Positive-negative Pairs Research Articles

Exploring Contrastive Pre-Training for Domain Connections in Medical Image Segmentation.

Unsupervised domain adaptation (UDA) in medical image segmentation aims to improve the generalization of deep models by alleviating domain gaps caused by inconsistency across equipment, imaging protocols, and patient conditions. However, existing UDA works remain insufficiently explored and present great limitations: 1) Exhibit cumbersome designs that prioritize aligning statistical metrics and distributions, which limits the model's flexibility and generalization while also overlooking the potential knowledge embedded in unlabeled data; 2) More applicable in a certain domain, lack the generalization capability to handle diverse shifts encountered in clinical scenarios. To overcome these limitations, we introduce MedCon, a unified framework that leverages general unsupervised contrastive pre-training to establish domain connections, effectively handling diverse domain shifts without tailored adjustments. Specifically, it initially explores a general contrastive pre-training to establish domain connections by leveraging the rich prior knowledge from unlabeled images. Thereafter, the pre-trained backbone is fine-tuned using source-based images to ultimately identify per-pixel semantic categories. To capture both intra- and inter-domain connections of anatomical structures, we construct positive-negative pairs from a hybrid aspect of both local and global scales. In this regard, a shared-weight encoder-decoder is employed to generate pixel-level representations, which are then mapped into hyper-spherical space using a non-learnable projection head to facilitate positive pair matching. Comprehensive experiments on diverse medical image datasets confirm that MedCon outperforms previous methods by effectively managing a wide range of domain shifts and showcasing superior generalization capabilities.

Randomization based evaluation of distinct topological and cancer expression characteristics of mutually acting gene pairs.

Small scale molecular network patterns and motifs are crucial for systems level understanding of cellular information transduction. Using randomizations, we statistically explored, previously overlooked basic patterns of mutually acting pairs, i.e. mutually positive (PP) or negative (NN) and positive-negative (PN) pairs, in two comprehensive and distinct large-scale molecular networks from literature; the human protein signaling network (PSN) and the human gene regulatory network (GRN). Only the positive and negative signs of all interacting pairs were randomized, while the gene pairs and the number of positive and negative signs in the original network were kept constant. While the numbers of NN and PN pairs were significantly higher, the number of PP pairs was significantly lower than randomly expected values. Genes participating in mutual pairs were more connected than other genes. NN genes were more connected than PP and PN in GRN for all types of degree values, including in, out, positive or negative connections, but less connected for in-degree and more connected for out-degree values in PSN. They also had significantly high number of intersections with each other and PN pairs than randomly expected values, indicating potential cooperative mechanisms. The three mutual interaction designs we examined had distinct RNA and protein expression correlation characteristics. NN protein pairs were uniquely over-represented across normal tissue samples, whose negative correlations were lost across cancer tissue samples. PP and PN pairs showed non-random positive RNA or protein expression correlation across normal or cancer tissue samples. Moreover, we developed an online tool, i.e. MGPNet, for further user specific analysis of mutual gene pairs. We identified SNCA with significantly enriched negatively correlated NN pairs. Unique non-random characteristics of mutual gene pairs identified in two different comprehensive molecular networks could provide valuable information for a better comparative understanding of molecular design principles between normal and cancer states. Insight Box/Paragraph Statement: This study provides a systems-level perspective on cellular information transduction by analyzing mutually acting pairs of genes. By examining mutually positive (PP), mutually negative (NN), and positive-negative (PN) pairs in the human protein signaling network (PSN) and the human gene regulatory network (GRN), we uncover significant variations in their connectivity and expression correlation. Our findings highlight the unique features of NN pairs across normal and cancer tissues and offer insights into molecular design principles. The development of the MGPNet tool further enhances user-specific analyses, enabling a deeper understanding of gene pair mechanisms and their potential cooperative roles in cellular processes.

FRCL-MNER: A Finer Grained Rank-Based Contrastive Learning Framework for Multimodal NER.

Multimodal named entity recognition (MNER) is an emerging field that aims to automatically detect named entities and classify their categories, utilizing input text and auxiliary resources such as images. While previous studies have leveraged object detectors to preprocess images and fuse textual semantics with corresponding image features, these methods often overlook the potential finer grained information within each modality and may exacerbate error propagation due to predetection. To address these issues, we propose a finer grained rank-based contrastive learning (FRCL) framework for MNER. This framework employs a global-level contrastive learning to align multimodal semantic features and a Top-K rank-based mask strategy to construct positive-negative pairs, thereby learning a finer grained multimodal interaction representation. Experimental results from three well-known social media datasets reveal that our approach surpasses existing strong baselines, and achieves up to a 1.54% improvement on the Twitter2015 dataset. Extensive discussions further confirm the effectiveness of our approach. We will release the source code on https://github.com/augusyan/FRCL.

