Cosine Similarity Research Articles

Novel CRM cosine similarity mapping strategy for simultaneous in-situ visual profiling lignocellulose in plant cell walls

Confocal Raman microscopy (CRM) is a promising in-situ visual technique that provides detailed insights into multiple lignocellulosic components and structures in plant cell walls at the micro-nano scale. In this study, we propose a novel CRM cosine similarity (CS) mapping strategy for the simultaneous in-situ visual profiling of lignin, cellulose, and hemicellulose in plant cell walls. The main stages of this strategy include: 1) a modified Otsu algorithm for extracting the regions of interest (ROI); 2) a modified subtraction method for cleaning the background signals in the ROI spectra; 3) a lignin signal subtraction method based on the pixel correction factor for eliminating the interference of strong lignin signals with weak cellulose and hemicellulose signals in the Raman full spectra of the cell walls; 4) second-order derivative spectral preprocessing for enhancing the discrimination between the characteristic peaks of cellulose and hemicellulose; 5) a CS mapping algorithm for simultaneous in-situ profiling of lignin, cellulose, and hemicellulose in plant cell walls. The effectiveness of the strategy is verified by characterizing the Brittle Culm1 (BC1) gene-mutant rice stem (IL349-BC1-KO) with known bioinformatics. This approach provides methodological support for in-situ visualization and analysis in fields such as plant or crop science at the micro-nano scale.

Tracking 3D motion of instruments in microsurgery: A comparative study of stereoscopic marker-based vs. deep learning method for objective analysis of surgical skills

Background:In microsurgery, integrating motor and visual information is crucial for achieving precise bimanual maneuvers. Computer vision serves as a useful tool for analyzing surgical maneuvers and enhancing situational awareness in medical assistive systems. Furthermore, the combination of deep learning with computer vision shows promise for objectively assessing surgical skills. Methods:This study compared two methods for tracking 3D movement during microsurgery: Mitracks3D, a stereoscopic tracker using color-marked instruments, and a deep learning method for instruments identification based on YOLOv8. Additionally, drawing an analogy between single-camera and new stereoscopic recording systems, 2D and 3D motion signals from microsurgery training videos on the peg transfer task were obtained to analyze surgeons’ psychomotor skills using six motion analysis parameters (MAPs). Results:When comparing the two 3D tracking methods, the motion signals revealed a relative error (ϵ) of less than 10%, with a correlation coefficient (ρ) exceeding 0.9892 ±0.0020, tracking loss below 2.0398%, and instruments detection time under 1 ms for Mitracks3D and 7.09 ±0.62 ms for YOLOv8. When comparing the performance between participants using 2D and 3D signals, statistically significant differences were found in 4 of the 6 explored MAPs. Additionally, a high cosine similarity (cs>0.99) was found between the performance scores of the 6 MAPs, indicating compatibility between the 2D and 3D recording modalities. Conclusions:The deep learning-based approach is as effective as stereo-tracking with markers; both methods produced remarkably similar 3D motion signals. The compatibility in evaluation between 2D and 3D was confirmed. In essence, adopting computer vision methods such as Mitracks3D and YOLOv8 provides a compelling alternative for constructing objective assessment tools for surgical skills.

Open-set Communication Scene Recognition Based on Metric Learning

Abstract In this paper, we study the problem of recognizing communication scenes under open set protocols, i.e., making the extracted features of communication scenes sufficiently separable under a suitable metric space. To achieve it, we design a feature fusion model based on a convolutional neural network to extract the features of communication scenes, and we take the real and imaginary parts of the channel frequency response as model inputs to extract the scene features sufficiently. Furthermore, in order to optimize the distribution of scene features in angular space, a Linear-Softmax loss function is designed to restrict the inter-class angles of extracted features to be larger than the intra-class angles by two hyperparameters. Meanwhile, a cosine distance-based communication scene recognition algorithm is designed to complete the communication scene recognition by calculating the cosine distance between features. Finally, it is tested on actual measurement data sets, and the experimental results show that compared with the state-of-the-art loss functions, the proposed Linear-Softmax loss function enables the learning of the maximum angular margin. The Linear-Softmax loss function combined with the proposed communication scenes recognition algorithm can complete the recognition of both closed-set and open-set test data, with a recognition accuracy as high as 97.28% on the closed-set test set and 85.65% on the open-set test set.

