Data Embedding Research Articles

The role of data embedding in equivariant quantum convolutional neural networks

AbstractGeometric deep learning refers to the scenario in which the symmetries of a dataset are used to constrain the parameter space of a neural network and thus, improve their trainability and generalization. Recently, this idea has been incorporated into the field of quantum machine learning, which has given rise to equivariant quantum neural networks (EQNNs). In this work, we investigate the role of classical-to-quantum embedding on the performance of equivariant quantum convolutional neural networks (EQCNNs) for the classification of images. We discuss the connection between the data embedding method and the resulting representation of a symmetry group and analyze how changing representation affects the expressibility of an EQCNN. We numerically compare the classification accuracy of EQCNNs with three different basis-permuted amplitude embeddings to the one obtained from a non-equivariant quantum convolutional neural network (QCNN). Our results show a clear dependence of classification accuracy on the underlying embedding, especially for initial training iterations. The improvement in classification accuracy of EQCNN over non-equivariant QCNN may be present or absent depending on the particular embedding and dataset used. The noisy simulation using simple noise models shows that certain EQCNNs are more robust to noise than non-equivariant QCNNs. It is expected that the results of this work can be useful to the community for a better understanding of the importance of data embedding choice in the context of geometric quantum machine learning.

Between innovative ambitions and erratic everyday practices: urban planners’ ambivalences towards digital transformation

Great expectations are being projected onto digital transformation as a solution to planning under uncertainty and complexity, while demanding new skills from and posing challenges to professionals. We address the ambivalent inclusion of information and communication technologies (ICTs) in everyday planning practices. How do ambivalences towards digitalisation as either a transformative process or as an intrusion upon everyday routines play out in urban planning as a situated knowledge practice? We focus on the erratic embedding of digital technologies and data into planning practices. We gather insights from interviews with planners in small and medium-sized cities in Germany, Austria and Switzerland. The analytical focus on knowing in practice and situated knowledge allows for insights on how digital transformations shape planning from professionals’ perspectives. Our research shows that digital transformation requires infrastructure and routines to bring forward innovative ambitions, which in many contexts are disrupted by limited resources and hierarchical decision-making procedures.

Automated mapping of electronic data capture fields to SDTM.

The goal of this work is to reduce the amount of manual work required to go from data capture to regulatory submission. It will be shown that the use of Siamese networks will allow for the generation of embeddings that can be used by traditional machine learning classifiers to perform the classification at much higher levels of accuracy than standard approaches. Siamese networks are a method for training data embeddings such that data within the same class are closer with respect to a given distance metric than they are to data points in another class. Because they are designed to learn similarity within pairs of data points, they work well in situations where the number of classes is relatively large compared to the number of training samples. In this work, we will show that embeddings generated via a Siamese network from metadata associated with electronic data capture forms can be used to predict the associated SDTM field. With a relatively simple network coupled with a basic classification algorithm, the proposed method can achieve accuracies greater than 90%, which is significantly higher than what has been achieved with traditional methods, with many of the inaccurate mappings due to a lack of training data. In many cases, there is a 15% increase in accuracy vs. more traditional methods. Leveraging Siamese networks, it is possible to generate embeddings that efficiently represent data fields in a lower dimensional space. This allows the creation of a system that can automatically map between data schemas at high levels of accuracy. Such systems represent the first step in automating one of the many labor-intensive data management tasks associated with clinical trials.

Dual-image reversible data hiding based on pixel value parity and multiple embedding strategy

Dual-image reversible data hiding (DI-RDH) has attracted a lot of attention for its excellent embedding capability. However, the stego image after data embedding is degraded, which limits its practical application. This paper proposes an innovative DI-RDH method based on pixel value parity (PVP), which prevents influences from the image texture features and ensures that the pixel value deviation between the temporary stego image and the original image is −1, 0, or 1, as it can partially offset the difference introduced when generating the cover image. Such characteristics enable us to establish a multiple embedding strategy (MES), which applies the difference image to embed the secret message, decreasing the number of invalid shifting pixels in the difference histogram to enhance embedding capacity and image quality. The average maximum embedding rate of our method combining PVP and two phases of MES is determined to be 1.37 bpp, corresponding to the theoretical results. The average peak signal-to-noise ratio of our method on the UCID database is increased by at least 2.19 dB for a given ER of 0.5 bpp compared with several state-of-the-art methods.

