Entropy Regularization Research Articles

Consistency Preservation and Feature Entropy Regularization for GAN Based Face Editing

Consistency Preservation and Feature Entropy Regularization for GAN Based Face Editing

Partial Domain Adaptation for Scene Classification From Remote Sensing Imagery

Although domain adaptation approaches have been proposed to tackle cross-regional, multitemporal, and multisensor remote sensing applications since they do not require any human interpretation in the target domain, most current works assume identical label space across the source and the target domains. However, in real-world applications, we often transfer knowledge from a large-scale dataset with rich annotations to a small-scale target dataset with scarcity of labels. In most cases, the label space of the source domain is usually large enough to subsume that of the target domain, which is termed partial domain adaptation. In this article, we propose a new partial domain adaptation algorithm for remote sensing scene classification and our proposed method contains three major parts. First, we employ a progressive auxiliary domain module to alleviate the negative transfer effect caused by outlier classes. Second, we adopt an improved domain adversarial neural network (DANN) with multiweights to better encourage domain confusion. Last but not least, we design an attentive complement entropy regularization to improve the prediction confidence for samples and avoid untransferable samples (such as the samples belonging to outlier classes in the source domain) being mistakenly classified. We collect three common remote sensing datasets to evaluate our proposed method. Our method achieves an average accuracy of 79.36%, which considerably outperforms other state-of-the-art partial domain adaptation methods with an average accuracy improvement of 1.90%–12.45% and attaining a 13.67% gain compared to the straightforward deep learning model (ResNet-50). The experiment results indicate that our approach shows promising prospects for solving more general and practical domain adaptation problems where the label space of the source domain subsumes that of the target domain.

A Novel Combined Regularization for Identifying Random Load Sources on Coal‐Rock Structure

This study discusses the practical engineering problem of determining random load sources on coal‐rock structures. A novel combined regularization technique combining mollification method (MM) and discrete regularization (DR), which was called MM‐DR technique, was proposed to reconstruct random load sources on coal‐rock structures. MM‐DR technology is compared with DR, Tikhonov regularization (TR), and maximum entropy regularization (MER) in load reconstruction. The results show that the reconstructed random load sources are more consistent with the real load sources using MM‐DR technique combined with particle swarm optimization (PSO) and L‐curve method, which was named as PSO‐L method, and selecting optimal ω value of kernel function is beneficial to overcome the ill‐posed of random load sources reconstruction and to obtain the stable and approximate solutions. The method proposed in this study provides a theoretical basis for the recognition of coal‐rock interface.

Relative entropy-regularized optimal transport on a graph: a new algorithm and an experimental comparison

The present work investigates a new relative entropy-regularized algorithm for solving the optimal transport on a graph problem within the randomized shortest paths formalism. More precisely, a unit flow is injected into a set of input nodes and collected from a set of output nodes with specified marginals, while minimizing the expected transportation cost, together with a paths-based relative entropy regularization term, providing a randomized routing policy. The main advantage of this new formulation is the fact that it can easily accommodate edge flow capacity constraints which commonly occur in real-world problems. The resulting optimal routing policy, i.e., the probability distribution of following an edge in each node, is Markovian and is computed after constraining the input and output flows to the prescribed marginal probabilities. In addition, experimental comparisons with other recently developed techniques show that the distance measure between nodes derived from the introduced model provides competitive results on semi-supervised classification tasks.

Entropy regularization methods for parameter space exploration

Entropy regularization is an important approach to improve exploration and enhance policy stability for reinforcement learning. However, in previous study, entropy regularization is applied only to action spaces. In this paper, we apply entropy regularization to parameter spaces. We use learnable noisy layers to parameterize the policy network to obtain a learnable entropy. Also, we derive the expression for the entropy of the noisy parameter and an upper bound on the joint entropy. Based on these, we propose a model-free method named deep pseudo deterministic policy gradients based on entropy regularization (DPGER). This method maximizes the entropy of each noisy parameter in the early learning process to promote exploration, and minimizes the joint entropy of the noisy parameters in the later learning process to facilitate the formation of stable policies. We test our method on four Mujoco environments with five random seeds. The results show that our method brings better performance compared to previous methods.

