Latent Representation Learning Research Articles

High efficiency deep image compression via channel-wise scale adaptive latent representation learning

Recent learning based neural image compression methods have achieved impressive rate–distortion (RD) performance via the sophisticated context entropy model, which performs well in capturing the spatial correlations of latent features. However, due to the dependency on the adjacent or distant decoded features, existing methods require an inefficient serial processing structure, which significantly limits its practicability. Instead of pursuing computationally expensive entropy estimation, we propose to reduce the spatial redundancy via the channel-wise scale adaptive latent representation learning, whose entropy coding is spatially context-free and parallelizable. Specifically, the proposed encoder adaptively determines the scale of the latent features via a learnable binary mask, which is optimized with the RD cost. In this way, lower-scale latent representation will be allocated to the channels with higher spatial redundancy, which consumes fewer bits and vice versa. The downscaled latent features could be well recovered with a lightweight inter-channel upconversion module in the decoder. To compensate for the entropy estimation performance degradation, we further develop an inter-scale hyperprior entropy model, which supports the high efficiency parallel encoding/decoding within each scale of the latent features. Extensive experiments are conducted to illustrate the efficacy of the proposed method. Our method achieves bitrate savings of 18.23%, 19.36%, and 27.04% over HEVC Intra, along with decoding speeds that are 46 times, 48 times, and 51 times faster than the baseline method on the Kodak, Tecnick, and CLIC datasets, respectively.

Skip-patching spatial–temporal discrepancy-based anomaly detection on multivariate time series

Anomaly detection in the Industrial Internet of Things (IIoT) is a challenging task that relies heavily on the efficient learning of multivariate time series representations. We introduce Skip-patching and Spatial–Temporal discrepancy mechanisms to improve the efficiency of detecting anomalies. Traditional feature extraction is hindered by redundant information in limited datasets. The situation is that feature generation from stable operational processes results in low-quality representations. To address this challenge, we propose the Skip-Patching mechanism. This approach involves selectively extracting features from partial data patches, prompting the model to learn more meaningful knowledge through self-supervised learning. It also effectively doubles the training sample size by creating independent sub-groups of patches. Despite the complex spatial and temporal relationships in IIoT systems, existing methods mainly extracted features from a single domain, either temporal or spatial (sensor-wise), or simply cascaded two features, i.e., one after one, which limited anomaly detection capabilities. To address this, we introduce the Spatial–Temporal Association Discrepancy component, which leverages discrepancies between spatial and temporal features to enhance latent representation learning. Our Skip-Patching Spatial–Temporal Anomaly Detection (SSAD) framework combines these two components to provide a more diverse and comprehensive learning process. Tested across four multivariate time series anomaly detection benchmarks, SSAD demonstrates superior performance, confirming the efficacy of combining Skip-patching and Spatial–Temporal features to enhance anomaly detection in IIoT systems.

Exploiting the capacity of deep networks only at the training stage for nonlinear black-box system identification

To benefit from the modeling capacity of deep models in system identification without worrying about inference time, this study presents a novel training strategy that uses deep models only during the training stage. For this purpose, two separate models with different structures and goals are employed. The first one is a deep generative model aiming at modeling the distribution of system output(s), called the teacher model, and the second one is a shallow basis function model, named the student model, fed by system input(s) to predict the system output(s). That means these isolated paths must reach the same ultimate target. As deep models show a great performance in modeling highly nonlinear systems, aligning the representation space learned by these two models makes the student model inherit the teacher model’s approximation power. The proposed objective function consists of the objective of each student and teacher model, adding up with a distance penalty between the learned latent representations. The simulation results on three nonlinear benchmarks show a comparative performance with examined deep architectures applied on the same benchmarks. Algorithmic transparency and structure efficiency are also achieved as byproducts.

Infrared-safe energy weighting does not guarantee small nonperturbative effects

Infrared and collinear (IRC) safety has long been used a proxy for robustness when developing new jet substructure observables. This guiding philosophy has been carried into the deep learning era, where IRC-safe neural networks have been used for many jet studies. For graph-based neural networks, the most straightforward way to achieve IRC safety is to weight particle inputs by their energies. However, energy-weighting by itself does not guarantee that perturbative calculations of machine-learned observables will enjoy small nonperturbative corrections. In this paper, we demonstrate the sensitivity of IRC-safe networks to nonperturbative effects, by training an energy flow network (EFN) to maximize its sensitivity to hadronization. We then show how to construct Lipschitz energy flow networks (L-EFNs), which are both IRC safe and relatively insensitive to nonperturbative corrections. We demonstrate the performance of L-EFNs on generated samples of quark and gluon jets, and showcase fascinating differences between the learned latent representations of EFNs and L-EFNs. Published by the American Physical Society 2024

Improved Dual Correlation Reduction Network With Affinity Recovery.

