification. No fine-tuning on convenet filters is required in our method. We use pre-trained convenet to extract features for both high- and low-resolution images, and then feed them into a two-layer feature transfer network for knowledge transfer. A SVM classifier is learned directly using these transferred low resolution features. Our network can be embedded into the state-of-the-art deep neural networks as a plug-in feature enhancement module. It preserves data structures in feature space for high resolution images, and transfers the distinguishing features from a well-structured source domain (high resolution features space) to a not well-organized target domain (low resolution features space). Extensive experiments on VOC2007 test set show that the proposed method achieves significant improvements over the baseline of using feature extraction.
Deep learning model trained by imbalanced data may not work satisfactorily since it could be determined by major classes and thus may ignore the classes with small amount of data. In this paper, we apply deep learning based imbalanced data classification for the first time to cellular macromolecular complexes captured by Cryo-electron tomography (Cryo-ET). We adopt a range of strategies to cope with imbalanced data, including data sampling, bagging, boosting, Genetic Programming based method and. Particularly, inspired from Inception 3D network, we propose a multi-path CNN model combining focal loss and mixup on the Cryo-ET dataset to expand the dataset, where each path had its best performance corresponding to each type of data and let the network learn the combinations of the paths to improve the classification performance. In addition, extensive experiments have been conducted to show our proposed method is flexible enough to cope with different number of classes by adjusting the number of paths in our multi-path model. To our knowledge, this work is the first application of deep learning methods of dealing with imbalanced data to the internal tissue classification of cell macromolecular complexes, which opened up a new path for cell classification in the field of computational biology.
Latent factor models (LFMs) such as matrix factorization have achieved the state-of-the-art performance among various collaborative filtering approaches for recommendation. Despite the high recommendation accuracy of LFMs, a critical issue to be resolved is their lack of interpretability. Extensive efforts have been devoted to interpreting the prediction results of LFMs. However, they either rely on auxiliary information which may not be available in practice, or sacrifice recommendation accuracy for interpretability. Influence functions, stemming from robust statistics, have been developed to understand the effect of training points on the predictions of black-box models. Inspired by this, we propose a novel explanation method named FIA (Fast Influence Analysis) to understand the prediction of trained LFMs by tracing back to the training data with influence functions. We present how to employ influence functions to measure the impact of historical user-item interactions on the prediction results of LFMs and provide intuitive neighbor-style explanations based on the most influential interactions. Our proposed FIA exploits the characteristics of two important LFMs, matrix factorization and neural collaborative filtering, and is capable of accelerating the overall influence analysis process. We provide a detailed complexity analysis for FIA over LFMs and conduct extensive experiments to evaluate its performance using real-world datasets. The results demonstrate the effectiveness and efficiency of FIA, and the usefulness of the generated explanations for the recommendation results.
This chapter presents the investigations and the results of feature learning using convolutional neural networks to automatically assess knee osteoarthritis (OA) severity and the associated clinical and diagnostic features of knee OA from X-ray images. Also, this chapter demonstrates that feature learning in a supervised manner is more effective than using conventional handcrafted features for automatic detection of knee joints and fine-grained knee OA image classification. In the general machine learning approach to automatically assess knee OA severity, the first step is to localize the region of interest that is to detect and extract the knee joint regions from the radiographs, and the next step is to classify the localized knee joints based on a radiographic classification scheme such as Kellgren and Lawrence grades. First, the existing approaches for detecting (or localizing) the knee joint regions based on handcrafted features are reviewed and outlined. Next, three new approaches are introduced: 1) to automatically detect the knee joint region using a fully convolutional network, 2) to automatically assess the radiographic knee OA using CNNs trained from scratch for classification and regression of knee joint images to predict KL grades in ordinal and continuous scales, and 3) to quantify the knee OA severity optimizing a weighted ratio of two loss functions: categorical cross entropy and mean-squared error using multi-objective convolutional learning and ordinal regression. Two public datasets: the OAI and the MOST are used to evaluate the approaches with promising results that outperform existing approaches. In summary, this work primarily contributes to the field of automated methods for localization (automatic detection) and quantification (image classification) of radiographic knee OA.
Music genre classification is a widely researched topic in music information retrieval (MIR). Being able to automatically tag genres will benefit music streaming service providers such as JOOX, Apple Music, and Spotify for their content-based recommendation. However, most studies on music classification have been done on western songs which differ from Thai songs. Lukthung, a distinctive and long-established type of Thai music, is one of the most popular music genres in Thailand and has a specific group of listeners. In this paper, we develop neural networks to classify such Lukthung genre from others using both lyrics and audios. Words used in Lukthung songs are particularly poetical, and their musical styles are uniquely composed of traditional Thai instruments. We leverage these two main characteristics by building a lyrics model based on bag-of-words (BoW), and an audio model using a convolutional neural network (CNN) architecture. We then aggregate the intermediate features learned from both models to build a final classifier. Our results show that the proposed three models outperform all of the standard classifiers where the combined model yields the best $F_1$ score of 0.86, allowing Lukthung classification to be applicable to personalized recommendation for Thai audience.
We describe a software toolbox for the configuration of deep neural networks in the domain of skin cancer classification. The implemented software architecture allows developers to quickly set up new convolutional neural network (CNN) architectures and hyper-parameter configurations. At the same time, the user interface, manageable as a simple spreadsheet, allows non-technical users to explore different configuration settings that need to be explored when switching to different data sets. In future versions, meta leaning frameworks can be added, or AutoML systems that continuously improve over time. Preliminary results, conducted with two CNNs in the context melanoma detection on dermoscopic images, quantify the impact of image augmentation, image resolution, and rescaling filter on the overall detection performance and training time.
We propose a novel approach for estimating the difficulty and transferability of supervised classification tasks. Unlike previous work, our approach is solution agnostic and does not require or assume trained models. Instead, we estimate these values using an information theoretic approach: treating training labels as random variables and exploring their statistics. When transferring from a source to a target task, we consider the conditional entropy between two such variables (i.e., label assignments of the two tasks). We show analytically and empirically that this value is related to the loss of the transferred model. We further show how to use this value to estimate task hardness. We test our claims extensively on three large scale data sets -- CelebA (40 tasks), Animals with Attributes 2 (85 tasks), and Caltech-UCSD Birds 200 (312 tasks) -- together representing 437 classification tasks. We provide results showing that our hardness and transferability estimates are strongly correlated with empirical hardness and transferability. As a case study, we transfer a learned face recognition model to CelebA attribute classification tasks, showing state of the art accuracy for tasks estimated to be highly transferable.