Adversarial artifact detection in EEG-based brain–computer interfaces

Objective. machine learning has achieved significant success in electroencephalogram (EEG) based brain-computer interfaces (BCIs), with most existing research focusing on improving the decoding accuracy. However, recent studies have shown that EEG-based BCIs are vulnerable to adversarial attacks, where small perturbations added to the input can cause misclassification. Detecting adversarial examples is crucial for both understanding this phenomenon and developing effective defense strategies.Approach. this paper, for the first time, explores adversarial detection in EEG-based BCIs. We extend several popular adversarial detection approaches from computer vision to BCIs. Two new Mahalanobis distance based adversarial detection approaches, and three cosine distance based adversarial detection approaches, are also proposed, which showed promising performance in detecting three kinds of white-box attacks.Main results. we evaluated the performance of eight adversarial detection approaches on three EEG datasets, three neural networks, and four types of adversarial attacks. Our approach achieved an area under the curve score of up to 99.99% in detecting white-box attacks. Additionally, we assessed the transferability of different adversarial detectors to unknown attacks.Significance. through extensive experiments, we found that white-box attacks may be easily detected, and differences exist in the distributions of different types of adversarial examples. Our work should facilitate understanding the vulnerability of existing BCI models and developing more secure BCIs in the future.

Serial Multimodal Biometrics Authentication and Liveness Detection Using Speech Recognition with Normalized Longest Word Subsequence Method

Biometric authentication aims to verify whether an entity matches the claimed identity based on biometric data. Despite its advantages, vulnerabilities, particularly those related to spoofing, still exist. Efforts to mitigate these vulnerabilities include multimodal approaches and liveness detection. However, these strategies may potentially increase resource requirements in the authentication process. This paper proposes a multimodal authentication process incorporating voice and facial recognition, with liveness detection applied to voice data using speech recognition. This paper introduces Normalized Longest Word Subsequence (NLWS), a combination of Intersection Over Union (IOU) and the longest common subsequence, to compare the prompted system sentence with the user's spoken sentence at speech recognition. Unlike the Word Error Rate (WER), NLWS has a measurable range between 1 and 0. Furthermore, the paper introduces decision-level fusion in the multimodal approach, employing two threshold levels in voice authentication. This approach aims to reduce resource requirements while enhancing the overall security of the authentication process. This paper uses cosine similarity, Euclidean distance, random forest, and extreme gradient boosting (XGBoost) to measure distance or similarity. The results show that the proposed method has better accuracy compared to unimodal approaches, achieving accuracies of 98.44%, 98.83%, 97.46%, and 99.22% using cosine similarity, Euclidean distance, random forest, and XGBoost calculations. The proposed method also demonstrates resource savings, reducing from 5.19 MB to 0.792 MB, from 7.3294 MB to 1.9437 MB, from 6.6512 MB to 1.3284 MB, and from 7.8632 MB to 2.1517 MB in different distance or similarity measurements

Industrial policy persistence and local economic performance: The role of subsidy allocation in China

The persistence of economic policies is crucial for their effectiveness and economic outcomes. However, debates on industrial policy effectiveness have primarily focused on policy tools and target sectors, with little attention given to policy persistence. This paper addresses this gap by evaluating how policy persistence during subsidy implementation affects policy effectiveness. We measure policy persistence using cosine similarity between vectors reflecting the cross-industry distribution of subsidies over consecutive years and instrument it with the occurrence of abrupt public safety incidents. Our results reveal a significantly positive effect of subsidy allocation persistence on local economic performance. This effect is stronger when local governments are influential or industries need consistent government support and weaker when policy changes are more predictable or local officials are less prone to decision-making mistakes. We further provide micro-level evidence showing that the persistence of industrial policies impacts local economic performance by influencing the activities of local businesses.