Deep graph clustering, which aims to reveal the underlying graph structure and divide the nodes into different clusters without human annotations, is a fundamental yet challenging task. However, we observe that the existing methods suffer from the representation collapse problem and tend to encode samples with different classes into the same latent embedding. Consequently, the discriminative capability of nodes is limited, resulting in suboptimal clustering performance. To address this problem, we propose a novel deep graph clustering algorithm termed improved dual correlation reduction network (IDCRN) through improving the discriminative capability of samples. Specifically, by approximating the cross-view feature correlation matrix to an identity matrix, we reduce the redundancy between different dimensions of features, thus improving the discriminative capability of the latent space explicitly. Meanwhile, the cross-view sample correlation matrix is forced to approximate the designed clustering-refined adjacency matrix to guide the learned latent representation to recover the affinity matrix even across views, thus enhancing the discriminative capability of features implicitly. Moreover, we avoid the collapsed representation caused by the oversmoothing issue in graph convolutional networks (GCNs) through an introduced propagation regularization term, enabling IDCRN to capture the long-range information with the shallow network structure. Extensive experimental results on six benchmarks have demonstrated the effectiveness and efficiency of IDCRN compared with the existing state-of-the-art deep graph clustering algorithms. The code of IDCRN is released at https://github.com/yueliu1999/IDCRN. Besides, we share a collection of deep graph clustering, including papers, codes, and datasets at https://github.com/yueliu1999/Awesome-Deep-Graph-Clustering.

Unsupervised Imputation of Non-Ignorably Missing Data Using Importance-Weighted Autoencoders

Deep Learning (DL) methods have dramatically increased in popularity in recent years. While its initial success was demonstrated in the classification and manipulation of image data, there has been significant growth in the application of DL methods to problems in the biomedical sciences. However, the greater prevalence and complexity of missing data in biomedical datasets present significant challenges for DL methods. Here, we provide a formal treatment of missing data in the context of Variational Autoencoders (VAEs), a popular unsupervised DL architecture commonly used for dimension reduction, imputation, and learning latent representations of complex data. We propose a new VAE architecture, NIMIWAE, that is one of the first to flexibly account for both ignorable and non-ignorable patterns of missingness in input features at training time. Following training, samples can be drawn from the approximate posterior distribution of the missing data can be used for multiple imputation, facilitating downstream analyses on high dimensional incomplete datasets. We demonstrate through statistical simulation that our method outperforms existing approaches for unsupervised learning tasks and imputation accuracy. We conclude with a case study of an EHR dataset pertaining to 12,000 ICU patients containing a large number of diagnostic measurements and clinical outcomes, where many features are only partially observed.

Low-Rank Tensor Regularized Views Recovery for Incomplete Multiview Clustering.

In real applications, it is often that the collected multiview data contain missing views. Most existing incomplete multiview clustering (IMVC) methods cannot fully utilize the underlying information of missing data or sufficiently explore the consistent and complementary characteristics. In this article, we propose a novel Low-rAnk Tensor regularized viEws Recovery (LATER) method for IMVC, which jointly reconstructs and utilizes the missing views and learns multilevel graphs for comprehensive similarity discovery in a unified model. The missing views are recovered from a common latent representation, and the recovered views conversely improve the learning of shared patterns. Based on the shared subspace representations and recovered complete multiview data, the multilevel graphs are learned by self-representation to fully exploit the consistent and complementary information among views. Besides, a tensor nuclear norm regularizer is introduced to pursue the global low-rank property and explore the interview correlations. An alternating direction minimization algorithm is presented to optimize the proposed model. Moreover, a new initialization method is proposed to promote the effectiveness of our method for latent representation learning and missing data recovery. Extensive experiments demonstrate that our method outperforms the state-of-the-art approaches.

Subspace learning via Hessian regularized latent representation learning with $${l}_{2,0}$$-norm constraint: unsupervised feature selection

Subspace learning via Hessian regularized latent representation learning with $${l}_{2,0}$$-norm constraint: unsupervised feature selection

Multi-view clustering via latent consistency multi-graph fusion

Multi-view clustering (MVC) algorithms are prevalent in machine learning tasks due to their superiority in mining complementary information from multi-view data. Despite their widespread use, existing MVC approaches ignore the presence of inconsistencies in the multi-view features during multi-view data fusion, which leads to suboptimal clustering performance. To address this challenge, the paper presents a novel approach called multi-view clustering via latent consistency multi-graph fusion. Specifically, a novel latent representation learning model is devised to learn a unified representation from multi-view features, simultaneously capturing the high-order relationships within the multi-view data. Additionally, to alleviate the adverse effects caused by the inconsistency multi-view feature, a consistent multi-graph fusion module is proposed to provide consistent constraints for latent representations. To be specific, the graphs that reflect the manifold structure for each view are fused to guide a consistent latent representation learning. After that, an alternating optimization algorithm is designed to update variables iteratively and alternatively, among which the fusion weights are assigned adaptively. Finally, comprehensive experiments conducted on seven widely recognized benchmark multi-view datasets affirm the efficacy of the proposed method, showcasing its superiority over state-of-the-art approaches.