Detection of pulmonary nodules in chest radiographs: novel cost function for effective network training with purely synthesized datasets

PurposeMany large radiographic datasets of lung nodules are available, but the small and hard-to-detect nodules are rarely validated by computed tomography. Such difficult nodules are crucial for training nodule detection methods. This lack of difficult nodules for training can be addressed by artificial nodule synthesis algorithms, which can create artificially embedded nodules. This study aimed to develop and evaluate a novel cost function for training networks to detect such lesions. Embedding artificial lesions in healthy medical images is effective when positive cases are insufficient for network training. Although this approach provides both positive (lesion-embedded) images and the corresponding negative (lesion-free) images, no known methods effectively use these pairs for training. This paper presents a novel cost function for segmentation-based detection networks when positive–negative pairs are available.MethodsBased on the classic U-Net, new terms were added to the original Dice loss for reducing false positives and the contrastive learning of diseased regions in the image pairs. The experimental network was trained and evaluated, respectively, on 131,072 fully synthesized pairs of images simulating lung cancer and real chest X-ray images from the Japanese Society of Radiological Technology dataset.ResultsThe proposed method outperformed RetinaNet and a single-shot multibox detector. The sensitivities were 0.688 and 0.507 when the number of false positives per image was 0.2, respectively, with and without fine-tuning under the leave-one-case-out setting.ConclusionTo our knowledge, this is the first study in which a method for detecting pulmonary nodules in chest X-ray images was evaluated on a real clinical dataset after being trained on fully synthesized images. The synthesized dataset is available at https://zenodo.org/records/10648433.

SALENet: Structure-Aware Lighting Estimations from a Single Image for Indoor Environments.

High Dynamic Range (HDR) lighting plays a pivotal role in modern augmented and mixed-reality (AR/MR) applications, facilitating immersive experiences through realistic object insertion and dynamic relighting. However, the acquisition of precise HDR environment maps remains cost-prohibitive and impractical when using standard devices. To bridge this gap, this paper introduces SALENet, a novel deep network for estimating global lighting conditions from a single image, to effectively mitigate the need for resource-intensive acquisition methods. In contrast to earlier studies, we focus on exploring the inherent structural relationships within the lighting distribution. We design a hierarchical transformer-based neural network architecture with a proposed cross-attention mechanism between different resolution lighting source representations, optimizing the spatial distribution of lighting sources simultaneously for enhanced consistency. To further improve accuracy, a structure-based contrastive learning method is proposed to select positive-negative pairs based on lighting distribution similarity. By harnessing the synergy of hierarchical transformers and structure-based contrastive learning, our framework yields a significant enhancement in lighting prediction accuracy, enabling high-fidelity augmented and mixed reality to achieve cost-effectively immersive and realistic lighting effects.

Efficient Self-Supervision using Patch-based Contrastive Learning for Histopathology Image Segmentation

Learning discriminative representations of unlabelled data is a challenging task. Contrastive self-supervised learning provides a framework to learn meaningful representations using learned notions of similarity measures from simple pretext tasks. In this work, we propose a simple and efficient framework for self-supervised image segmentation using contrastive learning on image patches, without using explicit pretext tasks or any further labeled fine-tuning. A fully convolutional neural network (FCNN) is trained in a self-supervised manner to discern features in the input images and obtain confidence maps which capture the network's belief about the objects belonging to the same class. Positive- and negative- patches are sampled based on the average entropy in the confidence maps for contrastive learning. Convergence is assumed when the information separation between the positive patches is small, and the positive-negative pairs is large. The proposed model only consists of a simple FCNN with 10.8k parameters and requires about 5 minutes to converge on the high resolution microscopy datasets, which is orders of magnitude smaller than the relevant self-supervised methods to attain similar performance. We evaluate the proposed method for the task of segmenting nuclei from two histopathology datasets, and show comparable performance with relevant self-supervised and supervised methods.

RLAS-BIABC: A Reinforcement Learning-Based Answer Selection Using the BERT Model Boosted by an Improved ABC Algorithm.