Quantum data encoding: a comparative analysis of classical-to-quantum mapping techniques and their impact on machine learning accuracy

This study explores the integration of quantum data embedding techniques into classical machine learning (ML) algorithms; to assess performance enhancements and computational implications across a spectrum of models. We explored various classical-to-quantum mapping methods; ranging from basis encoding and angle encoding to amplitude encoding; for encoding classical data. We conducted an extensive empirical study encompassing popular ML algorithms, including Logistic Regression, K-Nearest Neighbors, Support Vector Machines, and ensemble methods like Random Forest, LightGBM, AdaBoost, and CatBoost. Our findings reveal that quantum data embedding contributes to improved classification accuracy and F1 scores, particularly notable in models that inherently benefit from enhanced feature representation. We observed nuanced effects on running time, with low-complexity models exhibiting moderate increases and more computationally intensive models experiencing discernible changes. Notably, ensemble methods demonstrated a favorable balance between performance gains and computational overhead.This study underscores the potential of quantum data embedding to enhance classical ML models and emphasizes the importance of weighing performance improvements against computational costs. Future research may involve refining quantum encoding processes to optimize computational efficiency and explore scalability for real-world applications. Our work contributes to the growing body of knowledge on the intersection of quantum computing and classical machine learning, offering insights for researchers and practitioners seeking to harness the advantages of quantum-inspired techniques in practical scenarios.

Dynamic Spatial-Temporal Embedding via Neural Conditional Random Field for Multivariate Time Series Forecasting

How to capture dynamic spatial-temporal dependencies remains an open question in multivariate time series (MTS) forecasting. Although recent advanced spatial-temporal graph neural networks (STGNNs) achieve superior forecasting performance, they either consider pre-defined spatial correlations or simply learn static graphs. Some research has tried to learn many adjacent matrices to reveal time-varying spatial correlations, but they generate discrete graphs which cannot encode evolutionary information and also face computational complexity problem. In this article, we propose two significant plugins to help automatically learn enhanced dynamic spatial-temporal embedding of MTS data: (1) a novel neural conditional random field (CRF) layer. We find that the implicit time-varying spatial dependencies are reflected by the explicit changeable links between edges, and we propose the neural CRF to encode such pairwise changeable evolutionary inter-dependencies; (2) a structure adaptive graph convolution (SAGC) that does not require pre-defined graphs to capture semantically richer spatial correlations. Then, we integrate the neural CRF, SAGC with recurrent neural network to develop a new STGNN paradigm termed Adaptive Spatial-Temporal graph neural network with Conditional Random Field (ASTCRF), which can be trained in an end-to-end fashion. We validate the effectiveness, efficiency, and scalability of ASTCRF on five public benchmark MTS datasets.

PSO-based Quaternion Fourier Transform steganography: Enhancing imperceptibility and robustness through multi-dimensional frequency embedding

This paper presents a novel steganography technique using the Quaternion Fourier Transform (QFT) in the 4D frequency domain to enhance imperceptibility and robustness in digital image embedding. Steganography, the art of hiding information within media, faces challenges in balancing security, imperceptibility, and robustness. To address this, we leverage the multi-dimensional properties of quaternions, enabling the embedding of secret data in both grayscale and color images. For grayscale images, two quaternion dimensions are utilized for intensity and secret data, while for color images, all four dimensions are employed with one reserved for metadata. The research question centers on how to maximize spatial dispersion and color similarity while maintaining high imperceptibility and robustness against attacks. Experimental results show that the proposed method improves visual imperceptibility by more than 4 % and exhibits a 13 % increase in robustness against common steganalysis attacks compared to the best state-of-the-art existing technique. These advancements highlight the potential of this method for applications in secure communication, digital watermarking, and copyright protection. By combining the quaternion mathematical framework with a novel optimization strategy, this approach significantly improves upon traditional steganography methods.

Unraveling the developmental heterogeneity within the human retina to reconstruct the continuity of retinal ganglion cell maturation and stage-specific intrinsic and extrinsic factors.

