Recent success of text-to-image (T2I) generation and its increasing practical applications, enabled by diffusion models, require urgent consideration of erasing unwanted concepts, e.g., copyrighted, offensive, and unsafe ones, from the pre-trained models in a precise, timely, and low-cost manner. The twofold demand of concept erasure includes not only a precise removal of the target concept (i.e., erasure efficacy) but also a minimal change on non-target content (i.e., prior preservation), during generation. Existing methods face challenges in maintaining an effective balance between erasure efficacy and prior preservation, and they can be computationally costly. To improve, we propose a precise, fast, and low-cost concept erasure method, called Adaptive Vaule Decomposer (AdaVD), which is training-free. Our method is grounded in a classical linear algebraic operation of computing orthogonal complement, implemented in the value space of each cross-attention layer within the UNet of diffusion models. We design a shift factor to adaptively navigate the erasure strength, enhancing effectively prior preservation without sacrificing erasure efficacy. Extensive comparative experiments with both training-based and training-free state of the arts demonstrate that the proposed AdaVD excels in both single and multiple concept erasure, showing 2 to 10 times of improvement in prior preservation than the second best, meanwhile achieving the best or near best erasure efficacy. AdaVD supports a series of diffusion models and downstream image generation tasks, with code available on: https://github.com/WYuan1001/AdaVD.

Group-agent reinforcement learning (GARL) is a newly arising learning scenario, where multiple reinforcement learning agents study together in a group, sharing knowledge in an asynchronous fashion. The goal is to improve the learning performance of each individual agent. Under a more general heterogeneous setting where different agents learn using different algorithms, we advance GARL by designing novel and effective group-learning mechanisms. They guide the agents on whether and how to learn from action choices from the others, and allow the agents to adopt available policy and value function models sent by another agent if they perform better. We have conducted extensive experiments on a total of 43 different Atari 2600 games to demonstrate the superior performance of the proposed method. After the group learning, among the 129 agents examined, 96% are able to achieve a learning speed-up, and 72% are able to learn $\geq 100$ times faster. Also, around 41% of those agents have achieved a higher accumulated reward score by learning in $\leq 5%$ of the time steps required by a single agent when learning on its own.

Compared to conventional semantic segmentation with pixel-level supervision, weakly supervised semantic segmentation (WSSS) with image-level labels poses the challenge that it commonly focuses on the most discriminative regions, resulting in a disparity between weakly and fully supervision scenarios. A typical manifestation is the diminished precision on object boundaries, leading to deteriorated accuracy of WSSS. To alleviate this issue, we propose to adaptively partition the image content into certain regions (e.g., confident foreground and background) and uncertain regions (e.g., object boundaries and misclassified categories) for separate processing. For uncertain cues, we propose an adaptive masking strategy and seek to recover the local information with self-distilled knowledge. We further assume that confident regions should be robust enough to preserve the global semantics, and introduce a complementary self-distillation method that constrains semantic consistency between confident regions and an augmented view with the same class labels. Extensive experiments conducted on PASCAL VOC 2012 and MS COCO 2014 demonstrate that our proposed single-stage approach for WSSS not only outperforms state-of-the-art counterparts but also surpasses multi-stage methods that trade complexity for accuracy.

Financial resilience has been an important area of focus for the business sector since the outbreak of the pandemic. Currently, the assessment of financial resilience is typically completed through the review of financial statements. However, such resilience is commonly linked to negative life events and may be further impacted by the presence of mental health difficulties. As such, identifying and understanding these elements may provide a more complete understanding of an individual’s financial situation. We discuss the development of a challenging automated financial resilience detection system that aims to identify factors that may have negative impacts upon the resilience of individuals. This makes use of textual data to identify elements, such as the occurrence of negative life events or mental health difficulties, that indicate that individuals may be vulnerable to exploitation through the misselling of products. In addition to a traditional data-driven supervised approach, this work also demonstrates applying prompt-based learning to these tasks without the need for the training data (i.e., a dataless approach).

Cinemagraph creates captivating video experience by combining elements of still photography and subtle motion. However, most existing cinemagraph video generation lacks depth information, being restricted within 2-dimensional (2D) image space. We advance cinemagraph from 2D image space to 3-dimensional (3D) space with high quality by proposing LoopGaussian. It is based on 3D Gaussian modeling, taking advantage of the 3D Gaussian Splatting (3D-GS) technique that has significantly improved the field of novel view synthesis. Here is a brief overview of our new approach: It employs 3D-GS to reconstruct 3D Gaussian point clouds from multi-view images of static scenes, where shape regularization is used to prevent blurring or artifacts caused by object deformation. To maintain local continuity between scenes, it then clusters the 3D Gaussian points by the proposed SuperGaussian algorithm using features acquired by an autoencoder tailored for 3D Gaussian. Similarities between clusters are used to derive an Eulerian motion field for describing velocities across the entire scene. The estimated Eulerian motion field drives the movement of the 3D Gaussian points, based on which a 3D Cinemagraph is generated through bidirectional animation. The resulting 3D Cinemagraph exhibits natural and seamlessly loopable dynamics. Experiment results validate the effectiveness of the proposed approach, demonstrating high-quality and visually appealing video generation.

The semantic matching problem consists of recognizing if the candidate text is relevant to a particular input text. Semantic similarities can be determined from human-curated knowledge, but such knowledge may not be available in every language. Instead, statistical learning techniques have been applied, but these techniques circumvent the need for manual feature engineering by using large datasets to train models to perform semantic similarity scoring between portions of text or words. The pre-trained transformer provides a further mechanism to consolidate the information throughout a sentence into single sentence-level representations, but these representations may not be optimal for the matching task. As an alternative, we propose an interactive semantic transformer based on a greedy layer-wise framework to learn a distributed similarity representation for sentence pairs. The novelty of the architecture lies in an abstract representation of the semantic similarities created by three-stage learning strategies. Model training is accomplished through a greedy layer-wise training scheme, that incorporates both supervised and unsupervised learning. The proposed model is experimentally compared to state-of-the-art approaches on three different dataset types: the library TREC, the Yahoo!, and Stack Exchange community question datasets, and results show the proposed model outperforming other approaches.