Answer selection (AS) is a critical subtask of the open-domain question answering (QA) problem. The present paper proposes a method called RLAS-BIABC for AS, which is established on attention mechanism-based long short-term memory (LSTM) and the bidirectional encoder representations from transformers (BERT) word embedding, enriched by an improved artificial bee colony (ABC) algorithm for pretraining and a reinforcement learning-based algorithm for training backpropagation (BP) algorithm. BERT can be comprised in downstream work and fine-tuned as a united task-specific architecture, and the pretrained BERT model can grab different linguistic effects. Existing algorithms typically train the AS model with positive-negative pairs for a two-class classifier. A positive pair contains a question and a genuine answer, while a negative one includes a question and a fake answer. The output should be one for positive and zero for negative pairs. Typically, negative pairs are more than positive, leading to an imbalanced classification that drastically reduces system performance. To deal with it, we define classification as a sequential decision-making process in which the agent takes a sample at each step and classifies it. For each classification operation, the agent receives a reward, in which the prize of the majority class is less than the reward of the minority class. Ultimately, the agent finds the optimal value for the policy weights. We initialize the policy weights with the improved ABC algorithm. The initial value technique can prevent problems such as getting stuck in the local optimum. Although ABC serves well in most tasks, there is still a weakness in the ABC algorithm that disregards the fitness of related pairs of individuals in discovering a neighboring food source position. Therefore, this paper also proposes a mutual learning technique that modifies the produced candidate food source with the higher fitness between two individuals selected by a mutual learning factor. We tested our model on three datasets, LegalQA, TrecQA, and WikiQA, and the results show that RLAS-BIABC can be recognized as a state-of-the-art method.

Characteristics of sources and sinks of momentum in a turbulent boundary layer

In turbulent boundary layers, the wall-normal gradient of the Reynolds shear stress identifies momentum sources and sinks (T=∂[−uv]/∂y). These motions can be physically interpreted in two ways {i.)} as contributors to the turbulence term balancing the mean momentum equation, and {ii.)} as regions of strong local interaction between velocity and vorticity fluctuations. In this paper, the space-time evolution of momentum sources and sinks are investigated in a turbulent boundary layer at the Reynolds number (Reτ) = 2700, with time-resolved planar particle image velocimetry in a plane along the streamwise and wall-normal directions. Wavenumber--frequency power spectra of T fluctuations reveal that the wave velocities of momentum sources and sinks tend to match the local streamwise velocity in proximity to the wall. However, as the distance from the wall increases, the wave velocities of the T events are slightly lower than the local streamwise velocities of the flow, which is also confirmed from the tracking in time of the intense momentum sources and sinks. This evidences that momentum sources and sinks are preferentially located in low-momentum regions of the flow. The spectral content of the T fluctuations is maximum at the wall, but it decreases monotonically as the distance from the wall grows. The relative spectral contributions of the different wavelengths remains unaltered at varying wall-normal locations. From autocorrelation coefficient maps, the characteristic streamwise and wall-normal extents of the T motions are respectively 60 and 40 wall-units, independent of the wall distance. Both statistics and instantaneous visualizations show that momentum sources and sinks have a preferential tendency to be organized in positive-negative pairs in the wall-normal direction.

A Secret Confusion Based Energy‐Saving and Privacy‐Preserving Data Aggregation Algorithm

To reduce communication overhead on the premise of privacy protection, this study presents a novel secret Confusion based energy-saving and privacypreserving data aggregation algorithm (CESPT). In confusion phase, CESPT confuses real sensory data and their sources by positive-negative pairs and a confusion factor is introduced to determine the quantity of pairs generated by a sensor, the exchange rounds and the threshold of data exchange, which affect communication overhead and privacy intensity of a Wireless sensor network (WSN). In aggregation phase, CESPT adopts a positive-negative neutralization strategy and a well-designed time slice allocation mechanism to reduce network traffic and message collision. In a word, CESPT can greatly reduce data traffic and energy consumption and obtain accurate statistical results on the basis of data privacy.

Appraising Computing Self-Efficacy Emotions across a 5-Week Multimedia Authoring Project

<p>Emotion is a key aspect of how non-specialists learn computing. The emotions included in Computing Self-Efficacy (CSE) research were identified prior to the emergence of recent models of emotion. There has been no attempt to inventory attitudes elicited while learning computing, using contemporary psycholinguistic models of subjectivity. This study of 58 medical students in Saudi Arabia used Appraisal analysis of weekly written personal responses to gain a comprehensive overview of emotions elicited during five weeks’ instruction on website-building. A Before-After Survey identified gains made in reported frequency of tasks performed outside class. A Weekly Attitude Survey identified the strength of 6 previously-identified CSE emotions, framed as positive-negative pairs. Participant journals showed that many emotions included in previous CSE emotions are not frequently-realised, and attitudes are changeable across the learning process. Overall, most positive-negative pairs do not behave correlatively, some persist where others progress, and incidence is a better guide than polarity to an attitude’s significance. Capacity and confidence suggest three stages in learning a computing task.</p>

Dissection and prediction of RNA-binding sites on proteins.