An architecture entropy regularizer for differentiable neural architecture search

Differentiable architecture search (DARTS) is one of the prevailing paradigms of neural architecture search (NAS) due to allowing efficient gradient-based optimization during the search phase. However, its poor stability and generalizability are intolerable. We argue that the crux is the locally optimal architecture parameter caused by a dilemma, which is that the solutions to the Matthew effect and discretization discrepancy are inconsistent. To escape from the dilemma, we propose an architecture entropy to measure the discrepancy of the architecture parameters of different candidate operations and use it as a regularizer to control the learning of architecture parameters. Extensive experiments show that an architecture entropy regularizer with a negative or positive coefficient can effectively solve one side of the contradiction respectively, and the regularizer with a variable coefficient can relieve DARTS from the dilemma. Experimental results demonstrate that our architecture entropy regularizer can significantly improve different differentiable NAS algorithms on different datasets and different search spaces. Furthermore, we also achieve more accurate and more robust results on CIFAR-10 and ImageNet. The code is publicly available at https://github.com/kunjing96/DARTS-AER.

Entropy regularized actor-critic based multi-agent deep reinforcement learning for stochastic games

Multi-agent reinforcement learning (MARL) is an abstract framework modeling a dynamic environment that involves multiple learning and decision-making agents, each of which tries to maximize her cumulative reward. In MARL, each agent discovers a strategy alongside others and adapts her policy in response to the behavioural changes of others. A fundamental difficulty faced by MARL is that every agent is dynamically learning and changing to improve her reward, making the whole system unstable and agents’ policies difficult to converge. In this paper, we introduce the entropy regularizer into the Bellman equation and utilize Lagrange approach to optimize the entropy regularizer. We then propose a MARL algorithm based on the maximum entropy principle and the actor-critic method. This algorithm follows the policy gradient approach and uses a policy network and a value network. We call it Multi-Agent Deep Soft Policy Gradient (MADSPG). Then by using the Lagrange approach and dynamic minimax optimization, we propose the AUTO-MADSPG algorithm with an automatically adjusted entropy regularizer. These algorithms make multi-agent learning more stable while sufficient exploration is guaranteed. Finally, we also incorporate MADSPG with the recently proposed opponent modeling component into an integrated framework. This framework outperforms many state-of-the-art MARL algorithms in conventional cooperative and competitive game settings.

Domain adaptive tree crown detection using high-resolution remote sensing images

Rapid and accurate tree crown detection is significant to forestry management and precision forestry. However, the variability of the data is challenging for traditional deep learning-based methods in cross-regional scenarios. To solve the problem of low accuracy of existing methods due to the distribution characteristics of tree crowns and the special background in which they are located, we propose a multi-adversarial domain adaptive (DA) crown detection model, which saves much of labeling efforts needed for covering different features in different regions. First, we embed a multilevel transferable attention mechanism to extract the multiscale transferable features of the tree crown and combat the negative transfer caused by the background. Second, a multi-adversarial instance alignment module with entropy regularization and a weighting factor is designed to match the cross-regional data distributions to reduce the misdetection and misidentification of complex backgrounds. The above-mentioned adaptations are integrated into a two-stage object detector to realize end-to-end training. We conduct comprehensive experiments with four transfer tasks. Our method improves F1-score to 95.68%, 83.12%, 78.56%, and 66.89% and performs 3.61% and 15.38% better than the state-of-the-art domain adaptation approaches without extra inference costs. The results demonstrate the great potential of our method for DA tree crown detection, which provides a reference for further research on forestry detection.

Learning-Based Resource Allocation in Heterogeneous Ultradense Network

Learning-based resource allocation (LRA) is envisioned as an integral element of 6G. This article proposes a novel learning-based paradigm to address resource allocation problems in heterogeneous ultradense networks (HUDNs). Our paradigm is a highly efficient realization for utilizing the inherence properties in HUDN, which comprise the local validity and correlation attenuation. Concretely, we formulate the HUDNs as a heterogeneous bipartite graph model and propose the corresponding heterogeneous bipartite graph neural network (HBGNN). The local sampling characteristic of HBGNN matches the inherence properties. Meanwhile, our paradigm combines data-driven and model-driven learnings and employs online and offline trainings. Hence, two of LRA’s obstacles: 1) the overreliance on the perfect data set and 2) the low calculation efficiency are mitigated, and the realizability of our paradigm is improved. Besides, entropy regularization is utilized to guarantee the effectiveness of exploration in the configuration space. We apply our approach to a representative resource allocation problem, the jointly user association (UA) and power allocation (JUAPA) problem. We formulate JUAPA as a combination classification and regression problem and adopt a dynamical hyperparameter output layer to address the discrete variable of UA. Simulation results demonstrate that the proposed method has better performance and higher computational efficiency than traditional optimization algorithms.

