Composite Representation Research Articles

Comparative analyses of eight primer sets commonly used to target the bacterial 16S rRNA gene for marine metabarcoding-based studies

Next-generation sequencing (NGS), especially metabarcoding, is commonly used to study the diversity and distribution of microbes in diverse ecosystems. The choice of primer set is critical, given the drawbacks of short amplicons and amplicon sequencing bias inherent to metabarcoding. However, comparative analyses of primer sets have rarely been conducted using field samples. In this study, we compared eight commonly used primer sets, all targeting hypervariable regions in the bacterial 16S rRNA gene: 27F/338R (V1–V2), V2f/V3r (V2–V3), PRK341F/PRK806R (V3–V4), 341F/785R (V3–V4), 515F/806RB (V4), 515F/806R (V4), 515F-Y/926R (V4–V5), and B969F/BA1406R (V6–V8). We conducted NGS in triplicate, with >0.8 billion bases in total using coastal seawater samples. The representation of bacterial community composition varied significantly across the eight primer sets, despite being from the same sample. The 27F/338R primer set showed the highest number of operational taxonomic units (OTUs) and read counts, and accounted for 68% of all the order-level taxa found. Remarkably, a novel complementary combination of two primer sets, 27F/338R and 515F/806RB, covered 89% of all the orders that were present. Compared to other primer sets, this combination detected more OTUs of the orders Pelagibacterales and Rhodobacterales, which are ubiquitous in the oceans. As such, use of this combination in future studies may help to reduce diversity bias in ocean-derived samples, in particular temperate coastal samples.

Compositional Scene Representation Learning via Reconstruction: A Survey.

Visual scenes are composed of visual concepts and have the property of combinatorial explosion. An important reason for humans to efficiently learn from diverse visual scenes is the ability of compositional perception, and it is desirable for artificial intelligence to have similar abilities. Compositional scene representation learning is a task that enables such abilities. In recent years, various methods have been proposed to apply deep neural networks, which have been proven to be advantageous in representation learning, to learn compositional scene representations via reconstruction, advancing this research direction into the deep learning era. Learning via reconstruction is advantageous because it may utilize massive unlabeled data and avoid costly and laborious data annotation. In this survey, we first outline the current progress on reconstruction-based compositional scene representation learning with deep neural networks, including development history and categorizations of existing methods from the perspectives of the modeling of visual scenes and the inference of scene representations; then provide benchmarks, including an open source toolbox to reproduce the benchmark experiments, of representative methods that consider the most extensively studied problem setting and form the foundation for other methods; and finally discuss the limitations of existing methods and future directions of this research topic.

An evolutionary variational autoencoder for perovskite discovery

Machine learning (ML) techniques emerged as viable means for novel materials discovery and target property determination. At the vanguard of discoverable energy materials are perovskite crystalline materials, which are known for their robust design space and multifunctionality. Previous efforts for simulating the discovery of novel perovskites via ML have often been limited to straightforward tabular-dataset models and compositional phase-field representations. Therefore, the present study makes a contribution in expanding ML capability by demonstrating the efficacy of a new deep evolutionary learning framework for discovering stable and functional inorganic materials that adopts the complex A2BB′X6 and AA′BB′X6 double perovskite stoichiometries. The model design is called the Evolutionary Variational Autoencoder for Perovskite Discovery (EVAPD), which is comprised of a semi-supervised variational autoencoder (SS-VAE), an evolutionary-based genetic algorithm, and a one-to-one similarity analytical model. The genetic algorithm performs adaptive metaheuristic search operations for finding the most theoretically stable candidates emerging from a target-learnable latent space of the generative SS-VAE model. The integrated similarity analytical model assesses the deviation in three-dimensional atomic coordination between newly generated perovskites and proven standards, and as such, recommends the most promising and experimentally feasible candidates. Using Density Functional Theory (DFT), the novel perovskites are subjected to thorough variable-cell optimization and property determination. The current study presents 137 new perovskite materials generated by the proposed EVAPD model and identifies potential candidates for photovoltaic and optoelectronic applications. The new materials data are archived at NOMAD repository (doi.org/10.17172/NOMAD/2023.05.31-1) and are made openly available to interested users.