RNA-binding proteins are involved in many important regulatory processes in cells and their study is essential for a complete understanding of living organisms. They show a large variability from both structural and functional points of view. However, several recent studies performed on protein-RNA crystal structures have revealed interesting common properties. RNA-binding sites usually constitute patches of positively charged or polar residues that make most of the specific and non-specific contacts with RNA. Negatively charged or aliphatic residues are less frequent at protein-RNA interfaces, although they can also be found either forming aliphatic and positive-negative pairs in protein RNA-binding sites or contacting RNA through their main chains. Aromatic residues found within these interfaces are usually involved in specific base recognition at RNA single-strand regions. This specific recognition, in combination with structural complementarity, represents the key source for specificity in protein-RNA association. From all this knowledge, a variety of computational methods for prediction of RNA-binding sites have been developed based either on protein sequence or on protein structure. Some reported methods are really successful in the identification of RNA-binding proteins or the prediction of RNA-binding sites. Given the growing interest in the field, all these studies and prediction methods will undoubtedly contribute to the identification and comprehension of protein-RNA interactions.

Anomalous Effects of “Guest” Charges Immersed in Electrolyte: Exact 2D Results

We study physical situations when one or two "guest" arbitrarily-charged particles are immersed in the bulk of a classical electrolyte modelled by a Coulomb gas of positive/negative unit point-like charges, the whole system being in thermal equilibrium. The models are treated as two-dimensional with logarithmic pairwise interactions among charged constituents; the (dimensionless) inverse temperature $\beta$ is considered to be smaller than 2 in order to ensure the stability of the electrolyte against the collapse of positive-negative pairs of charges. Based on recent progress in the integrable (1+1)-dimensional sine-Gordon theory, exact formulas are derived for the chemical potential of one guest charge and for the asymptotic large-distance behavior of the effective interaction between two guest charges. The exact results imply, under certain circumstances, anomalous effects such as an effective attraction (repulsion) between like-charged (oppositely-charged) guest particles and the charge inversion in the electrolyte vicinity of a highly-charged guest particle. The adequacy of the concept of renormalized charge is confirmed in the whole stability region of inverse temperatures and the related saturation phenomenon is revised.

Exact Solution of a Charge-Asymmetric Two-Dimensional Coulomb Gas

The model under consideration is an asymmetric two-dimensional Coulomb gas of positively (q1=+1) and negatively (q2=−1/2) charged pointlike particles, interacting via a logarithmic potential. This continuous system is stable against collapse of positive-negative pairs of charges for the dimensionless coupling constant (inverse temperature) β<4. The mapping of the Coulomb gas is made onto the complex Bullough–Dodd model, and recent results about that integrable 2D field theory are used. The mapping provides the full thermodynamics (the free energy, the internal energy, the specific heat) and the large-distance asymptotics of the particle correlation functions, in the whole stability regime of the plasma. The results are checked by a small-β expansion and close to the collapse β=4 point. The comparison is made with the exactly solvable symmetric version of the model (q1=+1,q2=−1), and some fundamental changes in statistics caused by the charge asymmetry are pointed out.

Thermodynamic Properties of the Two-Dimensional Coulomb Gas in the Low-Density Limit

The model under consideration is the two-dimensional Coulomb gas of ± charged hard disks with diameter σ. For the case of pointlike charges (σ=0), the system is stable against collapse of positive-negative pairs of charges in the range of inverse temperatures 0≤β<2, where its full exact thermodynamics was obtained recently. In the present work, we derive the leading correction to the exact thermodynamics of pointlike charges due to presence of the hard core σ which enables us to extend the treatment beyond the collapse point β=2. Our results, which are conjectured to be exact in the low-density limit in the interval 0≤β<3, reproduce correctly the singularities of thermodynamic quantities at the collapse point and agree well with Monte-Carlo simulations. The “subtraction” mechanism within the ansatz proposed by M. E. Fisher et al. [J. Stat. Phys.79:1 (1995)], which excludes the existence of intermediate phases between the collapse point β=2 and the Kosterlitz–Thouless transition point βKT=4, is confirmed, however, a different analytic structure of this ansatz is suggested.

Large-distance behavior of particle correlations in the two-dimensional two-component plasma

The model under consideration is a two-dimensional two-component plasma, i.e., a continuous system of two species of pointlike particles of opposite charges ±1, interacting through the logarithmic Coulomb interaction. Using the exact results for the form-factors of an equivalent Euclidean sine-Gordon theory, we derive the large-distance behavior of the pair correlation functions between charged particles. This asymptotic behavior is checked on a few lower orders of its β-expansion (β is the inverse temperature) around the Debye–Huckel limit β→0, and at the free-fermion point β=2 at which the collapse of positive-negative pairs of charges occurs.