Oscillating latent dynamics in robot systems during walking and reaching.

Sensorimotor control of complex, dynamic systems such as humanoids or quadrupedal robots is notoriously difficult. While artificial systems traditionally employ hierarchical optimisation approaches or black-box policies, recent results in systems neuroscience suggest that complex behaviours such as locomotion and reaching are correlated with limit cycles in the primate motor cortex. A recent result suggests that, when applied to a learned latent space, oscillating patterns of activation can be used to control locomotion in a physical robot. While reminiscent of limit cycles observed in primate motor cortex, these dynamics are unsurprising given the cyclic nature of the robot's behaviour (walking). In this preliminary investigation, we consider how a similar approach extends to a less obviously cyclic behaviour (reaching). This has been explored in prior work using computational simulations. But simulations necessarily make simplifying assumptions that do not necessarily correspond to reality, so do not trivially transfer to real robot platforms. Our primary contribution is to demonstrate that we can infer and control real robot states in a learnt representation using oscillatory dynamics during reaching tasks. We further show that the learned latent representation encodes interpretable movements in the robot's workspace. Compared to robot locomotion, the dynamics that we observe for reaching are not fully cyclic, as they do not begin and end at the same position of latent space. However, they do begin to trace out the shape of a cycle, and, by construction, they are driven by the same underlying oscillatory mechanics.

Empowering Predictive Modeling by GAN-based Causal Information Learning

Generally speaking, we can easily specify many causal relationships in the prediction tasks of ubiquitous computing, such as human activity prediction, mobility prediction, and health prediction. However, most of the existing methods in these fields failed to take advantage of this prior causal knowledge. They typically make predictions only based on correlations in the data, which hinders the prediction performance in real-world scenarios, because a distribution shift between training data and testing data generally exists. To fill in this gap, we proposed a Generative Adversarial Network (GAN)-based Causal Information Learning prediction framework, which can effectively leverage causal information to improve the prediction performance of existing ubiquitous computing deep learning models. Specifically, faced with a unique challenge that the treatment variable, referring to the intervention that influences the target in a causal relationship, is generally continuous in ubiquitous computing, the framework employs a representation learning approach with a GAN-based deep learning model. By projecting all variables except the treatment into a latent space, it effectively minimizes confounding bias and leverages the learned latent representation for accurate predictions. In this way, it deals with the continuous treatment challenge, and in the meantime, it can be easily integrated with existing deep learning models to lift their prediction performance in practical scenarios with causal information. Extensive experiments on two large-scale real-world datasets demonstrate its superior performance over multiple state-of-the-art baselines. We also propose an analytical framework together with extensive experiments to empirically show that our framework achieves better performance gain under two conditions: when the distribution differences between the training data and the testing data are more significant and when the treatment effects are larger. Overall, this work suggests that learning causal information is a promising way to improve the prediction performance of ubiquitous computing tasks. We open both our dataset and code 1 and call for more research attention in this area.

Latent Representation Learning for Geospatial Entities

Representation learning has been instrumental in the success of machine learning, offering compact and performant data representations for diverse downstream tasks. In the spatial domain, it has been pivotal in extracting latent patterns from various data types, including points, polylines, polygons, and networked structures. However, existing approaches often fall short of explicitly capturing both semantic and spatial information, relying on proxies and synthetic features. This paper presents GeoNN, a novel graph neural network-based model designed to learn spatially-aware embeddings for geospatial entities. GeoNN leverages edge features generated from geodesic functions, dynamically selecting relevant features based on relative locations. It introduces both transductive (GeoNN-T) and inductive (GeoNN-I) models, ensuring effective encoding of geospatial features and scalability with entity changes. Extensive experiments demonstrate GeoNN’s effectiveness in location-sensitive superpixel-based graphs and real-world points of interest, outperforming baselines across various evaluation measures.