Tissue development is a complex spatiotemporal process with multiple interdependent components. Anatomical, histological, sequencing, and evolutional strategies can be used to profile and explain tissue development from different perspectives. The introduction of scRNAseq methods and the computational tools allows to deconvolute developmental heterogeneity and draw a decomposed uniform map. In this manuscript, we decomposed the development of a human retina with a focus on the retinal ganglion cells (RGC). To increase the temporal resolution of retinal cell classes maturation state we assumed the working hypothesis that that maturation of retinal ganglion cells is a continuous, non-discrete process. We have assembled the scRNAseq atlas of human fetal retina from fetal week 8 to week 27 and applied the computational methods to unravel maturation heterogeneity into a uniform maturation track. We align RGC transcriptomes in pseudotime to map RGC developmental fate trajectories against the broader timeline of retinal development. Through this analysis, we identified the continuous maturation track of RGC and described the cell-intrinsic (DEGs, maturation gene profiles, regulons, transcriptional motifs) and -extrinsic profiles (neurotrophic receptors across maturation, cell-cell interactions) of different RGC maturation states. We described the genes involved in the retina and RGC maturation, including de novo RGC maturation drivers. We demonstrate the application of the human fetal retina atlas as a reference tool, allowing automated annotation and universal embedding of scRNAseq data. Altogether, our findings deepen the current knowledge of the retina and RGC maturation by bringing in the maturation dimension for the cell class vs. state analysis. We show how the pseudotime application contributes to developmental-oriented analyses, allowing to order the cells by their maturation state. This approach not only improves the downstream computational analysis but also provides a true maturation track transcriptomics profile.

Imputing spatial transcriptomics through gene network constructed from protein language model.

Image-based spatial transcriptomic sequencing technologies have enabled the measurement of gene expression at single-cell resolution, but with a limited number of genes. Current computational approaches attempt to overcome these limitations by imputing missing genes, but face challenges regarding prediction accuracy and identification of cell populations due to the neglect of gene-gene relationships. In this context, we present stImpute, a method to impute spatial transcriptomics according to reference scRNA-seq data based on the gene network constructed from the protein language model ESM-2. Specifically, stImpute employs an autoencoder to create gene expression embeddings for both spatial transcriptomics and scRNA-seq data, which are used to identify the nearest neighboring cells between scRNA-seq and spatial transcriptomics datasets. According to the neighbored cells, the gene expressions of spatial transcriptomics cells are imputed through a graph neural network, where nodes are genes, and edges are based on cosine similarity between the ESM-2 embeddings of the gene-encoding proteins. The gene prediction uncertainty is further measured through a deep learning model. stImpute was shown to consistently outperform state-of-the-art methods across multiple datasets concerning imputation and clustering. stImpute also demonstrates robustness in producing consistent results that are insensitive to model parameters.

The art of seeing the elephant in the room: 2D embeddings of single-cell data do make sense.

A recent paper claimed that t-SNE and UMAP embeddings of single-cell datasets are "specious" and fail to capture true biological structure. The authors argued that such embeddings are as arbitrary and as misleading as forcing the data into an elephant shape. Here we show that this conclusion was based on inadequate and limited metrics of embedding quality. More appropriate metrics quantifying neighborhood and class preservation reveal the elephant in the room: while t-SNE and UMAP embeddings of single-cell data do not preserve high-dimensional distances, they can nevertheless provide biologically relevant information.

Latent space arithmetic on data embeddings from healthy multi-tissue human RNA-seq decodes disease modules

Latent space arithmetic on data embeddings from healthy multi-tissue human RNA-seq decodes disease modules

Forensic Investigation and Legal Issues: Hiding Information through Steganography