Items Outperform Adjectives in a Computational Model of Binary Semantic Classification.

Semantic memory encompasses one's knowledge about the world. Distributional semantic models, which construct vector spaces with embedded words, are a proposed framework for understanding the representational structure of human semantic knowledge. Unlike some classic semantic models, distributional semantic models lack a mechanism for specifying the properties of concepts, which raises questions regarding their utility for a general theory of semantic knowledge. Here, we develop a computational model of a binary semantic classification task, in which participants judged target words for the referent's size or animacy. We created a family of models, evaluating multiple distributional semantic models, and mechanisms for performing the classification. The most successful model constructed two composite representations for each extreme of the decision axis (e.g., one averaging together representations of characteristically big things and another of characteristically small things). Next, the target item was compared to each composite representation, allowing the model to classify more than 1,500 words with human-range performance and to predict response times. We propose that when making a decision on a binary semantic classification task, humans use task prompts to retrieve instances representative of the extremes on that semantic dimension and compare the probe to those instances. This proposal is consistent with the principles of the instance theory of semantic memory.

Multi-view enhanced zero-shot node classification

In recent years, Zero-shot Node Classification (ZNC), an emerging and more difficult task is starting to attract attention, where the classes of testing nodes are unobserved in the training stage. Existing studies for ZNC mainly utilize Graph Neural Networks (GNNs) to construct the feature subspace to align with the classes’ semantic subspace, thus enabling knowledge transfer from seen classes to unseen classes. However, the modeling of the node feature is single-view and unilateral, e.g., the bag-of-words vector, which is not enough to fully describe the characteristics of the node itself. To address this dilemma, we propose to develop the Multi-View Enhanced zero-shot node classification paradigm (MVE) to promote the machine’s generality to approach the human-like thinking mode. Specifically, multi-view features are obtained from different aspects such as pre-trained model embeddings, knowledge graphs, statistic methods, and then fused by a contrastive learning module into the compositional node representation. Meanwhile, a developed Graph Convolutional Network (GCN) is used to make the nodes fully absorb the information of neighbors while the over-smooth issue is alleviated by multi-view features and the proposed contrastive learning mechanism. Experimental results conducted on three public datasets show an average 25% improvement compared to baseline methods, proving the superiority of our multi-view learning framework. The code and data can be found at https://github.com/guaiqihen/MVE.

Hardness prediction of high entropy alloys with periodic table representation of composition, processing, structure and physical parameters

A rational and information-rich material representation is necessary for a machine learning model to successfully predict material properties. In this work, the periodic table representation (PTR) of composition, processing, structure and physical parameters is established and explored for hardness prediction of high entropy alloys. Based on convolutional neural network (CNN) with PTR, hardness prediction, hardness classification and alloy system extrapolation are achieved to demonstrate the prediction performance of the newly proposed model. By further introducing phase classification results with five mutually non-exclusive phases, predicted by CNN coupled with PTR, the results indicate that adding phase information is beneficial for hardness prediction. In addition, the stacking method is used to further improve the accuracy and stability of hardness prediction. The Diebold-Mariano test is used for model comparisons, and the results demonstrate that the improvement in prediction performance caused by information addition is statistically reliable.

Quantifying ant diversity and community in a subalpine forest mosaic: a comparison of two methods

Sampling efficiency, composition and detection biases associated with pitfall-trap and sample plot (standardised hand-collecting) methods were compared at seven high montane sites at Lasha Mountain, Yunnan, China. Rarefaction-interpolation curves showed total species diversity and sample coverage from each method were undifferentiated. Permutational multivariate analysis of variance showed highly overlapping communities. However, of all species collected, just 53% of all species found were shared between the two methods; 30% were exclusive to sample plot samples and 17% were exclusive to pitfall traps. Modelled comparisons of species richness (alpha) and proportion of species from the population (beta diversity) showed differences between methods. Permutational multivariate analysis of variance blocked by site showed community composition differed between methods, defined largely by the presence of a few dominant species.Implications for insect conservation: Our findings suggest that results from the two methods cannot be directly compared and are imperfect substitutes to one another. For long-term monitoring of biodiversity in complex forest mosaic systems, we suggest integrating a suite of complementary methods to achieve more complete representation of ant composition and diversity.