A principled framework for explainable multimodal disentanglement

Learning effective representations for data from multiple modalities is crucial in machine learning. Recent efforts focus on learning latent representations that integrate information from various modalities. These approaches generally assume simple or implicit relationships between different modalities and as a result are not able to accurately and explicitly depict the correlations among these modalities and lack explainability. To address this, we propose definitions and conditions for unsupervised multimodal disentanglement, offering guidelines for explicit disentanglement between modalities to enhance explainability. Furthermore, we have derived a novel objective function to explicitly separate multimodal data into components shared across modalities and components exclusive to each modality. The explicit guaranteed disentanglement is of great potential for downstream tasks. Benefiting from a cleverly designed network structure, we can visualize these disentangled representations, providing intuitive explainability. Experiments on a variety of multimodal datasets demonstrate that our objective can effectively disentangle information from different modalities while satisfying the disentangling conditions.

Long-term causal effects estimation via latent surrogates representation learning

Estimating long-term causal effects based on short-term surrogates is a significant but challenging problem in many real-world applications such as marketing and medicine. Most existing methods estimate causal effects in an idealistic and simplistic manner — disregarding unobserved surrogates and treating all short-term outcomes as surrogates. However, such methods are not well-suited to real-world scenarios where the partially observed surrogates are mixed with the proxies of unobserved surrogates among short-term outcomes. To address this issue, we develop our flexible method called LASER to estimate long-term causal effects in a more realistic situation where the surrogates are either observed or have observed proxies. In LASER, we employ an identifiable variational autoencoder to learn the latent surrogate representation by using all the surrogate candidates without the need to distinguish observed surrogates or proxies of unobserved surrogates. With the learned representation, we further devise a theoretically guaranteed and unbiased estimation of long-term causal effects. Extensive experimental results on the real-world and semi-synthetic datasets demonstrate the effectiveness of our proposed method.

Cluster structure augmented deep nonnegative matrix factorization with low-rank tensor learning

Clustering approaches based on deep nonnegative matrix factorization have received significant attention because it can learn hierarchical semantics from data in a layer-wise manner. However, latent representation learning and clustering are two separate processes in these models, leading to information loss and sub-optimal clusterings. Furthermore, if data contain complex structure or noise, the pre-defined graph in data space will not be precise enough for graph-regularized deep nonnegative matrix factorization models. Moreover, most of them maintain the similarity graph at the top layer, neglecting the information covered by other layers. In this paper, Cluster Structure Augmented Deep Nonnegative Matrix Factorization with Low-rank Tensor Learning is proposed. First, as representations at different layers cover different data abstractions, a learning mechanism is leveraged to generate multiple local similarity graphs based on representations from different layers, thus maintaining diversity across layers. Then, a low-rank third-order tensor is constructed by stacking these graphs to capture the high-order consistency across layers, thus capturing the intrinsic property of data. Third, clustering is integrated into the framework, allowing the cluster structure to facilitate multi-layer matrix factorization and graph structure learning, thereby generating discriminative representations. Experimental results on several datasets demonstrate that the proposed model outperforms several state-of-the-art methods.

Multi-view clustering via pseudo-label guide learning and latent graph structure recovery

Multi-view clustering (MvC) accomplishes sample classification tasks by exploring information from different views. Currently, researchers have paid greater attention to graph-based MvC methods. However, most existing methods only consider the original graph structure and pay relatively little attention to the graph structure in the latent space. In addition, most methods need to pay more attention to the consistency of information on different labels. Otherwise, this may lead to the loss of label information. This paper presents a new multi-view clustering framework to address the above issues. The proposed method considers both the information in the latent space and the original data space, which firstly obtains the pseudo-label by latent representation learning and then lets the pseudo-label guide the learning of the complementary information between the raw data views. To ensure the integrity of the data structure, a latent graph structure recovery strategy is designed in the latent space. Finally, an enhanced label fusion strategy is designed to fusion the different types of labels, yielding an information-rich label matrix for clustering. Experimental results demonstrate the proposed method’s effectiveness compared to other advanced approaches.

Multi-label feature selection via latent representation learning and dynamic graph constraints

As an effective method to deal with the curse of dimensionality, multi-label feature selection aims to select the most representative subset of features by eliminating unfavorable features. Although great progress has been made in this field, how to mine adequate supervisory information from multi-label data remains a key challenge. Compared to the latent information of instances, the latent information of instance relevance contains both the basic information of instances and the latent relevance between instances. Base on this knowledge, we propose a novel multi-label feature selection method named LRDG that explores latent representation learning and dynamic graph constraints. Specifically, we introduce the latent representation of instance relevance as supervisory information for pseudo-label learning, and minimize information loss during pseudo-label learning by means of the label manifold, the non-negative constraints, and the minimization of the Frobenius norm between pseudo-labels and ground-truth labels. In addition, considering the shortcomings brought by traditional graph regularization, we propose to use the dynamic graph constructed from low-dimensional pseudo-labels to constrain feature weights. Extensive experiments on various multi-label datasets demonstrate the effectiveness of the proposed method. The code is available at https://github.com/yunbao520/LRDG.