Zero-shot learning (ZSL) suffers intensely from the domain shift issue, i.e., the mismatch (or misalignment) between the true and learned data distributions for classes without training data (unseen classes). By learning additionally from unlabelled data collected for the unseen classes, transductive ZSL (TZSL) could reduce the shift but only to a certain extent. To improve TZSL, we propose a novel approach Bi-VAEGAN which strengthens the distribution alignment between the visual space and an auxiliary space. As a result, it can reduce largely the domain shift. The proposed key designs include (1) a bi-directional distribution alignment, (2) a simple but effective L2-norm based feature normalization approach, and (3) a more sophisticated unseen class prior estimation. Evaluated by four benchmark datasets, Bi-VAEGAN11Code is available at https://github.com/Zhicaiwww/Bi-VAEGAN achieves the new state of the art under both the standard and generalized TZSL settings.

The strategy of ensemble has become popular in adversarial defense, which trains multiple base classifiers to defend against adversarial attacks in a cooperative manner. Despite the empirical success, theoretical explanations on why an ensemble of adversarially trained classifiers is more robust than single ones remain unclear. To fill in this gap, we develop a new error theory dedicated to understanding ensemble adversarial defense, demonstrating a provable 0-1 loss reduction on challenging sample sets in adversarial defense scenarios. Guided by this theory, we propose an effective approach to improve ensemble adversarial defense, named interactive global adversarial training (iGAT). The proposal includes (1) a probabilistic distributing rule that selectively allocates to different base classifiers adversarial examples that are globally challenging to the ensemble, and (2) a regularization term to rescue the severest weaknesses of the base classifiers. Being tested over various existing ensemble adversarial defense techniques, iGAT is capable of boosting their performance by up to 17\% evaluated using CIFAR10 and CIFAR100 datasets under both white-box and black-box attacks.

When deploying machine learning estimators in science and engineering (SAE) domains, it is critical to avoid failed estimations that can have disastrous consequences, e.g., in aero engine design. This work focuses on detecting and correcting failed state estimations before adopting them in SAE inverse problems, by utilizing simulations and performance metrics guided by physical laws. We suggest to flag a machine learning estimation when its physical model error exceeds a feasible threshold, and propose a novel approach, GEESE, to correct it through optimization, aiming at delivering both low error and high efficiency. The key designs of GEESE include (1) a hybrid surrogate error model to provide fast error estimations to reduce simulation cost and to enable gradient based backpropagation of error feedback, and (2) two generative models to approximate the probability distributions of the candidate states for simulating the exploitation and exploration behaviours. All three models are constructed as neural networks. GEESE is tested on three real-world SAE inverse problems and compared to a number of state-of-the-art optimization/search approaches. Results show that it fails the least number of times in terms of finding a feasible state correction, and requires physical evaluations less frequently in general.

We present a theory of ensemble diversity, explaining the nature of diversity for a wide range of supervised learning scenarios. This challenge has been referred to as the “holy grail” of ensemble learning, an open research issue for over 30 years. Our framework reveals that diversity is in fact a hidden dimension in the bias-variance decomposition of the ensemble loss. We prove a family of exact bias-variance-diversity decompositions, for a wide range of losses in both regression and classification, e.g., squared, cross-entropy, and Poisson losses. For losses where an additive bias-variance decomposition is not available (e.g., 0/1 loss) we present an alternative approach: quantifying the effects of diversity, which turn out to be dependent on the label distribution. Overall, we argue that diversity is a measure of model fit, in precisely the same sense as bias and variance, but accounting for statistical dependencies between ensemble members. Thus, we should not be ‘maximising diversity’ as so many works aim to do---instead, we have a bias/variance/diversity trade-off to manage.

Abstract This work is on constrained large-scale non-convex optimization where the constraint set implies a manifold structure. Solving such problems is important in a multitude of fundamental machine learning tasks. Recent advances on Riemannian optimization have enabled the convenient recovery of solutions by adapting unconstrained optimization algorithms over manifolds. However, it remains challenging to scale up and meanwhile maintain stable convergence rates and handle saddle points. We propose a new second-order Riemannian optimization algorithm, aiming at improving convergence rate and reducing computational cost. It enhances the Riemannian trust-region algorithm that explores curvature information to escape saddle points through a mixture of subsampling and cubic regularization techniques. We conduct rigorous analysis to study the convergence behavior of the proposed algorithm. We also perform extensive experiments to evaluate it based on two general machine learning tasks using multiple datasets. The proposed algorithm exhibits improved computational speed, e.g., a speed improvement from $$12\% \:\text {to} \:227\%$$ 12 % to 227 % , and improved convergence behavior, e.g., an iteration number reduction from $$\mathcal{O}\left(\max\left(\epsilon_g^{-2}\epsilon_H^{-1},\epsilon_H^{-3}\right)\right) \,\text {to}\: \mathcal{O}\left(\max\left(\epsilon_g^{-2},\epsilon_H^{-3}\right)\right)$$ O max ϵ g - 2 ϵ H - 1 , ϵ H - 3 to O max ϵ g - 2 , ϵ H - 3 , compared to a large set of state-of-the-art Riemannian optimization algorithms.

