Clustering based Contrastive Learning for Improving Face Representations

This article presents a novel method for face clustering in videos using a video-centralised transformer. Previous works often employed contrastive learning to learn frame-level representation and used average pooling to aggregate the features along the temporal dimension. This approach may not fully capture the complicated video dynamics. In addition, despite the recent progress in video-based contrastive learning, few have attempted to learn a self-supervised clustering-friendly face representation that benefits the video face clustering task. To overcome these limitations, our method employs a transformer to directly learn video-level representations that can better reflect the temporally-varying property of faces in videos, while we also propose a video-centralised self-supervised framework to train the transformer model. We also investigate face clustering in egocentric videos, a fast-emerging field that has not been studied yet in works related to face clustering. To this end, we present and release the first large-scale egocentric video face clustering dataset named EasyCom-Clustering. We evaluate our proposed method on both the widely used Big Bang Theory (BBT) dataset and the new EasyCom-Clustering dataset. Results show the performance of our video-centralised transformer has surpassed all previous state-of-the-art methods on both benchmarks, exhibiting a self-attentive understanding of face videos.

In many countries, including those in the European Union, Japan, Australia, and China, labeling is required for foods that contain genetically modified material. Other countries, including the United States, do not require mandatory labeling of GM foods. The United States, however, does allow firms to voluntarily label their products as non-GM. This raises the question of whether a mandatory labeling or voluntary labeling policy is more efficient. Proponents of voluntary labeling policies say they are less expensive, because only the firms that wish to label their products must incur the labeling costs. In a mandatory labeling regime, all firms would need to incur additional costs, whether the costs are due to product testing, label design, segregation, mistakes in labeling, etc. Many proponents of mandatory labeling of GM foods say that consumers have a right to know what they are eating and proponents claim that governments should mandate labels for foods made through genetic modification. This paper presents empirical evidence on consumers' value for foods with positive (i.e. this food is made with genetic engineering) labels versus and for foods with negative (i.e. this food is made without genetic engineering) labels. The positive label is likely to occur under a mandatory labeling system, while the negative label is likely to occur in a voluntary system. Thus, we present evidence on the consumer value of different labeling policies. We estimate values using a laboratory auction experiment performed on 142 randomly chosen adult consumers in the Des Moines, IA, and Saint Paul, MN, areas, grouped in 8 experimental units. They participated in a random nth-price auction experiment, in which they bid on three familiar neutral food items (an oil, a processed food, and a fresh food) that may be genetically modified. Experimental units are randomly assigned to the labeling treatment (i.e., food items with accurate GM food labels versus no food labels). Using statistical design and econometric analysis, this paper will estimate the average value of food label regimes to consumers. This will be measured as the difference in the auction price of a particular food item with and without a GM food label. We present evidence that shows that a voluntary labeling system elicits the same consumer reaction as a mandatory labeling system. Because a voluntary labeling policy would be less expensive to implement, this paper provides evidence that the United States should not abandon the current voluntary labeling system for a mandatory one. This information could be useful to U.S. public policy makers who must decide on GM food labeling legislation.

In this paper, we focus on face clustering in videos. To promote the performance of video clustering by multiple intrinsic cues, i.e., pairwise constraints and multiple views, we propose a constrained multi-view video face clustering method under a unified graph-based model. First, unlike most existing video face clustering methods which only employ these constraints in the clustering step, we strengthen the pairwise constraints through the whole video face clustering framework, both in sparse subspace representation and spectral clustering. In the constrained sparse subspace representation, the sparse representation is forced to explore unknown relationships. In the constrained spectral clustering, the constraints are used to guide for learning more reasonable new representations. Second, our method considers both the video face pairwise constraints as well as the multi-view consistence simultaneously. In particular, the graph regularization enforces the pairwise constraints to be respected and the co-regularization penalizes the disagreement among different graphs of multiple views. Experiments on three real-world video benchmark data sets demonstrate the significant improvements of our method over the state-of-the-art methods.

The rapid growth of tag-free user-generated videos (on the Internet), surgical recorded videos, and surveillance videos has necessitated the need for effective content-based video retrieval systems. Earlier methods for video representations are based on hand-crafted, which hardly performed well on the video retrieval tasks. Subsequently, deep learning methods have successfully demonstrated their effectiveness in both image and video-related tasks, but at the cost of creating massively labeled datasets. Thus, the economic solution is to use freely available unlabeled web videos for representation learning. In this regard, most of the recently developed methods are based on solving a single pretext task using 2D or 3D convolutional network. However, this paper designs and studies a 3D convolutional autoencoder (3D-CAE) for video representation learning (since it does not require labels). Further, this paper proposes a new unsupervised video feature learning method based on joint learning of past and future prediction using 3D-CAE with temporal contrastive learning. The experiments are conducted on UCF-101 and HMDB-51 datasets, where the proposed approach achieves better retrieval performance than state-of-the-art. In the ablation study, the action recognition task is performed by fine-tuning the unsupervised pre-trained model where it outperforms other methods, which further confirms the superiority of our method in learning underlying features. Such an unsupervised representation learning approach could also benefit the medical domain, where it is expensive to create large label datasets.