Model-Based Imitation Learning Using Entropy Regularization of Model and Policy

Approaches based on generative adversarial networks for imitation learning are promising because they are sample efficient in terms of expert demonstrations. However, training a generator requires many interactions with the actual environment because model-free reinforcement learning is adopted to update a policy. To improve the sample efficiency using model-based reinforcement learning, we propose model-based Entropy-Regularized Imitation Learning (MB-ERIL) under the entropy-regularized Markov decision process to reduce the number of interactions with the actual environment. MB-ERIL uses two discriminators. A policy discriminator distinguishes the actions generated by a robot from expert ones, and a model discriminator distinguishes the counterfactual state transitions generated by the model from the actual ones. We derive structured discriminators so that the learning of the policy and the model is efficient. Computer simulations and real robot experiments show that MB-ERIL achieves a competitive performance and significantly improves the sample efficiency compared to baseline methods.

Self-supervised knowledge distillation for complementary label learning

In this paper, we tackle a new learning paradigm called learning from complementary labels, where the training data specifies classes that instances do not belong to, instead of the accuracy labels. In general, it is more efficient to collect the complementary labels compared with collecting the supervised ones, with no need for selecting the correct one from a number of candidates. While current state-of-the-art methods design various loss functions to train competitive models by the limited supervised information, they overlook learning from the data and model themselves, which always contain fruitful information that can improve the performance of complementary label learning. In this paper, we propose a novel learning framework, which seamlessly integrates self-supervised and self-distillation to complementary learning. Based on the general complementary learning framework, we employ an entropy regularization term to guarantee the network outputs exhibit a sharper state. Then, to intensively learn information from the data, we leverage the self-supervised learning based on rotation and transformation operations as a plug-in auxiliary task to learn better transferable representations. Finally, knowledge distillation is introduced to further extract the “dark knowledge” from a network to guide the training of a student network. In the extensive experiments, our method surprisingly demonstrates compelling performance in accuracy over several state-of-the-art approaches.

Multiview Latent Representation Learning with Feature Diversity for Clustering

To analyze and manage data, multiple features are extracted for robust and accurate description. Taking account of the complementary nature of multiview data, it is urgent to find a low-dimensional compact representation by leveraging multiple views and generating a better performance for clustering. In this paper, we present multiview latent representation learning with feature diversity for clustering (MvLRFD). To get rid of noise and maintain the information of each view, matrix factorization is adopted to get the latent representation. For fusing the diverse latent representations, we argue that the final representation is constructed by concatenating the latent representation of each view. With the help of the Hilbert–Schmidt Independence Criterion (HSIC), the diversity of features in the final representation is maximized to exploit the complementary information. A new coregularization strategy is introduced to maximize the distance between the similarity between data of the final representation and the latent one. The partition entropy regularization is adopted to control the uniformity level of the value of the similarity matrix and the weight of each view. To verify the availability of MvLRFD compared to state-of-the-art algorithms, several experiments are conducted on four real datasets.

Efficient Optimal Transport Algorithm by Accelerated Gradient Descent

Optimal transport (OT) plays an essential role in various areas like machine learning and deep learning. However, computing discrete optimal transport plan for large scale problems with adequate accuracy and efficiency is still highly challenging. Recently, methods based on the Sinkhorn algorithm add an entropy regularizer to the prime problem and get a trade off between efficiency and accuracy. In this paper, we propose a novel algorithm to further improve the efficiency and accuracy based on Nesterov's smoothing technique. Basically, the non-smooth c-transform of the Kantorovich potential is approximated by the smooth Log-Sum-Exp function, which finally smooths the original non-smooth Kantorovich dual functional. The smooth Kantorovich functional can be optimized by the fast proximal gradient algorithm (FISTA) efficiently. Theoretically, the computational complexity of the proposed method is lower than current estimation of the Sinkhorn algorithm in terms of the precision. Empirically, compared with the Sinkhorn algorithm, our experimental results demonstrate that the proposed method achieves faster convergence and better accuracy with the same parameter.