The question answering system in open domain enables a machine to automatically select and generate the answer for questions posed by humans in a natural language form on the website. Previous approaches seek effective ways of extracting the semantic features between question and answer, but the contextual information effects in semantic matching are still limited by short-term memory. As an alternative, we propose an internal knowledge-based end-to-end model, enhanced by an attentive memory network for both answer selection and answer generation tasks by considering the full advantages of the semantics and multifacts (i.e., timescales, topics, and context). In detail, we design a long-term memory to learn the top-$k$fine-grained similarity representations, where two memory-aware mechanisms aggregate the series of semantic word-level and sentence-level similarities to support the coarse contextual information. Furthermore, we propose a novel memory refinement mechanism with the two-dimensional of writing heads that offer an efficient approach to multiview selection of the salient word pairs. In the training stage, we adopt the transformer-based transfer learning skill to effectively pretrain the model. Experimentally, we compare the state-of-the-art approaches on four public datasets, the experimental results show that the proposed model achieves competitive performance.

We introduce a novel bias-variance decomposition for a range of strictly convex margin losses, including the logistic loss (minimized by the classic LogitBoost algorithm) as well as the squared margin loss and canonical boosting loss. Furthermore we show that, for all strictly convex margin losses, the expected risk decomposes into the risk of a "central" model and a term quantifying variation in the functional margin with respect to variations in the training data. These decompositions provide a diagnostic tool for practitioners to understand model overfitting/underfitting, and have implications for additive ensemble models—for example, when our bias-variance decomposition holds, there is a corresponding "ambiguity" decomposition, which can be used to quantify model diversity.

Vision multi-layer perceptrons (MLPs) have shown promising performance in computer vision tasks, and become the main competitor of CNNs and vision Transformers. They use token-mixing layers to capture cross-token interactions, as opposed to the multi-head self-attention mechanism used by Transformers. However, the heavily parameterized token-mixing layers naturally lack mechanisms to capture local information and multi-granular non-local relations, thus their discriminative power is restrained. To tackle this issue, we propose a new positional spacial gating unit (PoSGU). It exploits the attention formulations used in the classical relative positional encoding (RPE), to efficiently encode the cross-token relations for token mixing. It can successfully reduce the current quadratic parameter complexity O(N2) of vision MLPs to $O(N)$ and O(1). We experiment with two RPE mechanisms, and further propose a group-wise extension to improve their expressive power with the accomplishment of multi-granular contexts. These then serve as the key building blocks of a new type of vision MLP, referred to as PosMLP. We evaluate the effectiveness of the proposed approach by conducting thorough experiments, demonstrating an improved or comparable performance with reduced parameter complexity. For instance, for a model trained on ImageNet1K, we achieve a performance improvement from 72.14% to 74.02% and a learnable parameter reduction from 19.4M to 18.2M. Code could be found at https://github.com/Zhicaiwww/PosMLP https://github.com/Zhicaiwww/PosMLP.

Image reconstruction enhanced by regularizers, e.g., to enforce sparsity, low rank or smoothness priors on images, has many successful applications in vision tasks such as computer photography, biomedical and spectral imaging. It has been well accepted that non-convex regularizers normally perform better than convex ones in terms of the reconstruction quality. But their convergence analysis is only established to a critical point, rather than the global optima. To mitigate the loss of guarantees for global optima, we propose to apply the concept of invexity and provide the first list of proved invex regularizers for improving image reconstruction. Moreover, we establish convergence guarantees to global optima for various advanced image reconstruction techniques after being improved by such invex regularization. To the best of our knowledge, this is the first practical work applying invex regularization to improve imaging with global optima guarantees. To demonstrate the effectiveness of invex regularization, numerical experiments are conducted for various imaging tasks using benchmark datasets.

Knowledge graphs have received intensive research interests. When the labels of most nodes or datapoints are missing, anchor graph and hierarchical anchor graph models can be employed. With an anchor graph or hierarchical anchor graph, we only need to optimize the labels of the coarsest anchors, and the labels of datapoints can be inferred from these anchors in a coarse-to-fine manner. The complexity of optimization is therefore reduced to a cubic cost with respect to the number of the coarsest anchors. However, to obtain a high accuracy when a data distribution is complex, the scale of this anchor set still needs to be large, which thus inevitably incurs an expensive computational burden. As such, a challenge in scaling up these models is how to efficiently estimate the labels of these anchors while keeping classification performance. To address this problem, we propose a novel approach that adds an anchor label predictor in the conventional anchor graph and hierarchical anchor graph models. In the proposed approach, the labels of the coarsest anchors are not directly optimized, and instead, we learn a label predictor which estimates the labels of these anchors with their spectral representations. The predictor is optimized with a regularization on all datapoints based on a hierarchical anchor graph, and we show that its solution only involves the inversion of a small-size matrix. Built upon the anchor hierarchy, we design a sparse intra-layer adjacency matrix over these anchors, which can simultaneously accelerate spectral embedding and enhance effectiveness. Our approach is named Faster Learning on Anchor Graph (FLAG) as it improves conventional anchor-graph-based methods in terms of efficiency. Experiments on a variety of publicly available datasets with sizes varying from thousands to millions of samples demonstrate the effectiveness of our approach.

The type of neural networks widely used in artificial intelligence applications mixes its computation and memory modules in neuron weights and activities. The previously learned information are stored in network weights. When dealing with complex data, e.g., those possessing diverse content or containing long-sequences, some information stored in the weights can be altered drastically or wiped as the training goes, but they are not necessarily unimportant. External memory is a recent technique proposed to prevent from forgetting significant previously learned information. In this work, we aim at taking advantage of this recent technique to advance the similarity learning task that is critical in many real-world artificial intelligence applications. We propose suitable external memory design supported by extended attention mechanism. Two different kinds of memory modules are proposed so that the similarity learning process can dynamically shift focus over a wide range of diverse content contained by the training data. Effectiveness of the proposed method is demonstrated through evaluations based on different image retrieval tasks and compared against various state-of-the-art algorithms in the field.