Abstract: This paper presents information hiding, also known as data embedding, for transmitting secret information through a network. Data security is the prime aspect of conveying information through communication channels. The data hiding and information procedure can be done using several methods, i.e., cryptography, steganography, and digital watermarking. Most people need clarification about cryptography and steganography because both methods are used for hiding information, but there is a minor difference between them. In cryptography, the secret message is hidden, known as “Secret Writing”; in steganography, secret communication is known as “covered Writing”. This paper mainly focuses on steganography for data and information hiding. Steganography is used primarily on terrorism, where the terrorists secretly transmit their data to others globally. The existence of the steganography evaluation model is dominant for stronger development of preferable techniques and the way to cover the extant ones. Despite the ordinary parameters, they are generally acknowledged for the comprehensive assessment of the steganography technique, which is deficient. Nowadays there are many forensic investigations are ongoing on terrorist attacks, which are done through secret transmission of data or information. Many challenges are faced by steganography forensics during the analysis of embedded data. This paper urges digital forensics to be more forward than criminals who use data-hiding techniques that can take advantage of and obscure their criminal activities. This paper concludes the need to develop more robust tools and steganalysis techniques.

Information hiding using approximate POSIT representation

AbstractPOSIT is a numerical system consists of sign, regime, exponential, and fraction bits to overcome some limitations of the IEEE‐754 floating point (FP) representation. This work proposes a technique to hide critical information in the least significant bits (LSBs) of the POSIT FP representation by exploiting approximate computing (AC). The proposed technique, called information hiding using POSIT and LSB (IHUPL), explores the opportunities offered by both the POSIT representation and the AC to hide information with a minimum loss in accuracy and other performance metrics. IHUPL offers two options for hiding information: either by embedding one digit of each character into each pixel or by embedding all the digits of the character at the same time into each pixel of the alpha‐red, green, blue, or black/white image. Experimental results are evaluated for benchmark images and showed that IHUPL enhanced the accuracy of embedding data into LSB of POSIT FP by an average of 5%, the image quality improvement rates of IHUPL are 19% and 16% for options 1 and 2, respectively. Besides the encoding method using IHUPL, the paper outlines an extraction decoding technique that saves the original replacement bits in a key‐image to recover the hidden security message.

A best-match approach for gene set analyses in embedding spaces.

Embedding methods have emerged as a valuable class of approaches for distilling essential information from complex high-dimensional data into more accessible lower-dimensional spaces. Applications of embedding methods to biological data have demonstrated that gene embeddings can effectively capture physical, structural, and functional relationships between genes. However, this utility has been primarily realized by using gene embeddings for downstream machine-learning tasks. Much less has been done to examine the embeddings directly, especially analyses of gene sets in embedding spaces. Here, we propose an Algorithm for Network Data Embedding and Similarity (ANDES), a novel best-match approach that can be used with existing gene embeddings to compare gene sets while reconciling gene set diversity. This intuitive method has important downstream implications for improving the utility of embedding spaces for various tasks. Specifically, we show how ANDES, when applied to different gene embeddings encoding protein-protein interactions, can be used as a novel overrepresentation- and rank-based gene set enrichment analysis method that achieves state-of-the-art performance. Additionally, ANDES can use multiorganism joint gene embeddings to facilitate functional knowledge transfer across organisms, allowing for phenotype mapping across model systems. Our flexible, straightforward best-match methodology can be extended to other embedding spaces with diverse community structures between set elements.

Multi-modal contrastive learning of subcellular organization using DICE.

The data deluge in biology calls for computational approaches that can integrate multiple datasets of different types to build a holistic view of biological processes or structures of interest. An emerging paradigm in this domain is the unsupervised learning of data embeddings that can be used for downstream clustering and classification tasks. While such approaches for integrating data of similar types are becoming common, there is scarcer work on consolidating different data modalities such as network and image information. Here, we introduce DICE (Data Integration through Contrastive Embedding), a contrastive learning model for multi-modal data integration. We apply this model to study the subcellular organization of proteins by integrating protein-protein interaction data and protein image data measured in HEK293 cells. We demonstrate the advantage of data integration over any single modality and show that our framework outperforms previous integration approaches. Availability: https://github.com/raminass/protein-contrastive Contact: raminass@gmail.com.