Phytoplankton diversity explained by connectivity across a mesoscale frontal system in the open ocean

Phytoplankton community composition is important in establishing ecosystem structure and function. Intuitively, we recognize that water movements must be important for modifying spatial gradients and plankton diversity. However, identifying boundaries and exchange between habitats in the open ocean is not straightforward. Here, we use the abundance of nine phytoplankton species closely sampled in a mesoscale frontal system in the northeastern North Sea as a proxy for community composition and explore the relationship between phytoplankton biogeography and transport patterns. Subsurface community distributions could be related to modeled patterns in water movement. A methodology for analyzing pelagic diversity that includes a representation of plankton community composition and an Eulerian connectivity tracer was developed, and the relative importance of connectivity and geographical distance for phytoplankton species composition analyzed. The connectivity tracer identifies timescales and dispersal barriers in the open ocean. Connectivity was found to be superior in explaining pelagic plankton diversity and found to be a prerequisite for understanding the pelagic phytoplankton composition. This approach is a valuable tool for establishing the link between ocean transports, ecosystem structure and biodiversity and for informing the placement of marine protected areas.

AvatarReX: Real-time Expressive Full-body Avatars

We present AvatarReX, a new method for learning NeRF-based full-body avatars from video data. The learnt avatar not only provides expressive control of the body, hands and the face together, but also supports real-time animation and rendering. To this end, we propose a compositional avatar representation, where the body, hands and the face are separately modeled in a way that the structural prior from parametric mesh templates is properly utilized without compromising representation flexibility. Furthermore, we disentangle the geometry and appearance for each part. With these technical designs, we propose a dedicated deferred rendering pipeline, which can be executed at a real-time framerate to synthesize high-quality free-view images. The disentanglement of geometry and appearance also allows us to design a two-pass training strategy that combines volume rendering and surface rendering for network training. In this way, patch-level supervision can be applied to force the network to learn sharp appearance details on the basis of geometry estimation. Overall, our method enables automatic construction of expressive full-body avatars with real-time rendering capability, and can generate photo-realistic images with dynamic details for novel body motions and facial expressions.

Risk Assessment of the Possible Intermediate Host Role of Pigs for Coronaviruses with a Deep Learning Predictor.

Swine coronaviruses (CoVs) have been found to cause infection in humans, suggesting that Suiformes might be potential intermediate hosts in CoV transmission from their natural hosts to humans. The present study aims to establish convolutional neural network (CNN) models to predict host adaptation of swine CoVs. Decomposing of each ORF1ab and Spike sequence was performed with dinucleotide composition representation (DCR) and other traits. The relationship between CoVs from different adaptive hosts was analyzed by unsupervised learning, and CNN models based on DCR of ORF1ab and Spike were built to predict the host adaptation of swine CoVs. The rationality of the models was verified with phylogenetic analysis. Unsupervised learning showed that there is a multiple host adaptation of different swine CoVs. According to the adaptation prediction of CNN models, swine acute diarrhea syndrome CoV (SADS-CoV) and porcine epidemic diarrhea virus (PEDV) are adapted to Chiroptera, swine transmissible gastroenteritis virus (TGEV) is adapted to Carnivora, porcine hemagglutinating encephalomyelitis (PHEV) might be adapted to Primate, Rodent, and Lagomorpha, and porcine deltacoronavirus (PDCoV) might be adapted to Chiroptera, Artiodactyla, and Carnivora. In summary, the DCR trait has been confirmed to be representative for the CoV genome, and the DCR-based deep learning model works well to assess the adaptation of swine CoVs to other mammals. Suiformes might be intermediate hosts for human CoVs and other mammalian CoVs. The present study provides a novel approach to assess the risk of adaptation and transmission to humans and other mammals of swine CoVs.