We propose a novel framework named ViOCE that integrates ontology-based background knowledge in the form of n-ball concept embeddings into a neural network based vision architecture. The approach consists of two main components: (1) converting symbolic knowledge of an ontology into continuous space by learning n-ball embeddings that capture properties of subsumption and disjointness, (2) guiding the training and inference of a vision model using the learnt embeddings. We propose techniques to measure the quality of n-ball embeddings and evaluate ViOCE using the task of few-shot image classification, where it demonstrates superior performance in two standard benchmarks. We further introduce a metric to use background knowledge to measure the degree of incorrect predictions.

In order to protect vulnerable people in telemarketing, organisations have to investigate the speech recordings to identify them first. Typically, the investigation is manually conducted. As such, the procedure is costly and time-consuming. With an automatic vulnerability detection system, more vulnerable people can be identified and protected. A standard telephone conversation lasts around 5 minutes, the detection system is expected to be able to identify such a potential vulnerable speaker from speech segments. Due to the complexity of the vulnerability definition and the unavailable annotated vulnerability examples, this paper attempts to address the detection problem as three classification tasks: age classification, accent classification and patient/non-patient classification utilising publicly available datasets. In the proposed system, we trained three sub models using acoustic and textual features for each sub task. Each trained model was evaluated on multiple datasets and achieved competitive results compared to a strong baseline (i.e. in-dataset accuracy).

introduction Share on Introduction to Big Multimodal Multimedia Data with Deep Analytics Authors: Yang Wang Hefei University of Technology Hefei University of TechnologyView Profile , Meng Fang Tecent AI Tecent AIView Profile , Joey Tianyi Zhou A-Star A-StarView Profile , Tingting Mu The University of Manchester The University of ManchesterView Profile , Dacheng Tao The UBTECH Sydney Artificial Intelligence Centre, the University of Sydney, Australia The UBTECH Sydney Artificial Intelligence Centre, the University of Sydney, AustraliaView Profile Authors Info & Claims ACM Transactions on Multimedia Computing, Communications, and ApplicationsVolume 17Issue 1sJanuary 2021 Article No.: 1pp 1–3https://doi.org/10.1145/3447530Online:31 March 2021Publication History 0citation212DownloadsMetricsTotal Citations0Total Downloads212Last 12 Months93Last 6 weeks4 Get Citation AlertsNew Citation Alert added!This alert has been successfully added and will be sent to:You will be notified whenever a record that you have chosen has been cited.To manage your alert preferences, click on the button below.Manage my Alerts New Citation Alert!Please log in to your account Save to BinderSave to BinderCreate a New BinderNameCancelCreateExport CitationPublisher SiteGet Access

In general, development of adequately complex mathematical models, such as deep neural networks, can be an effective way to improve the accuracy of learning models. However, this is achieved at the cost of reduced post-hoc model interpretability, because what is learned by the model can become less intelligible and tractable to humans as the model complexity increases. In this paper, we target a similarity learning task in the context of image retrieval, with a focus on the model interpretability issue. An effective similarity neural network (SNN) is proposed to offer not only to seek robust retrieval performance but also to achieve satisfactory post-hoc interpretability. The network is designed by linking the neuron architecture with the organization of a concept tree and by formulating neuron operations to pass similarity information between concepts. Various ways of understanding and visualizing what is learned by the SNN neurons are proposed. We also exhaustively evaluate the proposed approach using a number of relevant datasets against a number of state-of-the-art approaches to demonstrate the effectiveness of the proposed network. Our results show that the proposed approach can offer superior performance when compared against state-of-the-art approaches. Neuron visualization results are demonstrated to support the understanding of the trained neurons.

Sentiment analysis is an important topic concerning identification of feelings, attitudes, emotions and opinions from text. To automate such analysis, a large amount of example text needs to be manually annotated for model training. This is laborious and expensive, but the cross-domain technique is a key solution to reducing the cost by reusing annotated reviews across domains. However, its success largely relies on the learning of a robust common representation space across domains. In the recent years, significant effort has been invested to improve the cross-domain representation learning by designing increasingly more complex and elaborate model inputs and architectures. We support that it is not necessary to increase design complexity as this inevitably consumes more time in model training. Instead, we propose to explore the word polarity and occurrence information through a simple mapping and encode such information more accurately whilst managing lower computational costs. The proposed approach is unique and takes advantage of the stochastic embedding technique to tackle cross-domain sentiment alignment. Its effectiveness is benchmarked with over ten data tasks constructed from two review corpora and it is compared against ten classical and state-of-the-art methods.

In this paper we investigate the problems of class imbalance and irrelevant relationships in Visual Relationship Detection (VRD). State-of-the-art deep VRD models still struggle to predict uncommon classes, limiting their applicability. Moreover, many methods are incapable of properly filtering out background relationships while predicting relevant ones. Although these problems are very apparent, they have both been overlooked so far. We analyse why this is the case and propose modifications to both model and training to alleviate the aforementioned issues, as well as suggesting new measures to complement existing ones and give a more holistic picture of the efficacy of a model.

We propose a set of highly scalable algorithms for the combinatorial data analysis problem of seriating similarity matrices. Seriation consists of finding a permutation of data instances, such that similar instances are nearby in the ordering. Applications of the seriation problem can be found in various disciplines such as in bioinformatics for genome sequencing, data visualization and exploratory data analysis. Our algorithms attempt to minimize certain p-SUM objectives, which also arise in the problem of envelope reduction of sparse matrices. In particular, we present a set of graduated non-convexity algorithms for vector-based relaxations of the general p-SUM problem for $$p \in \left\{ 2, 1, \tfrac{1}{2}\right\} $$ that can scale to very large problem sizes. Different choices of p emphasize global versus local similarity pattern structure. We conduct a number of experiments to compare our algorithms to various state-of-the-art combinatorial optimization methods on real and synthetic datasets. The experimental results demonstrate that compared to other approaches, the proposed algorithms are very competitive and scale well with large problem sizes.