Traceable Group-Wise Self-Optimizing Feature Transformation Learning: A Dual Optimization Perspective

Feature transformation aims to reconstruct an effective representation space by mathematically refining the existing features. It serves as a pivotal approach to combat the curse of dimensionality, enhance model generalization, mitigate data sparsity, and extend the applicability of classical models. Existing research predominantly focuses on domain knowledge-based feature engineering or learning latent representations. However, these methods, while insightful, lack full automation and fail to yield a traceable and optimal representation space. An indispensable question arises: Can we concurrently address these limitations when reconstructing a feature space for a machine learning task? Our initial work took a pioneering step towards this challenge by introducing a novel self-optimizing framework. This framework leverages the power of three cascading reinforced agents to automatically select candidate features and operations for generating improved feature transformation combinations. Despite the impressive strides made, there was room for enhancing its effectiveness and generalization capability. In this extended journal version, we advance our initial work from two distinct yet interconnected perspectives: 1) We propose a refinement of the original framework, which integrates a graph-based state representation method to capture the feature interactions more effectively and develop different Q-learning strategies to alleviate Q-value overestimation further. 2) We utilize a new optimization technique (actor-critic) to train the entire self-optimizing framework in order to accelerate the model convergence and improve the feature transformation performance. Finally, to validate the improved effectiveness and generalization capability of our framework, we perform extensive experiments and conduct comprehensive analyses. These provide empirical evidence of the strides made in this journal version over the initial work, solidifying our framework’s standing as a substantial contribution to the field of automated feature transformation. To improve the reproducibility, we have released the associated code and data by the Github link https://github.com/coco11563/TKDD2023_code.

Prediction of drug-disease associations based on reinforcement symmetric metric learning and graph convolution network.

Accurately identifying novel indications for drugs is crucial in drug research and discovery. Traditional drug discovery is costly and time-consuming. Computational drug repositioning can provide an effective strategy for discovering potential drug-disease associations. However, the known experimentally verified drug-disease associations is relatively sparse, which may affect the prediction performance of the computational drug repositioning methods. Moreover, while the existing drug-disease prediction method based on metric learning algorithm has achieved better performance, it simply learns features of drugs and diseases only from the drug-centered perspective, and cannot comprehensively model the latent features of drugs and diseases. In this study, we propose a novel drug repositioning method named RSML-GCN, which applies graph convolutional network and reinforcement symmetric metric learning to predict potential drug-disease associations. RSML-GCN first constructs a drug-disease heterogeneous network by integrating the association and feature information of drugs and diseases. Then, the graph convolutional network (GCN) is applied to complement the drug-disease association information. Finally, reinforcement symmetric metric learning with adaptive margin is designed to learn the latent vector representation of drugs and diseases. Based on the learned latent vector representation, the novel drug-disease associations can be identified by the metric function. Comprehensive experiments on benchmark datasets demonstrated the superior prediction performance of RSML-GCN for drug repositioning.

A Novel Feature Engineering Method Based on Latent Representation Learning for Radiomics: Application in NSCLC Subtype Classification.

Radiomics refers to the high-throughput extraction of quantitative features from medical images, and is widely used to construct machine learning models for the prediction of clinical outcomes, while feature engineering is the most important work in radiomics. However, current feature engineering methods fail to fully and effectively utilize the heterogeneity of features when dealing with different kinds of radiomics features. In this work, latent representation learning is first presented as a novel feature engineering approach to reconstruct a set of latent space features from original shape, intensity and texture features. This proposed method projects features into a subspace called latent space, in which the latent space features are obtained by minimizing a unique hybrid loss function including a clustering-like loss and a reconstruction loss. The former one ensures the separability among each class while the latter one narrows the gap between the original features and latent space features. Experiments were performed on a multi-center non-small cell lung cancer (NSCLC) subtype classification dataset from 8 international open databases. Results showed that compared with four traditional feature engineering methods (baseline, PCA, Lasso and L2,1-norm minimization), latent representation learning could significantly improve the classification performance of various machine learning classifiers on the independent test set (all p<0.001). Further on two additional test sets, latent representation learning also showed a significant improvement in generalization performance. Our research shows that latent representation learning is a more effective feature engineering method, which has the potential to be used as a general technology in a wide range of radiomics researches.