A Composite Analogy to Study the Linear Elasticity of a Pressurized Latex Tube with Application to a Mechanical Vocal Fold Replica

Mechanical deformable vocal fold replicas are an inherent part of physical studies of the fluid–structure interaction underlying vocal folds auto-oscillation during voiced speech sound production. In this context, the current work considers the linear stress–strain characterization of a pressurized latex tube vocal fold replica. An imaging approach is developed to measure the effective low-strain linear Young’s moduli along the streamwise (49[Formula: see text]kPa) and transverse main auto-oscillation (44[Formula: see text]kPa) directions. Next, a composite analogy is proposed to model the replica’s structure as an equivalent de-homogenized multi-layer material with two, three or four layers. This way equivalent low-strain Young’s moduli of each equivalent single layer can be estimated. Both measured effective and modeled equivalent low-strain Young’s moduli are within the range up to 65 kPa associated with human vocal folds. Resulting equivalent composite representations are of interest for the future design of pressurized latex tube replicas. This is illustrated considering the influence of outer layer latex properties on the overall estimate of the effective Young’s modulus. The proposed analogy is thus efficient in contributing to the direct comparison, in terms of low-strain elastic behavior, between replicas.

CSR-SVM: Compositional semantic representation for intelligent identification of engineering change documents based on SVM

Timely and accurate identification of engineering change documents plays an important role in the document management system of the Architecture, Engineering, and Construction (AEC) industry, which facilitates decision making. The current way of manual review and analysis may cause a delay or omission in information transmission, which may severely affect the project schedule and cost. This paper adopts text classification to explore the intelligent identification of unstructured engineering change documents. However, the paucity of available corpus and the limited number of engineering change texts are challenges in practice. Additionally, the semantic information of texts cannot be elegantly exploited. Especially there are unregistered words that interfere with semantics in engineering change texts. To tackle these problems, we propose a compositional semantic representation (CSR) and develop an SVM-based method named CSR-SVM. We introduce a language model to produce word embeddings; here, a domain dictionary is established for unregistered words. The embeddings are then exploited in CSR so that CSR can incorporate both key and global semantic representations. The former is obtained based on dependency parsing and word embeddings, and the latter is obtained according to all the word representations of a text. The CSR provides sufficient semantic representation for SVM in an efficient way. The advantages of CSR-SVM have been validated by experiments on a real-world dataset.

Efficient Decoding of Compositional Structure in Holistic Representations.

We investigate the task of retrieving information from compositional distributed representations formed by hyperdimensional computing/vector symbolic architectures and present novel techniques that achieve new information rate bounds. First, we provide an overview of the decoding techniques that can be used to approach the retrieval task. The techniques are categorized into four groups. We then evaluate the considered techniques in several settings that involve, for example, inclusion of external noise and storage elements with reduced precision. In particular, we find that the decoding techniques from the sparse coding and compressed sensing literature (rarely used for hyperdimensional computing/vector symbolic architectures) are also well suited for decoding information from the compositional distributed representations. Combining these decoding techniques with interference cancellation ideas from communications improves previously reported bounds (Hersche et al., 2021) of the information rate of the distributed representations from 1.20 to 1.40 bits per dimension for smaller codebooks and from 0.60 to 1.26 bits per dimension for larger codebooks.

Genomic representation predicts an asymptotic host adaptation of bat coronaviruses using deep learning.

Coronaviruses (CoVs) are naturally found in bats and can occasionally cause infection and transmission in humans and other mammals. Our study aimed to build a deep learning (DL) method to predict the adaptation of bat CoVs to other mammals. The CoV genome was represented with a method of dinucleotide composition representation (DCR) for the two main viral genes, ORF1ab and Spike. DCR features were first analyzed for their distribution among adaptive hosts and then trained with a DL classifier of convolutional neural networks (CNN) to predict the adaptation of bat CoVs. The results demonstrated inter-host separation and intra-host clustering of DCR-represented CoVs for six host types: Artiodactyla, Carnivora, Chiroptera, Primates, Rodentia/Lagomorpha, and Suiformes. The DCR-based CNN with five host labels (without Chiroptera) predicted a dominant adaptation of bat CoVs to Artiodactyla hosts, then to Carnivora and Rodentia/Lagomorpha mammals, and later to primates. Moreover, a linear asymptotic adaptation of all CoVs (except Suiformes) from Artiodactyla to Carnivora and Rodentia/Lagomorpha and then to Primates indicates an asymptotic bats-other mammals-human adaptation. Genomic dinucleotides represented as DCR indicate a host-specific separation, and clustering predicts a linear asymptotic adaptation shift of bat CoVs from other mammals to humans via deep learning.