Correction‐enabled reversible data hiding with pixel repetition for high embedding rate and quality preservation

AbstractA novel correction‐enabled Pixel Repetition (PR)‐based Reversible Data Hiding (RDH) framework, featuring a new embedding scheme is presented. The proposed RDH scheme uses contextually redundant block pixels, generated via PR, in a two‐phase adaptive embedding process, enhancing both image quality and data embedding rates. Specifically, each block encodes 4 bits of data using new mapping conditions that facilitate seed pixel reconstruction from remaining block pixels and provide additional embedding opportunities. Additionally, an innovative post‐embedding error correction technique, based on ‐bit error‐correction, minimises post‐embedding distortion, further improving image quality. This error correction approach augments data embedding robustness, vital for applications like medical imaging, telemedicine, and digital watermarking that requires high embedding capacity with minimum possible distortion. The proposed scheme surpasses existing state‐of‐the‐art methods in embedding rate‐distortion performance, validated through subjective and objective analyses. Furthermore, statistical analysis, including histogram and fragility testing, confirms the scheme's potential for image authentication across diverse multimedia applications. The correction‐enabled RDH with PR offers enhanced embedding capacity and image quality preservation, making it particularly advantageous for applications requiring robust data hiding while maintaining visual fidelity.

Research on blind reversible database watermarking algorithm based on dual embedding strategy

Abstract Massive databases encounter security such as data theft, illegal copying, and copyright infringement during creating, transmitting, and sharing of big data. The reversible data watermarking technology can effectively solve these problems, which can extract the watermark information accurately and recover the original carrier data without any distortion. However, most existing methods extract watermark information non-blindly and cannot effectively achieve a balance between watermark embedding capacity and data distortion. This paper proposes a blind reversible database watermarking method based on dual embedding, which combines histogram shifting and distortion-free watermarking methods to achieve an adaptive selection of histogram bins, blind extraction of watermark information, and carrier data recovery. The proposed method preprocesses the database tuples by scrambling them and constructs a prediction error histogram using first-layer tuples in square prediction within each group. The watermark information is embedded through adaptive selection and expansion of histogram bins, while the distortion-free watermarking method is used in another layer to assist in the recovery of original carrier data. The experimental results show that the proposed method can achieve an embedding capacity of more than three times the capacity of existing methods. It can also achieve blind watermark extraction and outperform some other state-of-the-art methods.

METAPLANTCODE: Harmonizing Plant Metabarcoding Pipelines in Europe

The METAPLANTCODE project is dedicated to advancing and optimizing pan-European case studies on metabarcoding. The project's objectives include providing best practice recommendations, optimizing analysis pipelines for species identification, and creating user-friendly reference databases. To accomplish these objectives, METAPLANTCODE will identify and address gaps in current methodologies, publish best practice documents on FAIR (Findable, Accessible, Interoperable, Reusable) data publishing for plant metabarcode data to GBIF (Global Biodiversity Information Facility) and the INSDC (International Nucleotide Sequence Database Collaboration), and implement ELIXIR-compatible multimodal deep learning (DL) models in novel tools for standalone metabarcoding analyses using various data sources. A significant focus of the project is enhancing species identification accuracy through GBIF records and metadata. This involves mapping regional, national, and international botanical taxonomic checklists, red lists, and floras to the Catalogue of Life (COL) via the COL ChecklistBank. Additionally, taxonomic and floristic literature will be semantically enriched with new entity recognition and relationship extraction modules, supporting the enhanced identification of species through domain-specific descriptive and phenotypic features. An interface will link taxonomic names to treatments, identify homonyms and synonyms, and facilitate the conversion and annotation of floras, red lists, and ecological treatments. All METAPLANTCODE products will adhere to FAIR standards by the project's end. The project emphasizes knowledge transfer from the outset, engaging with associated partners and stakeholders. Key stakeholders will be identified, priorities set, and communication channels established, monitored, and adjusted as necessary. Efforts to enhance stakeholder engagement, training, and outreach will ensure that plant metabarcoding becomes a routine standard for biodiversity monitoring in Europe and beyond. Deep Learning for Plant Metabarcoding Within the METAPLANTCODE project, our team is tasked with improving taxonomic precision by integrating deep learning on metabarcoding data and metadata. Previous studies have demonstrated the applicability of deep learning to non-plant barcoding data and its computational efficiency compared to traditional bioinformatics approaches (Flück et al. 2022). Deep Learning Models for Metabarcoding Data Our approach involves evaluating the efficacy of several deep learning models—such as Convolutional Neural Networks (CNN)(LeCun et al. 2015), Transformer models (Vaswani et al. 2017), Hyena (Poli et al. 2023), and Mamba architectures (Gu et al. 2023)—on plant barcoding datasets. Preliminary results will be presented, highlighting the application of these models and the proposed ensemble method (Mohammed and Kora 2023), which combines multiple barcode sequence representations and learning strategies. The ensemble approach, when integrated with classical machine learning models such as logistic regression and Support Vector Machines (SVM) (Noble 2006), is anticipated to offer improved precision and robustness compared to individual model applications (Fig. 1). Multimodal Refinement of Predictions In the subsequent phase, we aim to refine genetic sequence classifications by employing a multimodal strategy. This approach will integrate genetic information with traditional botanical knowledge. We will utilize biological interaction lists (e.g., species-species, species-habitat) provided by the METAPLANTCODE project to train a large language model (LLM) on relevant scientific literature. This LLM, specifically tailored for plant biodiversity, will incorporate metadata associated with genetic samples (including location, temporality, and climatic conditions). By merging embeddings of both metadata and genetic data, we aim to enhance the accuracy of taxonomic predictions (Fig. 2). Conclusion Through this research, we aim to develop an effective method for integrating genetic data with textual information from various sources. We anticipate that this approach will not only enhance plant metabarcoding but also be applicable to other barcoding fields, such as bacteria, fish, fungi, and more. Additionally, we expect this methodology to find broader applications in genomic research, providing valuable insights and improvements across diverse biological disciplines.