A typical objective of data visualization is to generate low-dimensional plots that maximally convey the information within the data. The visualization output should help the user not only identify the local neighborhood structure of individual samples, but also obtain a global view of the relative positioning and separation between cohorts. Here, we propose a novel visualization framework designed to satisfy these needs. By incorporating additional cohort positioning and discriminative constraints into local neighbor preservation models through the use of computed cohort prototypes, effective control over the arrangements and proximities of data cohorts can be obtained. We introduce various embedding and projection algorithms based on objective functions addressing the different visualization requirements. Their underlying models are optimized effectively using matrix manifold procedures to incorporate the problem constraints. Additionally, to facilitate large-scale applications, a matrix decomposition based model is also proposed to accelerate the computation. The improved capabilities of the new methods are demonstrated using various state-of-the-art dimensionality reduction algorithms. We present many qualitative and quantitative comparisons, on both synthetic problems and real-world tasks of complex text and image data, that show notable improvements over existing techniques.

Descriptive clustering consists of automatically organizing data instances into clusters and generating a descriptive summary for each cluster. The description should inform a user about the contents of each cluster without further examination of the specific instances, enabling a user to rapidly scan for relevant clusters. Selection of descriptions often relies on heuristic criteria. We model descriptive clustering as an auto-encoder network that predicts features from cluster assignments and predicts cluster assignments from a subset of features. The subset of features used for predicting a cluster serves as its description. For text documents, the occurrence or count of words, phrases, or other attributes provides a sparse feature representation with interpretable feature labels. In the proposed network, cluster predictions are made using logistic regression models, and feature predictions rely on logistic or multinomial regression models. Optimizing these models leads to a completely self-tuned descriptive clustering approach that automatically selects the number of clusters and the number of features for each cluster. We applied the methodology to a variety of short text documents and showed that the selected clustering, as evidenced by the selected feature subsets, are associated with a meaningful topical organization.

Motoki Sato, Austin J. Brockmeier, Georgios Kontonatsios, Tingting Mu, John Y. Goulermas, Jun’ichi Tsujii, Sophia Ananiadou. Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics: Volume 1, Long Papers. 2017.

In this work we propose a highly scalable algorithm for solving the combinatorial data analysis problem of seriation. Seriation is a technique for optimizing a permutation of data instances, with respect to some proximity measure such that nearby instances in the linear arrangement are more similar. One consistent objective function for seriation is the 2-SUM minimization problem, which uses the 2-norm between instance locations to penalize non-zero similarity values, and can be written as a quadratic function of the permutation vector. Recently, two convex relaxations of the 2-SUM problem have been proposed, which can be solved as constrained quadratic programs using interior point methods; however, the interior point solvers become expensive when the problem size increases. In this paper we present a graduated non-convexity method for vector-based relaxations of the 2-SUM that yields better approximate solutions and scales to very large problem sizes. We conduct a number of experiments on real and synthetic datasets. The experimental results demonstrate that our proposed algorithm outperforms other approaches that solve the 2-SUM, and is the only competitive approach that can scale to large problem sizes.

Citation screening, an integral process within systematic reviews that identifies citations relevant to the underlying research question, is a time-consuming and resource-intensive task. During the screening task, analysts manually assign a label to each citation, to designate whether a citation is eligible for inclusion in the review. Recently, several studies have explored the use of active learning in text classification to reduce the human workload involved in the screening task. However, existing approaches require a significant amount of manually labelled citations for the text classification to achieve a robust performance. In this paper, we propose a semi-supervised method that identifies relevant citations as early as possible in the screening process by exploiting the pairwise similarities between labelled and unlabelled citations to improve the classification performance without additional manual labelling effort. Our approach is based on the hypothesis that similar citations share the same label (e.g., if one citation should be included, then other similar citations should be included also). To calculate the similarity between labelled and unlabelled citations we investigate two different feature spaces, namely a bag-of-words and a spectral embedding based on the bag-of-words. The semi-supervised method propagates the classification codes of manually labelled citations to neighbouring unlabelled citations in the feature space. The automatically labelled citations are combined with the manually labelled citations to form an augmented training set. For evaluation purposes, we apply our method to reviews from clinical and public health. The results show that our semi-supervised method with label propagation achieves statistically significant improvements over two state-of-the-art active learning approaches across both clinical and public health reviews.

In this paper, we describe an unsupervised measure for quantifying the 'informativeness' of correlation matrices formed from the pairwise similarities or relationships among data instances. The measure quantifies the heterogeneity of the correlations and is defined as the distance between a correlation matrix and the nearest correlation matrix with constant off-diagonal entries. This non-parametric notion generalizes existing test statistics for equality of correlation coefficients by allowing for alternative distance metrics, such as the Bures and other distances from quantum information theory. For several distance and dissimilarity metrics, we derive closed-form expressions of informativeness, which can be applied as objective functions for machine learning applications. Empirically, we demonstrate that informativeness is a useful criterion for selecting kernel parameters, choosing the dimension for kernel-based nonlinear dimensionality reduction, and identifying structured graphs. We also consider the problem of finding a maximally informative correlation matrix around a target matrix, and explore parameterizing the optimization in terms of the coordinates of the sample or through a lower-dimensional embedding. In the latter case, we find that maximizing the Bures-based informativeness measure, which is maximal for centered rank-1 correlation matrices, is equivalent to minimizing a specific matrix norm, and present an algorithm to solve the minimization problem using the norm's proximal operator. The proposed correlation denoising algorithm consistently improves spectral clustering. Overall, we find informativeness to be a novel and useful criterion for identifying non-trivial correlation structure.