Size fractionated biogeochemical constituents across adjacent coastal systems informs approaches for integrating small catchment studies into regional models

AbstractObserved and predicted hydrological changes in C‐rich boreal ecosystems have the ability to change the transport trajectory of biogeochemical constituents from the land to ecologically, economically, and culturally important coastal systems. Yet, most of our current understanding of biogeochemical fluxes and cycling across salinity gradients stem from observations of large and urbanized riverine systems, which overlooks the numerically abundant smaller systems. In this study, we conducted a baseline assessment of the biogeochemical constituents across salinity gradients among two adjacent small systems in the boreal zone. Dissolved iron (DFe) and its ratio with dissolved organic carbon (DFe : DOC) were the most sensitive indicators for small catchment heterogeneity. These parameters were the best indicators of change among coastal systems across regional and seasonal scales. Our results also confirm consistencies in common optical measures (SUVA254 and S[275–295/350–400]) and DOC to nitrogen ratios that may adequately provide representation of biogeochemical composition on a regional scale. Simultaneous variation in biogeochemical parameters across particulate and dissolved pools during the summer‐to‐fall transition period indicate this as an important timeframe for targeted investigation of the linkages between biogeochemical parameters and coastal ecosystem functioning. By providing some key spatial and temporal constraints on biogeochemical fluxes among boreal river‐estuaries, our findings indicate that DFe and DFe : DOC ratios should be used to design research aimed at capturing regional and coastal ecosystem scale biogeochemical fluxes to inform Earth System Models.

A composite trend representation-based tracking system with historical portfolio data for portfolio optimization

In this paper, we propose a novel tracking system based on composite trend representation and historical portfolio data (CTRHP) for portfolio optimization (PO). In the part of obtaining prediction of price, we introduce the important data of historical portfolio, which is rarely utilized before, to improve the accuracy of measuring investment performance. In addition, we propose a set of correlation coefficient-based similarity measurement functions (CSMFs) to automatically assign different weights to different trend representations, which enables each trend representation to have an impact on future price predictions and set the strength according to their investment performance. In the part of portfolio optimization, a new optimization objective based on generalized increasing factors is proposed to optimize the investment portfolio, and a fast solution algorithm is presented. Extensive experiments on six standard datasets from real financial markets across different assets and different time horizons show that our CTRHP achieves significantly better performance compared with previous state-of-the-art PO systems in investing returns and risk control. Moreover, it has the advantages of being able to tolerate certain transaction fees and running fast, which shows that it is suitable for real financial environments.

A multi-view-CNN framework for deep representation learning in image classification

Deep representation learning in image classification is an area in computer vision where deep Convolutional Neural Networks (CNNs) have flourished. Nevertheless, developing an efficient image recognition model for real world applications is a challenging task, since image datasets are characterized by instances with a large amount of noise and redundant information. Thus, it is essential to incorporate an intelligent feature extraction and filtering method in order to create robust and efficient image representations. In this work, we propose a Multi-View-CNN framework which drastically boosts the performance of pre-trained CNN models, such as ResNet and VGG in image classification applications. In this approach different type of views of the same initial image are used in order to extract different types of features utilizing pre-trained CNN models. However, in order to reduce the huge dimensional size of the raw CNN’s output features and create a robust image representation, the Principal Component Analysis (PCA) dimension reduction method is applied. Then, all these extracted feature vectors are concatenated building a final composite feature representation of the initial image dataset. Finally, this augmented feature vector is used for training a linear model (Logistic Regression) in order to perform the final classification tasks. The main findings of this work are summarized as follows. First, the proposed Multi-View-CNN framework managed to drastically increase the performance results of pre-trained CNN models. Second, the incorporation of PCA as a final layer into the main CNN topology, instead of using the classical dimension reduction layer components such as Averaging and Max Pooling operations, managed to significantly improve the performance. The whole implementation code of this framework alongside with the datasets used in our experimental simulations was uploaded to our public GitHub repository to the following link: https://github.com/EmmanuelPintelas/A-Multi-View-CNN-Framework-for-Deep-Representation-Learning-in-Image-Classification.

In the fast lane: Receptor trafficking during status epilepticus.