Comparison of Latent Semantic Analysis (LSA) and Doc2Vec Algorithms of Thesis Similarity Detection

This study aims to develop a system for detecting similarities in thesis titles and content to prevent plagiarism and support student originality. The high level of similarity in final projects is a significant concern in academic environments. Two text vectorization methods, Latent Semantic Analysis (LSA) and Doc2Vec, were compared to measure document similarity. Results showed that LSA achieved a very high cosine similarity (99.94%) due to dimensionality reduction that preserved semantic correlations. In contrast, Doc2Vec produced lower similarity scores, with 7.17% for PV-DM and 39.07% for PV-DBOW, indicating richer text representations. This study adopted the CRISP-DM model, which includes Business Understanding, Data Understanding, Data Preparation, Modelling, and Evaluation. The model is expected to strengthen academic integrity and encourage valuable scientific contributions.

Foreign Accent and Forensic Speaker Identification in Voice Lineups: The Influence of Acoustic Features Based on Prosody.

This research aims to examine both the prosodic-acoustic features and the perceptual correlates of foreign-accented English and foreign-accented Brazilian Portuguese and check how the speakers' productions of foreign and native accents are correlated to the listeners' perception. In the Methodology, we conducted a speech production procedure with a group of American speakers of L2 Brazilian Portuguese and a group of Brazilian speakers of L2 English, and a speech perception procedure in which we performed voice lineups for both languages.For the speech production statistical analysis, we ran Generalized Additive Models to evaluate the effect of the language groups on each class (metric or prosodic-acoustic) of features controlled for the smoothing effect of the covariate(s) of the opposite class. For the speech perception statistical analysis, we ran a Kruskal-Wallis test and a post-hoc Dunn's test to evaluate the effect of the voices of the lineups on the scores judged by the listeners. We nevertheless conducted acoustic (voice) similarity tests based on Cosine and Euclidean distances. Results showed significant acoustic differences between the language groups in terms of variability of the f0, duration, and voice quality. For the lineups, the results indicated that prosodic features of f0, intensity, and voice quality correlated to the listeners' perceived judgments.

A multi-entity reinforced main path analysis: Heterogeneous network embedding considering knowledge proximity

Main path analysis (MPA) is an important approach in detecting the trajectory of knowledge diffusion in a specific research domain. Previous studies always focus on citation-based relationships, overlooking other structural forms in citation network. This study introduces a multi-entity reinforced MPA model by constructing a knowledge graph from paper metadata, including citations, authors, journals, and keywords. We construct heterogeneous network to reveal relationships among various entities. Different knowledge graph embedding models are employed to train the network, thereby obtaining entity and relation embeddings. The cosine similarity algorithm is adopted to measure the knowledge proximity between these embeddings. We take the Internet of Thing domain as an example to verify the performance of the multi-entity reinforced MPA through both quantitative and qualitative analysis. Our findings indicate that the adjusted MPA exhibits stronger topic relevance, demonstrating the effectiveness of the method in capturing complex knowledge relationships.

Optimizing hierarchical classifiers with parameter tuning and confidence scoring

Hierarchical classifiers play a crucial role in addressing complex classification tasks by breaking them down into smaller, more manageable sub-tasks. This paper continues a series of works, focused on the technical Ukrainian texts hierarchical classification, specifically the classification of repair works and spare parts used in automobile maintenance and servicing. We tackle the challenges posed by multilingual data inputs – specifically Ukrainian, Russian, and their hybrid – and the lack of standard data cleaning models for the Ukrainian language. We developed a novel classification algorithm, which employs TF-IDF victimization with unigrams and bigrams, keyword selection, and cosine similarity for classification. This paper describes a method for training and evaluating a hierarchical classification model using parameter tuning for each node in a tree structure. The training process involves initializing weights for tokens in the class tree nodes and input strings, followed by iterative parameter tuning to optimize classification accuracy. Initial weights are assigned based on predefined rules, and the iterative process adjusts these weights to achieve optimal performance. The paper also addresses the challenge of interpreting multiple confidence scores from the classification process, proposing a machine learning approach using Scikit-learn's GradientBoostingClassifier to calculate a unified confidence score. This score helps assess the classification reliability, particularly for unlabeled data, by transforming input values, generating polynomial parameters, and using logarithmic transformations and scaling. The classifier is fine-tuned using hyper parameter optimization techniques, and the final model provides a robust confidence score for classification tasks, enabling the verification and classification results optimization across large datasets. Our experimental results demonstrate significant improvements in classification performance. Overall classification accuracy nearly doubled after training, reaching 92.38 %. This research not only advances the theoretical framework of hierarchical classifiers but also provides practical solutions for processing large-scale, unlabeled datasets in the automotive industry. The developed methodology can enhance various applications, including automated customer support systems, predictive maintenance, and decision-making processes for stakeholders like insurance companies and service centers. Future work will extend this approach to more complex tasks, such as extracting and classifying information from extensive text sources like telephone call transcriptions.