Adaptive Preference Measurement with Unstructured Data

Many products are most meaningfully described using unstructured data such as text or images. Unstructured data are also common in e-commerce, in which products are often described by photos and text but not with standardized sets of attributes. Whereas much is known about how to efficiently measure consumer preferences when products can be meaningfully described by structured attributes, there is scant research on doing the same for unstructured data. This paper introduces a real-time, adaptive survey design framework for measuring preferences over unstructured data, leveraging Bayesian optimization. By adaptively choosing items to display based on uncertainty around a nonparametric utility model, the proposed method maximizes information gain per question, enabling quick estimation of individual-level preferences. The approach operates on embeddings of the unstructured data, thereby eliminating the requirement for manual coding of product attributes. We apply the method to measuring preferences over clothing and highlight its potential for both the general task of marketing research and the specific task of designing customer onboarding surveys to mitigate the cold-start recommendation problem. We also develop methods for interpreting the nonparametric utility functions, which allow us to reconstruct consumer valuations of discrete attributes, even for attributes that were not considered or available a priori. This paper was accepted by Duncan Simester, marketing. Fundings: Funding for this project was provided by Analytics at Wharton, the Wharton Behavioral Lab, and the Wharton Dean’s Fund. The author also thanks the Govil Family for financial support. Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2023.03775 .

Multi-Frequency Entropy for Quantifying Complex Dynamics and Its Application on EEG Data.

Multivariate entropy algorithms have proven effective in the complexity dynamic analysis of electroencephalography (EEG) signals, with researchers commonly configuring the variables as multi-channel time series. However, the complex quantification of brain dynamics from a multi-frequency perspective has not been extensively explored, despite existing evidence suggesting interactions among brain rhythms at different frequencies. In this study, we proposed a novel algorithm, termed multi-frequency entropy (mFreEn), enhancing the capabilities of existing multivariate entropy algorithms and facilitating the complexity study of interactions among brain rhythms of different frequency bands. Firstly, utilizing simulated data, we evaluated the mFreEn's sensitivity to various noise signals, frequencies, and amplitudes, investigated the effects of parameters such as the embedding dimension and data length, and analyzed its anti-noise performance. The results indicated that mFreEn demonstrated enhanced sensitivity and reduced parameter dependence compared to traditional multivariate entropy algorithms. Subsequently, the mFreEn algorithm was applied to the analysis of real EEG data. We found that mFreEn exhibited a good diagnostic performance in analyzing resting-state EEG data from various brain disorders. Furthermore, mFreEn showed a good classification performance for EEG activity induced by diverse task stimuli. Consequently, mFreEn provides another important perspective to quantify complex dynamics.