In this paper, we study the problem of face clustering in videos. Specifically, given automatically extracted faces from videos and two kinds of prior knowledge (the face track that each face belongs to, and the pairs of faces that appear in the same frame), the task is to partition the faces into a given number of disjoint groups, such that each group is associated with one subject. To deal with this problem, we propose a new method called weighted block-sparse low rank representation (WBSLRR) which considers the available prior knowledge while learning a low rank data representation, and also develop a simple but effective approach to obtain the clustering result of faces. Moreover, after using several acceleration techniques, our proposed method is suitable for solving large-scale problems. The experimental results on two benchmark datasets demonstrate the effectiveness of our approach.

The impact of the COVID-19 pandemic has affected the unemployment problem in Indonesia, and it has become one of the public's concerns in the past two years. In August 2020, which was 9.8 million people and August 2021, which was 9.1 million people. Given these conditions, the government needs to make improvements regarding the current unemployment problem. The main objective of this research is to find out how public opinion is regarding the government's efforts to overcome the unemployment problem during the COVID-19 pandemic in Indonesia. Sentiment analysis was carried out on public opinion using Twitter as a data source. To measure the performance of the model, three algorithms are used, namely Naïive Bayes, Decision Tree, and Random Forest. The results of this study indicate that there are positive labels that have 1710 sentiments, and negative labels that have 1553 sentiments. The best algorithm obtained in this study is Random Forest, with an accuracy value of 79%. This study produces 15 features that affect the unemployment problem, the highest positive weight is 'ignore', and the influential feature with the highest negative weight is 'stamp'.

Twitter is one of the most popular and rapidly growing platforms. Through Twitter, users can write and share various activities and opinions, including opinions about 2024 presidential candidates. Several candidates who are suitable to replace the president of Indonesia in 2024 have become the talk of the news media. Anies Baswedan is one of the presidential candidates who has been proposed by the National Democratic Party (NasDem) on October 3, 2022. The opinions of Twitter users can be seen through tweets about Anies Baswedan as a 2024 presidential candidate. These tweets can be analyzed to obtain information on public sentiment towards Anies Baswedan as a 2024 presidential candidate. Therefore, this study aims to apply the Support Vector Machine method in classifying sentiment towards Anies Baswedan as a 2024 presidential candidate. The dataset amounted to 3400 with positive labels as many as 2130 tweets and negative labels as many as 1270 tweets. Labeling is done manually with crowdsourced labelling techniques, obtained a kappa value of 0.68 which shows the level of agreement is relatively strong. Text preprocessing process is carried out. The dataset is divided into training data and test data with a ratio of 90:10. The SVM model with RBF kernel using C=9 and γ=2 parameter pairs has successfully produced good results in validation and evaluation. The accuracy results obtained were 90.61%, precision of 90.67%, recall of 90.61% and f1-score of 90.61%.

By examining the extant literature evaluating small business failure, this paper seeks to show that research in the area has been 'framed', leading to predictable outcomes. Presentation (framing) blinkers the way a problem is perceived and reviewed. Positive labels are more likely to evoke positive associations; negative labels are more likely to evoke negative associations leading to evaluations dependent on how the situation has been labelled. Much prior literature has focused on the negative area of small business failure. An established framework is used to analyse and illustrate that framing has occurred within that literature with respect to assessment of small business failure. Many researchers have accepted that small business is likely to fail with the result that their research is aimed at supporting this contention. Such acceptance impacts on policy decisions and decisions of venture capitalists, bankers and potential entrepreneurs.

Multi-modal clustering (MMC) aims to explore complementary information from diverse modalities for clustering performance facilitating. This article studies challenging problems in MMC methods based on deep neural networks. On one hand, most existing methods lack a unified objective to simultaneously learn the inter- and intra-modality consistency, resulting in a limited representation learning capacity. On the other hand, most existing processes are modeled for a finite sample set and cannot handle out-of-sample data. To handle the above two challenges, we propose a novel Graph Embedding Contrastive Multi-modal Clustering network (GECMC), which treats the representation learning and multi-modal clustering as two sides of one coin rather than two separate problems. In brief, we specifically design a contrastive loss by benefiting from pseudo-labels to explore consistency across modalities. Thus, GECMC shows an effective way to maximize the similarities of intra-cluster representations while minimizing the similarities of inter-cluster representations at both inter- and intra-modality levels. So, the clustering and representation learning interact and jointly evolve in a co-training framework. After that, we build a clustering layer parameterized with cluster centroids, showing that GECMC can learn the clustering labels with given samples and handle out-of-sample data. GECMC yields superior results than 14 competitive methods on four challenging datasets. Codes and datasets are available: https://github.com/xdweixia/GECMC.