A Two-Stage Fault Localization Method for Active Distribution Networks Based on COA-SVM Model and Cosine Similarity

A Two-Stage Fault Localization Method for Active Distribution Networks Based on COA-SVM Model and Cosine Similarity

Comparison of text information from information sources based on the cosine similarity algorithm

This article presents research aimed at developing an optimal concept of analysis and comparison of information sources based on large volumes of textual information using natural language processing (NLP) methods. The object of the study was Telegram news channels, which are used as sources of text data. Texts were pre-processed, including cleaning, tokenization, and lemmatization, to form a global dictionary consisting of unique words from all information sources. For each source, a vector representation of the texts was built, the dimension of which corresponds to the number of unique words in the global dictionary. The frequency of use of each word in the channel's texts was displayed in the corresponding positions of the vector. By applying the cosine similarity algorithm to pairs of vectors, a square matrix was obtained that demonstrates the degree of similarity between different sources. The results of the study show the effectiveness of the proposed approach for quantitative assessment of the similarity of textual data from different sources. The need for further optimization of the algorithm was identified, in particular by parameterization to achieve a balance between accuracy and computational cost, as well as the separation of words with excessive weight, such as specific terms or channel names. The proposed method can be applied to the analysis of information flows, the identification of relationships between sources and the study of the socio-cultural influence of media content in the conditions of the modern information environment.

A graph representation framework for encrypted network traffic classification

Network Traffic Classification (NTC) is crucial for ensuring internet security, but encryption presents significant challenges to this task. While Machine Learning (ML) and Deep Learning (DL) methods have shown promise, issues such as limited representativeness leading to sub-optimal generalizations and performance remain prevalent. These problems become more pronounced with advanced obfuscation, network security, and privacy technologies, indicating a need for improved model robustness. To address these issues, we focus on feature extraction and representation in NTC by leveraging the expressive power of graphs to represent network traffic at various granularity levels. By modeling network traffic as interconnected graphs, we can analyze both flow-level and packet-level data. Our graph representation method for encrypted NTC effectively preserves crucial information despite encryption and obfuscation. We enhance the robustness of our approach by using cosine similarity to exploit correlations between encrypted network flows and packets, defining relationships between abstract entities. This graph structure enables the creation of structural embeddings that accurately define network traffic across different encryption levels. Our end-to-end process demonstrates significant improvements where traditional NTC methods struggle, such as in Tor classification, which employs anonymization to further obfuscate traffic. Our packet-level classification approach consistently outperforms existing methods, achieving accuracies exceeding 96%.

Variable frequency selective structure (CFRP/Cu/CFRP) micro blind hole laser high-quality drilling method guided by acoustic emission technology

The laminated structure of carbon fiber reinforced plastic (CFRP)-Cu-CFRP is an important structure for aerospace radome. The blind holes manufacturing is core process, but there are technological challenges in achieving the exposure of a 10 μm thick copper foil at the bottom of the blind hole. The exact position of the 10 μm thick copper foil within the material’s internal space is unknown, and the uneven heat accumulation and conduction can easily cause burning and damage to the ultra-thin copper foil. To date, there is no publicly accessible literature that documents the execution of this specific process. In this paper, novel variable inner diameter spiral scanning guided by Cu-CFRP interface accurately extracted using cosine similarity (CS) algorithm is proposed to obtain a blind hole with near-zero ablation damage. The similarity of acoustic emission signals generated by laser interaction with Cu and CFRP, as well as thermal accumulation and conduction behavior under different scanning radius, are described. The high-quality blind holes have been successfully drilled, with a large area copper foil exposed at the bottom, while the surface of the copper foil remaining intact without any damage or heat affected zone. Compared with traditional drilling method, the average exposed copper foil area at the bottom of the blind holes is increased by 238.4 % (from 0.464 mm2 to 1.570 mm2). And the average taper of blind hole is reduced by 41.7 % (from 0.470 to 0.274). This method can effectively increase the area of exposed copper foil and reduces the taper of blind holes, improve the welding strength and conductivity of wires in aerospace radomes, thereby enhancing the stealth performance and operational reliability of fighter jets.