Status epilepticus (SE) remains a significant cause of morbidity and mortality and often is refractory to standard first-line treatments. A rapid loss of synaptic inhibition and development of pharmacoresistance to benzodiazepines (BZDs) occurs early during SE, while NMDA and AMPA receptor antagonists remain effective treatments after BZDs have failed. Multimodal and subunit-selective receptor trafficking within minutes to an hour of SE involves GABA-A, NMDA, and AMPA receptors and contributes to shifts in the number and subunit composition of surface receptors with differential impacts on the physiology, pharmacology, and strength of GABAergic and glutamatergic currents at synaptic and extrasynaptic sites. During the first hour of SE, synaptic GABA-A receptors containing γ2 subunits move to the cell interior while extrasynaptic GABA-A receptors with δ subunits are preserved. Conversely, NMDA receptors containing N2B subunits are increased at synaptic and extrasynaptic sites, and homomeric GluA1 ("GluA2-lacking") calcium permeant AMPA receptor surface expression also is increased. Molecular mechanisms, largely driven by NMDA receptor or calcium permeant AMPA receptor activation early during circuit hyperactivity, regulate subunit-specific interactions with proteins involved with synaptic scaffolding, adaptin-AP2/clathrin-dependent endocytosis, endoplasmic reticulum (ER) retention, and endosomal recycling. Reviewed here is how SE-induced shifts in receptor subunit composition and surface representation increase the excitatory to inhibitory imbalance that sustains seizures and fuels excitotoxicity contributing to chronic sequela such as "spontaneous recurrent seizures" (SRS). A role for early multimodal therapy is suggested both for treatment of SE and for prevention of long-term comorbidities.

Effects of lead and lean in multi-axis directed energy deposition

The present study examines the effect of varying laser incidence angles on textural, microstructural, and geometric characteristics of directed energy deposition (DED) processed materials, providing a more comprehensive outlook on participating laser-matter interaction phenomena and ultimately devising strategies to ameliorate print performance. In this study, single-layer, single-/multi-track specimens were processed to examine the effect of non-orthogonal angular configurations on bead morphology, microstructure, phase composition, and textural representation of DED-processed 316L stainless steel materials. It was observed that bead size decreased at increasing lead and lean angles. Asymmetry in the distribution of the bead morphology as a function of lead angle indicates better catchment for acute lead angle configurations over obtuse configurations. No significant differences in phase composition, texture, and microstructure were observed in moderate off-axis configurations. When the penetration depth for the deposits was below 20 μm, columnar structures dominated the microstructure of the deposited material. At deeper penetration depths, columnar and equiaxed structures were observed at the bead-substrate interface and center of the bead, respectively. Compared to powder-blown DED, wire-DED dilution profiles were found to be asymmetric in both orthogonal and non-orthogonal wire DED samples.

Efficient On-Device Session-Based Recommendation

On-device session-based recommendation systems have been achieving increasing attention on account of the low energy/resource consumption and privacy protection while providing promising recommendation performance. To fit the powerful neural session-based recommendation models in resource-constrained mobile devices, tensor-train decomposition and its variants have been widely applied to reduce memory footprint by decomposing the embedding table into smaller tensors, showing great potential in compressing recommendation models. However, these model compression techniques significantly increase the local inference time due to the complex process of generating index lists and a series of tensor multiplications to form item embeddings. The resultant on-device recommender fails to provide real-time responses and recommendations. To improve the online recommendation efficiency, we propose to learn compositional encoding-based compact item representations. Specifically, each item is represented by a compositional code that consists of several codewords, and we learn embedding vectors to represent each codeword instead of each item. Then the composition of the codeword embedding vectors from different embedding matrices (i.e., codebooks) forms the item embedding. Since the size of codebooks can be extremely small, the recommender model is thus able to fit in resource-constrained devices and save the codebooks for fast local inference. Besides, to prevent the loss of model capacity caused by compression, we propose a bidirectional self-supervised knowledge distillation framework. Extensive experimental results on two benchmark datasets demonstrate that compared with existing methods, the proposed on-device recommender not only achieves an 8x inference speedup with a large compression ratio but also shows superior recommendation performance. The code is released at https://github.com/xiaxin1998/EODRec.