Economic Development Analysis of China’s Coastal Industries Based on Deep Learning and SDA Model Weighting

China’s coast is the main spatial carrier and important geographical unit for future industrial development in China. We propose a weighted average decomposition method based on three basic problems in the theory of SDA model decomposition method, investigate the relevant properties of this method, and prove that two decomposition methods widely used in the literature—bipolar decomposition method and midpoint power decomposition method—are the approximate solutions of this decomposition method. Based on the idea of the Wasserstein distance algorithm for machine learning, the Wasserstein distance algorithm and its solution are improved by using the matrix expansion Sinkhorn algorithm and the entropy regularization constraint method, and the industry copolymerization index is constructed through hypothesis testing and Monte Carlo simulation to measure the level of industry copolymerization in coastal China. The results show that the level of industry copolymerisation within the same quantile industry is greater than that between industries across quantile industries in China’s coasts.

Maximum Entropy Technique and Regularization Functional for Determining the Pharmacokinetic Parameters in DCE-MRI

This paper aims to solve the arterial input function (AIF) determination in dynamic contrast-enhanced MRI (DCE-MRI), an important linear ill-posed inverse problem, using the maximum entropy technique (MET) and regularization functionals. In addition, estimating the pharmacokinetic parameters from a DCE-MR image investigations is an urgent need to obtain the precise information about the AIF–the concentration of the contrast agent on the left ventricular blood pool measured over time. For this reason, the main idea is to show how to find a unique solution of linear system of equations generally in the form of y=Ax+b, named an ill-conditioned linear system of equations after discretization of the integral equations, which appear in different tomographic image restoration and reconstruction issues. Here, a new algorithm is described to estimate an appropriate probability distribution function for AIF according to the MET and regularization functionals for the contrast agent concentration when applying Bayesian estimation approach to estimate two different pharmacokinetic parameters. Moreover, by using the proposed approach when analyzing simulated and real datasets of the breast tumors according to pharmacokinetic factors, it indicates that using Bayesian inference—that infer the uncertainties of the computed solutions, and specific knowledge of the noise and errors—combined with the regularization functional of the maximum entropy problem, improved the convergence behavior and led to more consistent morphological and functional statistics and results. Finally, in comparison to the proposed exponential distribution based on MET and Newton’s method, or Weibull distribution via the MET and teaching–learning-based optimization (MET/TLBO) in the previous studies, the family of Gamma and Erlang distributions estimated by the new algorithm are more appropriate and robust AIFs.

A Semisupervised End-to-End Framework for Transportation Mode Detection by Using GPS-Enabled Sensing Devices

As an essential component of Internet of Things, GPS-enabled devices record tremendous digital traces, which provide a great convenience for understanding human mobility. How to discover transportation modes efficiently from such valuable sources has come into the spotlight. In this article, the transportation mode detection is treated as a dense classification task, and a similarity entropy-based encoder-decoder (SEED) model is proposed. We first design an encoder-decoder backbone for end-to-end mode detection. Then, a semi-supervised learning module based on similarity entropy is proposed to exploit numerous unlabeled data. Specifically, we stack several convolutional layers as an encoder to capture hierarchical features from fixed-length trajectories, and then adopt transposed convolutional layers as a decoder. For a semi-supervised module, inspired by entropy regularization, we use the <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${K}$ </tex-math></inline-formula> -Means algorithm to cluster prototype vectors from the encoder’s predictions. We then fine-tune the encoder by sharpening the similarity distribution between unlabeled predictions and prototypes, aiming to make the former close to one prototype only while staying away from others. A majority-voting post-processing method is used to alleviate jitter impact when inferring. The Experimental results show that SEED significantly outperforms segmentation-then-inference methods. Furthermore, the similarity entropy-based module can improve the generalization performance of the model, and the metrics such as intersection over union can be increased by 5% over baselines. All of these verify the superiority of our method.

Multi-Model Adaptation Learning With Possibilistic Clustering Assumption for EEG-Based Emotion Recognition.