Advancing Kidney Transplantation: A Machine Learning Approach to Enhance Donor-Recipient Matching.

(1) Background: Globally, the kidney donor shortage has made the allocation process critical for patients awaiting a kidney transplant. Adopting Machine Learning (ML) models for donor-recipient matching can potentially improve kidney allocation processes when compared with traditional points-based systems. (2) Methods: This study developed an ML-based approach for donor-recipient matching. A comprehensive evaluation was conducted using ten widely used classifiers (logistic regression, decision tree, random forest, support vector machine, gradient boosting, boost, CatBoost, LightGBM, naive Bayes, and neural networks) across three experimental scenarios to ensure a robust approach. The first scenario used the original dataset, the second used a merged version of the dataset, and the last scenario used a hierarchical architecture model. Additionally, a custom ranking algorithm was designed to identify the most suitable recipients. Finally, the ML-based donor-recipient matching model was integrated into a web-based platform called Nephron. (3) Results: The gradient boost model was the top performer, achieving a remarkable and consistent accuracy rate of 98% across the three experimental scenarios. Furthermore, the custom ranking algorithm outperformed the conventional cosine and Jaccard similarity methods in identifying the most suitable recipients. Importantly, the platform not only facilitated efficient patient selection and prioritisation for kidney allocation but can be flexibly adapted for other solid organ allocation systems built on similar criteria. (4) Conclusions: This study proposes an ML-based approach to optimize donor-recipient matching within the kidney allocation process. Successful implementation of this methodology demonstrates significant potential to enhance both efficiency and fairness in kidney transplantation.

Fractal Self-Similarity in Semantic Convergence: Gradient of Embedding Similarity across Transformer Layers

This paper presents a mathematical analysis of semantic convergence in transformer-based language models, drawing inspiration from the concept of fractal self-similarity. We introduce and prove a novel theorem characterizing the gradient of embedding similarity across layers. Specifically, we establish that there exists a monotonically increasing function that provides a lower bound on the rate at which the average cosine similarity between token embeddings at consecutive layers and the final layer increases. This establishes a fundamental property: semantic alignment of token representations consistently increases through the network, exhibiting a pattern of progressive refinement, analogous to fractal self-similarity. The key challenge addressed is the quantification and generalization of semantic convergence across diverse model architectures and input contexts. To validate our findings, we conduct experiments on BERT and DistilBERT models, analyzing embedding similarities for diverse input types. While our experiments are limited to these models, we empirically demonstrate consistent semantic convergence within these architectures. Quantitatively, we find that the average rates of semantic convergence are approximately 0.0826 for BERT and 0.1855 for DistilBERT. We observe that the rate of convergence varies based on token frequency and model depth, with rare words showing slightly higher similarities (differences of approximately 0.0167 for BERT and 0.0120 for DistilBERT). This work advances our understanding of transformer models’ internal mechanisms and provides a mathematical framework for comparing and optimizing model architectures.

Geometry of Textual Data Augmentation: Insights from Large Language Models

Data augmentation is crucial for enhancing the performance of text classification models when labelled training data are scarce. For natural language processing (NLP) tasks, large language models (LLMs) are able to generate high-quality augmented data. But a fundamental understanding of the reasons for their effectiveness remains limited. This paper presents a geometric and topological perspective on textual data augmentation using LLMs. We compare the augmentation data generated by GPT-J with those generated through cosine similarity from Word2Vec and GloVe embeddings. Topological data analysis reveals that GPT-J generated data maintains label coherence. Convex hull analysis of such data represented by their two principal components shows that they lie within the spatial boundaries of the original training data. Delaunay triangulation reveals that increasing the number of augmented data points that are connected within these boundaries correlates with improved classification accuracy. These findings provide insights into the superior performance of LLMs in data augmentation. A framework for predicting the usefulness of augmentation data based on geometric properties could be formed based on these techniques.