In this paper, a novel unsupervised hashing algorithm, referred to as t-USMVH, and its extension to unsupervised deep hashing, referred to as t-UDH, are proposed to support large-scale video-to-video retrieval. To improve robustness of the unsupervised learning, the t-USMVH combines multiple types of feature representations and effectively fuses them by examining a continuous relevance score based on a Gaussian estimation over pairwise distances, and also a discrete neighbor score based on the cardinality of reciprocal neighbors. To reduce sensitivity to scale changes for mapping objects that are far apart from each other, Student t-distribution is used to estimate the similarity between the relaxed hash code vectors for keyframes. This results in more accurate preservation of the desired unsupervised similarity structure in the hash code space. By adapting the corresponding optimization objective and constructing the hash mapping function via a deep neural network, we develop a robust unsupervised training strategy for a deep hashing network. The efficiency and effectiveness of the proposed methods are evaluated on two public video collections via comparisons against multiple classical and the state-of-the-art methods.

Near-duplicate video retrieval (NDVR) has been a significant research task in multimedia given its high impact in applications, such as video search, recommendation, and copyright protection. In addition to accurate retrieval performance, the exponential growth of online videos has imposed heavy demands on the efficiency and scalability of the existing systems. Aiming at improving both the retrieval accuracy and speed, we propose a novel stochastic multiview hashing algorithm to facilitate the construction of a large-scale NDVR system. Reliable mapping functions, which convert multiple types of keyframe features, enhanced by auxiliary information such as video-keyframe association and ground truth relevance to binary hash code strings, are learned by maximizing a mixture of the generalized retrieval precision and recall scores. A composite Kullback-Leibler divergence measure is used to approximate the retrieval scores, which aligns stochastically the neighborhood structures between the original feature and the relaxed hash code spaces. The efficiency and effectiveness of the proposed method are examined using two public near-duplicate video collections and are compared against various classical and state-of-the-art NDVR systems.

Descriptive document clustering aims at discovering clusters of semantically interrelated documents together with meaningful labels to summarize the content of each document cluster. In this work, we propose a novel descriptive clustering framework, referred to as CEDL. It relies on the formulation and generation of 2 types of heterogeneous objects, which correspond to documents and candidate phrases, using multilevel similarity information. CEDL is composed of 5 main processing stages. First, it simultaneously maps the documents and candidate phrases into a common co‐embedded space that preserves higher‐order, neighbor‐based proximities between the combined sets of documents and phrases. Then, it discovers an approximate cluster structure of documents in the common space. The third stage extracts promising topic phrases by constructing a discriminant model where documents along with their cluster memberships are used as training instances. Subsequently, the final cluster labels are selected from the topic phrases using a ranking scheme using multiple scores based on the extracted co‐embedding information and the discriminant output. The final stage polishes the initial clusters to reduce noise and accommodate the multitopic nature of documents. The effectiveness and competitiveness of CEDL is demonstrated qualitatively and quantitatively with experiments using document databases from different application fields.

This work is related to the combinatorial data analysis problem of seriation used for data visualization and exploratory analysis. Seriation re-sequences the data, so that more similar samples or objects appear closer together, whereas dissimilar ones are further apart. Despite the large number of current algorithms to realize such re-sequencing, there has not been a systematic way for analyzing the resulting sequences, comparing them, or fusing them to obtain a single unifying one. We propose a new positional proximity measure that evaluates the similarity of two arbitrary sequences based on their agreement on pairwise positional information of the sequenced objects. Furthermore, we present various statistical properties of this measure as well as its normalized version modeled as an instance of the generalized correlation coefficient. Based on this measure, we define a new procedure for consensus seriation that fuses multiple arbitrary sequences based on a quadratic assignment problem formulation and an efficient way of approximating its solution. We also derive theoretical links with other permutation distance functions and present their associated combinatorial optimization forms for consensus tasks. The utility of the proposed contributions is demonstrated through the comparison and fusion of multiple seriation algorithms we have implemented, using many real-world datasets from different application domains.

Unsupervised Cross-domain Sentiment Classification is the task of adapting a sentiment classifier trained on a particular domain (source domain), to a different domain (target domain), without requiring any labeled data for the target domain. By adapting an existing sentiment classifier to previously unseen target domains, we can avoid the cost for manual data annotation for the target domain. We model this problem as embedding learning, and construct three objective functions that capture: (a) distributional properties ofpivots(i.e., common features that appear in both source and target domains), (b) label constraints in the source domain documents, and (c) geometric properties in the unlabeled documents in both source and target domains. Unlike prior proposals that first learn a lower-dimensional embedding independent of the source domain sentiment labels, and next a sentiment classifier in this embedding, our joint optimisation method learns embeddings that are sensitive to sentiment classification. Experimental results on a benchmark dataset show that by jointly optimising the three objectives we can obtain better performances in comparison to optimising each objective function separately, thereby demonstrating the importance of task-specific embedding learning for cross-domain sentiment classification. Among the individual objective functions, the best performance is obtained by (c). Moreover, the proposed method reports cross-domain sentiment classification accuracies that are statistically comparable to the current state-of-the-art embedding learning methods for cross-domain sentiment classification.

Recent research has shown the effectiveness of rich feature representation for tasks in natural language processing (NLP). However, exceedingly large number of features do not always improve classification performance. They may contain redundant information, lead to noisy feature presentations, and also render the learning algorithms intractable. In this paper, we propose a supervised embedding framework that modifies the relative positions between instances to increase the compatibility between the input features and the output labels and meanwhile preserves the local distribution of the original data in the embedded space. The proposed framework attempts to support flexible balance between the preservation of intrinsic geometry and the enhancement of class separability for both interclass and intraclass instances. It takes into account characteristics of linguistic features by using an inner product‐based optimization template. (Dis)similarity features, also known as empirical kernel mapping, is employed to enable computationally tractable processing of extremely high‐dimensional input, and also to handle nonlinearities in embedding generation when necessary. Evaluated on two NLP tasks with six data sets, the proposed framework provides better classification performance than the support vector machine without using any dimensionality reduction technique. It also generates embeddings with better class discriminability as compared to many existing embedding algorithms.