In machine learning community, graph-based semi-supervised learning (GSSL) approaches have attracted more extensive research due to their elegant mathematical formulation and good performance. However, one of the reasons affecting the performance of the GSSL method is that the training data and test data need to be independently identically distributed (IID); any individual user may show a completely different encephalogram (EEG) data in the same situation. The EEG data may be non-IID. In addition, noise/outlier sensitiveness still exist in GSSL approaches. To these ends, we propose in this paper a novel clustering method based on structure risk minimization model, called multi-model adaptation learning with possibilistic clustering assumption for EEG-based emotion recognition (MA-PCA). It can effectively minimize the influence from the noise/outlier samples based on different EEG-based data distribution in some reproduced kernel Hilbert space. Our main ideas are as follows: (1) reducing the negative impact of noise/outlier patterns through fuzzy entropy regularization, (2) considering the training data and test data are IID and non-IID to obtain a better performance by multi-model adaptation learning, and (3) the algorithm implementation and convergence theorem are also given. A large number of experiments and deep analysis on real DEAP datasets and SEED datasets was carried out. The results show that the MA-PCA method has superior or comparable robustness and generalization performance to EEG-based emotion recognition.

A new weakly supervised approach for ALS point cloud semantic segmentation

Although novel point cloud semantic segmentation schemes that continuously surpass state-of-the-art results exist, the success of learning an effective model typically relies on the availability of abundant labeled data. However, data annotation is a time-consumng and labor-intensive task, particularly for large-scale airborne laser scanning (ALS) point clouds involving multiple classes in urban areas. Therefore, simultaneously obtaining promising results while significantly reducing labeling is crucial. In this study, we propose a deep-learning-based weakly supervised framework for the semantic segmentation of ALS point clouds. This is to exploit implicit information from unlabeled data subject to incomplete and sparse labels. Entropy regularization is introduced to penalize class overlap in the predictive probability. Additionally, a consistency constraint is designed to improve the robustness of the predictions by minimizing the difference between the current and ensemble predictions. Finally, we propose an online soft pseudo-labeling strategy to create additional supervisory sources in an efficient and nonparametric manner. Extensive experimental analysis using three benchmark datasets demonstrates that our proposed method significantly boosts the classification performance without compromising the computational efficiency, considering the sparse point annotations. It outperforms the current weakly supervised methods and achieves a result comparable to that of full supervision competitors. Considering the ISPRS Vaihingen 3D data, using only 1‰ labels, our method achieved an overall accuracy of 83.0% and an average F1 score of 70.0%. These increased by 6.9% and 12.8%, respectively, compared to the model trained only using sparse label information.

Entropy-Enhanced Attention Model for Explanation Recommendation

Most of the existing recommendation systems using deep learning are based on the method of RNN (Recurrent Neural Network). However, due to some inherent defects of RNN, recommendation systems based on RNN are not only very time consuming but also unable to capture the long-range dependencies between user comments. Through the sentiment analysis of user comments, we can better capture the characteristics of user interest. Information entropy can reduce the adverse impact of noise words on the construction of user interests. Information entropy is used to analyze the user information content and filter out users with low information entropy to achieve the purpose of filtering noise data. A self-attention recommendation model based on entropy regularization is proposed to analyze the emotional polarity of the data set. Specifically, to model the mixed interactions from user comments, a multi-head self-attention network is introduced. The loss function of the model is used to realize the interpretability of recommendation systems. The experiment results show that our model outperforms the baseline methods in terms of MAP (Mean Average Precision) and NDCG (Normalized Discounted Cumulative Gain) on several datasets, and it achieves good interpretability.

Robust deep kernel-based fuzzy clustering with spatial information for image segmentation

Clustering algorithms with deep neural network has attracted wide attention to scholars. A deep fuzzy K-means clustering algorithm model on adaptive loss function and entropy regularization (DFKM) is proposed by combining automatic encoder and clustering algorithm. Although it introduces adaptive loss function and entropy regularization to improve the robustness of the model, its segmentation effect is not ideal for high noise. The research purpose of this paper is to focus on the anti-noise performance of image segmentation. Therefore, on the basis of DFKM, this paper focus on image segmentation, combine neighborhood median and mean information of current pixel, introduce neighborhood information of membership degree, and extend Euclidean distance to kernel space by using kernel function, propose a dual-neighborhood information constrained deep fuzzy clustering based on kernel function (KDFKMS). A large number of experimental results show that compared with DFKM and classical image segmentation algorithms, this algorithm has stronger anti-noise robustness.