Creating the Slider Tester Repair Recommendation System to Enhance the Repair Step by Using Machine Learning

This project aims to develop a recommendation system to mitigate looping issues in HDD slider testing using the Amber testing machine (Machine A). Components simulating the HDD often fail and require repair before re-testing. However, post-repair, there is a 34% probability that the component (referred to as Product A) will experience looping, characterized by repeated failures with error code A. This recurring issue significantly hampers testing efficiency by reducing the number of successful slider tests. To address this challenge, we propose a dual-approach recommendation system that provides technicians with actionable insights to minimize the occurrence of looping. For previously analyzed components, a collaborative filtering technique utilizing implicit ratings is employed to generate recommendations. For new components, for which prior data are unavailable, a cosine similarity approach is applied to suggest optimal actions. An automatic training system is implemented to retrain the model as new data become available, ensuring that the recommendation system remains robust and effective over time. The proposed system is expected to offer precise guidance to technicians, thereby improving the overall efficiency of the testing process by reducing the frequency of looping issues. This work represents a significant advancement in enhancing operational reliability and productivity in HDD slider testing.

LDAPrototype: a model selection algorithm to improve reliability of latent Dirichlet allocation

Latent Dirichlet allocation (LDA) is a popular method for analyzing large text corpora, but it suffers from instability due to its reliance on random initialization. This results in different outcomes for replicated runs, hindering reproducibility. To address this, we introduce LDAPrototype, a new approach for selecting the most representative LDA run from multiple replications on the same dataset. LDAPrototype enhances the reliability of LDA conclusions by ensuring greater similarity between replications compared to traditional LDA runs or models chosen based on perplexity or NPMI. A key feature of LDAPrototype is its use of a novel model similarity measure called S-CLOP (Similarity of multiple sets by Clustering with LOcal Pruning). It is based on topic similarities, for which we compare the usage of measures like the thresholded Jaccard coefficient, cosine similarity, Jensen-Shannon divergence, and rank-biased overlap. The effectiveness of LDAPrototype is demonstrated through its application to six real datasets, including newspaper articles and tweets. The results show improved reproducibility and reliability in topic modeling outcomes. LDAPrototype’s approach is noteworthy for its practical applicability, comprehensibility, ease of implementation, and computational efficiency. Furthermore, the algorithm’s concept can be generalized to other topic modeling procedures that characterize topics through word distributions, making it a versatile tool in text data analysis.

Snapshot computed tomographic microscopic imaging spectrometer and its video-level tracking of poisonous Microcystis aeruginosa cells in mixed algae

Conventional microscopic spectral imaging suffers from extended scanning times across wavelength or spatial dimension. To improve capabilities of dynamic microscopic spectral imaging, we developed a snapshot computed tomographic microscopic imaging spectrometer (CTMIS) based on the zeroth and first orders dispersive diffraction of a two-dimensional grating. Utilizing the CTMIS-UNET reconstruction algorithm, we can reconstruct a spectral cube (541x541x26) for each frame of micro spectral imaging video. Experimental results demonstrate a sub-4 μm spatial resolution achievable through a 20x objective lens and a spectral resolution better than 10 nm among 450–700 nm, while maintaining spectral cosine similarities exceeding 0.9989 when comparing reconstructed spectra with ground truth data. Spectral imaging videos of four species of algae and mixed algae were captured under 10 ms exposure time using the CTMIS system. Leveraging the self-developed UNET-SI26 algae recognition network, precise identification and tracking of four types of algae and poisonous microcysts aeruginosa in mixed algae were conducted. The pixel-level recognition accuracy exceeds 95 %, while the accuracy for counting the numbers of different types of cells surpasses 85 %, offering an efficient and accurate spectral imaging method for real-time monitoring and early warning of harmful algae at the cellular level.