This paper proposes a novel way for generating reliable low-dimensional features with improved class separability in a kernel-induced feature space. The feature projections rely on a very efficient sequential projection pursuit method, adapted to support nonlinear projections using a new kernel matrix update scheme. This enables the gradual removal of structure from the space of residual dimensions to allow the recovery of multiple projections. An adaptive kernel function is employed to unfold different types of data characteristics. We follow a holistic model selection procedure that, together with the optimal projections, dimensionality, and kernel parameters, additionally optimizes symbolically the projection index that controls the actual measurement of the data interestingness without user interaction. We tackle the underlying complex bi-level optimization model as a mixture of evolutionary and gradient search. The effectiveness of the proposed algorithm over existing approaches is demonstrated with benchmark evaluations and comparisons.

In this paper, we study the co-embedding problem of how to map different types of patterns into one common low-dimensional space, given only the associations (relation values) between samples. We conduct a generic analysis to discover the commonalities between existing co-embedding algorithms and indirectly related approaches and investigate possible factors controlling the shapes and distributions of the co-embeddings. The primary contribution of this work is a novel method for computing co-embeddings, termed the automatic co-embedding with adaptive shaping (ACAS) algorithm, based on an efficient transformation of the co-embedding problem. Its advantages include flexible model adaptation to the given data, an economical set of model variables leading to a parametric co-embedding formulation, and a robust model fitting criterion for model optimization based on a quantization procedure. The secondary contribution of this work is the introduction of a set of generic schemes for the qualitative analysis and quantitative assessment of the output of co-embedding algorithms, using existing labeled benchmark datasets. Experiments with synthetic and real-world datasets show that the proposed algorithm is very competitive compared to existing ones.

In this paper, we study travel time and energy cost prediction at any future departure time for a targeted road segment and vehicle. These two prediction tasks play an important part in the design of advanced driver-assistance systems (ADAS) that can automatically manage battery charging, energy saving, and route planning for fully electric vehicles. Compared with the fundamental problem of travel time prediction, which usually learns from the historical and current data of travel time itself, energy cost prediction is a more complex problem that involves multiple context conditions and vehicle status measured by various time-invariant and time-variant data. We define a general learning problem based on multiple time-invariant and time-variant inputs to unify these two prediction tasks. To solve the defined learning problem, we propose heterogeneous delay embedding (HDE), which extracts an informative feature space for regression analysis and aims at achieving satisfactory prediction for any future departure time. The proposed HDE first categorizes the historical and current data of a time-variant measurement into different types, then incorporates different delay settings for embedding multiple types of time-series data, and finally removes redundant information and noise from the generated features using orthogonal locality preserving projection. Experimental results demonstrate the effectiveness of the proposed method for both short- and long-term predictions of travel time and energy cost.

Route planning for fully electric vehicles (FEVs) must take energy efficiency into account due to limited battery capacity and time-consuming recharging. In addition, the planning algorithm should allow for negative energy costs in the road network due to regenerative braking, which is a unique feature of FEVs. In this paper, we propose a framework for energy-driven and context-aware route planning for FEVs. It has two novel aspects: 1) It is context aware, i.e., the framework has access to real-time traffic data for routing cost estimation; and it is energy driven, i.e., both time and energy efficiency are accounted for; which implies a biobjective nature of the optimization. In addition, in the case of insufficient energy on board, an optimal detour via recharge points is computed. Our main contributions to address these issues can be highlighted as follows: A vehicle-to-vehicle (V2V) communication protocol is proposed to realize the context awareness, and we replace the original biobjective form of optimality with two single-objective forms and propose a constrained A* ( CA*) algorithm to find the solutions. The algorithm maintains a Pareto front while it confines its search by energy constraints. The best recharging detour can be also found using the algorithm. We first compared the performance of the CA* algorithm with other algorithms. We then evaluate the impact of the context awareness on road traffic by simulations using a realistic road network regarding different forms of optimality. Finally, we show that the CA* algorithm can effectively produce optimal recharging detours.

Spectral embedding methods have played a very important role in dimensionality reduction and feature generation in machine learning. Supervised spectral embedding methods additionally improve the classification of labeled data, using proximity information that considers both features and class labels. However, these calculate the proximity information by treating all intraclass similarities homogeneously for all classes, and similarly for all interclass samples. In this paper, we propose a very novel and generic method which can treat all the intra- and interclass sample similarities heterogeneously by potentially using a different proximity function for each class and each class pair. To handle the complexity of selecting these functions, we employ evolutionary programming as an automated powerful formula induction engine. In addition, for computational efficiency and expressive power, we use a compact matrix tree representation equipped with a broad set of functions that can build most currently used similarity functions as well as new ones. Model selection is data driven, because the entire model is symbolically instantiated using only problem training data, and no user-selected functions or parameters are required. We perform thorough comparative experimentations with multiple classification datasets and many existing state-of-the-art embedding methods, which show that the proposed algorithm is very competitive in terms of classification accuracy and generalization ability.

This paper is about supervised and semi-supervised dimensionality reduction (DR) by generating spectral embeddings from multi-output data based on the pairwise proximity information. Two flexible and generic frameworks are proposed to achieve supervised DR (SDR) for multilabel classification. One is able to extend any existing single-label SDR to multilabel via sample duplication, referred to as MESD. The other is a multilabel design framework that tackles the SDR problem by computing weight (proximity) matrices based on simultaneous feature and label information, referred to as MOPE, as a generalization of many current techniques. A diverse set of different schemes for label-based proximity calculation, as well as a mechanism for combining label-based and feature-based weight information by considering information importance and prioritization, are proposed for MOPE. Additionally, we summarize many current spectral methods for unsupervised DR (UDR), single/multilabel SDR, and semi-supervised DR (SSDR) and express them under a common template representation as a general guide to researchers in the field. We also propose a general framework for achieving SSDR by combining existing SDR and UDR models, and also a procedure of reducing the computational cost via learning with a target set of relation features. The effectiveness of our proposed methodologies is demonstrated with experiments with document collections for multilabel text categorization from the natural language processing domain.