Hyperspectral band selection plays an important role in overcoming the curse of dimensionality. Recently, clustering-based band selection methods have shown promise in the selection of informative and representative bands from hyperspectral images (HSIs). However, most existing clustering-based band selection methods involve the clustering of original HSIs, limiting their performance because of the high dimensionality of hyperspectral bands. To tackle this problem, a novel hyperspectral band selection method termed joint learning of correlation-constrained fuzzy clustering and discriminative non-negative representation for hyperspectral band selection (CFNR) is presented. In CFNR, graph regularized non-negative matrix factorization (GNMF) and constrained fuzzy C-means (FCM) are integrated into a unified model to perform clustering on the learned feature representation of bands rather than on the original high-dimensional data. Specifically, the proposed CFNR aims to learn the discriminative non-negative representation of each band for clustering by introducing GNMF into the model of the constrained FCM and making full use of the intrinsic manifold structure of HSIs. Moreover, based on the band correlation property of HSIs, a correlation constraint, which enforces the similarity of clustering results between neighboring bands, is imposed on the membership matrix of FCM in the CFNR model to obtain clustering results that meet the needs of band selection. The alternating direction multiplier method is adopted to solve the joint optimization model. Compared with existing methods, CFNR can obtain a more informative and representative band subset, thus can improve the reliability of hyperspectral image classifications. Experimental results on five real hyperspectral datasets demonstrate that CFNR can achieve superior performance compared with several state-of-the-art methods.

Can We Trust Studies Published by Authors With Financial Conflicts? Using the Decline of Metal-on-Metal Total Hip Arthroplasty to Investigate

This paper describes our contribution in Opinion Target Extraction OTE and Sentiment Polarity sub tasks of SemEval 2015 ABSA task. A CRF model with IOB notation has been adopted for OTE with several groups of features including syntactic, lexical, semantic, sentiment lexicon features. Our submission for OTE is ranked fifth over twenty submissions. A Logistic Regression model with a weighting schema of positive and negative labels have been used for sentiment polarity; several groups of features (lexical, syntactic, semantic, lexicon and Z score) are extracted. Our submission for Sentiment Polarity is ranked third over ten submissions on the restaurant data set, third over thirteen on the laptops data set, but the first over eleven on the hotel data set that is out-of-domain set.

This paper focuses on building up distributed representation of words in cause and effect spaces, a task-specific word embedding technique for causality. The causal embedding model is trained on a large set of cause-effect phrase pairs extracted from raw text corpus via a set of high-precision causal patterns. Three strategies are proposed to transfer the positive or negative labels from the level of phrase pairs to the level of word pairs, leading to three causal embedding models (Pairwise-Matching, Max-Matching, and AttentiveMatching) correspondingly. Experimental results have shown that Max-Matching and Attentive-Matching models significantly outperform several state-of-the-art competitors by a large margin on both English and Chinese corpora.

The problem examined in this study is about the user’s trust in using digital learning applications that are downloaded on playstore. Many reviews are given by the public about the application that has been downloaded on playstore. This review is very influential on their trust in using the application. The purpose of this study is to classify data according to labels and find out the best choice between the classification method and the proposed selection feature as a consideration in determining the use of digital learning applications. This study compares the classification method, the Naïve Bayes algorithm and the genetic algorithm (GA) as feature selection with the Naïve Bayes algorithm classification method and the particle swarm optimization (PSO) as feature selection to categorize the reviews in the playstore. The experimental results show that the Naïve Bayes algorithm and PSO as feature selection is the best model between the two models proposed in this study. Reviews can be classified into positive and negative labels well. The accuracy is 98.00%. The results of the classification are expected to help in making decisions when going to use digital learning application.

Recently, object detection has gained significant progress in remote sensing images. Nevertheless, we conclude two defects in remote sensing image object detection. At first, most methods rely on feature pyramid, but the features of different levels would influence each other when we use top-down operation. Second, the traditional label assignment strategy cannot assign suitable labels, as it adopts the fixed intersection over union (IoU) threshold to divide positive samples and negative samples during training. According to the problems we pointed out, a simple yet effective framework is employed to eliminate these two limitations. It integrates two novel components: aware feature pyramid network (AFPN) and group assignment strategy (GAS). AFPN is to mitigate the adverse effects caused by the first problem. Specifically, it learns a vector for the higher level features in the feature pyramid to obtain clean features. As for the second limitation, we recommend a new label assignment strategy named GAS to address this problem. Samples will be grouped according to their overlaps with ground truth, and then, they are assigned to positive or negative labels in each group. Extensive experiments are conducted on the large-scale object detection dataset DIOR and DOTA. With the newly introduced two key components, our model significantly improves the detection accuracy. Without bells and whistles, our proposed method achieves 2.0% and 1.9% higher mean average precision (mAP) than Faster R-CNN with FPN when using ResNet50 and ResNet101 as the backbones, respectively. Finally, we obtain 73.3% mAP on the DIOR dataset without any tricks. Our code is available at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/hm-better/dior_detect</uri> .