The objective of this paper is the design of an engine for the automatic generation of supervised manifold embedding models. It proposes a modular and adaptive data embedding framework for classification, referred to as DEFC, which realizes in different stages including initial data preprocessing, relation feature generation and embedding computation. For the computation of embeddings, the concepts of friend closeness and enemy dispersion are introduced, to better control at local level the relative positions of the intraclass and interclass data samples. These are shown to be general cases of the global information setup utilized in the Fisher criterion, and are employed for the construction of different optimization templates to drive the DEFC model generation. For model identification, we use a simple but effective bilevel evolutionary optimization, which searches for the optimal model and its best model parameters. The effectiveness of DEFC is demonstrated with experiments using noisy synthetic datasets possessing nonlinear distributions and real-world datasets from different application fields.

A key question in medical decision support is how best to visualise a patient database, with especial reference to cohort labelling, whether this is an indicator function for classification or a cluster index. We propose the use of the kernel trick to visualise complete patient databases, in low-dimensional projections, with class labelling, given a non-linear classifier of choice. The results show that this method is useful both to see how individual patient cases relate to each other with reference to the classification boundary, and also to obtain a visual indication of the separation that can be obtained with difference choices of kernel functions.

Plantar lesions induced by biomechanical dysfunction pose a considerable socioeconomic health care challenge, and failure to detect lesions early can have significant effects on patient prognoses. Most of the previous works on plantar lesion identification employed the analysis of biomechanical microenvironment variables like pressure and thermal fields. This paper focuses on foot kinematics and applies kernel principal component analysis (KPCA) for nonlinear dimensionality reduction of features, followed by Fisher's linear discriminant analysis for the classification of patients with different types of foot lesions, in order to establish an association between foot motion and lesion formation. Performance comparisons are made using leave-one-out cross-validation. Results show that the proposed method can lead to approximately 94% correct classification rates, with a reduction of feature dimensionality from 2100 to 46, without any manual preprocessing or elaborate feature extraction methods. The results imply that foot kinematics contain information that is highly relevant to pathology classification and also that the nonlinear KPCA approach has considerable power in unraveling abstract biomechanical features into a relatively low-dimensional pathology-relevant space.

Projection techniques are frequently used as the principal means for the implementation of feature extraction and dimensionality reduction for machine learning applications. A well established and broad class of such projection techniques is the projection pursuit (PP). Its core design parameter is a projection index, which is the driving force in obtaining the transformation function via optimization, and represents in an explicit or implicit way the user's perception of the useful information contained within the datasets. This paper seeks to address the problem related to the design of PP index functions for the linear feature extraction case. We achieve this using an evolutionary search framework, capable of building new indices to fit the properties of the available datasets. The high expressive power of this framework is sustained by a rich set of function primitives. The performance of several PP indices previously proposed by human experts is compared with these automatically generated indices for the task of classification, and results show a decrease in the classification errors.

Abstract: We show that tuning of multiple parameters for a 2‐norm support vector machine (L2‐SVM) could be viewed as an identification problem of a nonlinear dynamic system. Benefiting from the reachable smooth nonlinearity of an L2‐SVM, we propose to employ the extended Kalman filter to tune the kernel and regularization parameters automatically for the L2‐SVM. The proposed method is validated using three public benchmark data sets and compared with the gradient descent approach as well as the genetic algorithm in measures of classification accuracy and computing time. Experimental results demonstrate the effectiveness of the proposed method in higher classification accuracies, faster training speed and less sensitivity to the initial settings.

We propose two variations of the support vector data description (SVDD) with negative samples (NSVDD) that learn a closed spherically shaped boundary around a set of samples in the target class by involving different forms of slack vectors, including the two-norm NSVDD and nu-NSVDD. We extend the NSVDDs to solve the multiclass classification problems based on the distances between the samples and the centers of the learned spherically shaped boundaries in a kernel-defined feature space by using a combination of linear discriminant analysis (LDA) and nearest-neighbor (NN) rule. Extensive simulations are developed with one real-world data set on the automatic monitoring of roller bearings with vibration signals and eight benchmark data sets for both binary and multiclass classification. The benchmark testing results show that our proposed methods provide lower classification error rates and smaller standard deviations with the cross-validation procedure. The two-norm NSVDD with the LDA-NN rule recorded a test accuracy of 100.0% for the binary fault detection of roller bearings and 99.9% for the multiclass classification of roller bearings under six conditions.

We propose a 2-plane learning method for binary classification, named as the strict 2-surface proximal (S2SP) classifier, by seeking two cross proximal planes based on two strict optimization objectives with a "square of sum" optimization factor, of which the nonlinearity is achieved by employing kernel functions. We apply the S2SP classifier for both linear and nonlinear classification to recognize malignant tumors from a set of 57 regions in mammograms, of which 20 are related to malignant tumors and 37 to benign masses. Ten different feature combinations are studied. Experimental results demonstrate that the linear S2SP classifier provides results comparable to those obtained by Fisher linear discriminant analysis (FLDA). For one feature set (FSs), the linear classification performance was significantly improved to 0.97 by using the S2SP classifier, as compared to the FLDA performance of 0.82, in terms of the area under the receiver operating characteristics (ROC) curve. In the case of nonlinear classification, the S2SP classifier with the triangle kernel provided a perfect performance of 1.0 for all of the ten feature combinations, also evaluated in terms of the area under the ROC curve, but with good robustness limited to the setting of the kernel parameter in a certain range.