Contrast Features Research Articles

Electronic Structures of Late versus Early Transition Metal Imido Complexes from 15N NMR Signatures

Imido ligand is a ubiquitous motif in organometallic chemistry, serving roles spanning from ancillary ligands to reactive sites. The nature of M=N bond is highly depended on the metal centres and their d‐electron configuration, with late‐transition metal (LTM) imido complexes exhibiting contrasting features when compared to their early‐transition metal (ETM) analogues. Envisioning to uncover general electronic descriptor for the nature of imido ligands, we computationally investigate the solid‐state 15N NMR signatures of LTM imido complexes with various central metals, geometries and d‐electron counts, and compare them against these of the corresponding ETM systems. The spectroscopic signatures are mostly driven by the presence of filled, π‐symmetry orbitals in LTM imido complexes, suggesting the development of high‐lying π(M=N) and low‐lying σ/σ*(M=N) orbitals. This contrasts with what is observed for the reported ETM systems, for which high‐lying σ‐type orbitals determine the NMR signature. Noteworthily, Ni‐ and Pd‐imido complexes with d8 configurations exhibit highly asymmetric nitrogen‐15 NMR signature with extremely deshielded principal components, because of the presence of filled, high‐lying antibonding π*(M=N) orbitals, consistent with their high reactivity. The sensitive response of 15N NMR signature to the nature of metal sites further highlights that chemical shift is a useful reactivity descriptor.

Short‐term effects of food waste composts on physicochemical soil quality and horticultural crop production

AbstractAimComposts made from food waste will soon become more widespread on the market thanks to the upcoming enforcement of the legal obligation to sort biowaste. Our experiment aims at improving knowledge on the short‐term effects of these composts on soils’ physicochemical properties and vegetable crops.MethodsThree composts with contrasting characteristics were tested: a 100% v/v green waste compost (C1), and two composts composed of 50% v/v food waste and 50% v/v green waste, one prepared directly on the soil (C2) and the other from a competing producer who have the French NFU 44‐051 (AFNOR NF U 44‐051, 2006) certification for an organic amendment (C3). They were applied at a rate of 100 t ha−1 (dry matter) on two cropped soils with contrasting textures. Soil‐and‐compost mixes and compost‐free soil were planted with lettuce, radish, and potato.ResultsSeventy‐four days after planting, composts improved some soil physicochemical properties. The compost‐amended soils had better saturated hydraulic conductivity (Ks, 1 10−3–2.5 10−3 cm s−1) than the compost‐free soil (0.5 10−3 cm s−1), and water‐stable aggregates were higher than the initial value in C3 soil, equal to it in C2 soil, and lower in C1 soil. pH, total nitrogen, and organic carbon increased in all compost‐amended soils. Food waste compost stimulated crop production. The yields (dry matter) of all three crops were two to three times higher in the two soils amended with food waste compost compared to unamended soil, whereas they decreased almost two times in the soil amended with green waste compost due to nitrogen immobilization. Trace metals (particularly Pb and Cd) added by the composts, although present in edible parts of the plants, did not exceed the European rules for trace metals.ConclusionThus, food waste composts have positive effects on soils and vegetable crops, and the higher their organic matter content, the higher these positive effects.

Irreversibility analysis of radiative TiO2+Al2O3+Cu/H2O ternary nanofluid flow over a bidirectional stretching sheet with cattaneo-christov heat flux: nanotechnology applications

Ternary nanofluids demonstrate better heat transfer characteristics in contrast to conventional liquids and typical hybrid nanofluids. These are applied in sophisticated cooling systems for electronics, heat exchangers, and automotive engines, along with renewable energy systems such as solar collectors, where efficient heat transfer plays a crucial role. The aim of this research is to investigate the movement of a Casson ternary nanofluid passing through a bidirectional exponential sheet by employing the innovative Cattaneo-Christov heat flux concept. The utilization of the energy equation considers thermal radiation, viscous dissipation, and heat source/sink effects, with the integration of chemical reaction effects into the concentration equation. An analysis of entropy generation is utilized to evaluate the thermodynamic irreversibility within the system. The conversion of transport equations involves a transformation from partial to ordinary differential equations, followed by a numerical solution utilizing the BVP4C solver embedded in the MATLAB package R2022b. The impacts of developed factors on thermal, concentration, and velocity behavior, as well as engineering quantities, are thoroughly explored graphically. The outcome reveals that the velocity gradients diminish as magnetic fields intensify, while it amplified by the Hall factor. The rise in temperature of ternary nanofluid correlates with elevated levels of radiation, and Brownian motion. Concentration intensifies with the rapid development of thermophoresis influences. Heightened values of the Reynolds and Brinkman numbers give rise to amplified entropy production but a decrease in the Bejan number. The ternary nanofluid displays a remarkable 8.92% increase in skin friction on the x-axis and y-axis, influenced by the potent Darcy-Forchheimer factor. The rates of mass and heat transfer in nanofluids undergo a decrease of 8.58% and 12.52%, respectively, due to the heightened influence of Brownian motion and Eckert number. The results could provide valuable insights into the performance of ternary nanofluids in various industrial environments under specific conditions.

An efficient multiscale enhancement network with attention mechanism for aluminium defect detection

ABSTRACT Real-time defect detection is required to efficiently control the quality of aluminium. However, aluminium defects have the characteristics of small-size, low contrast, and multiscale variations, which pose great challenges to defect detection. This article aims to improve the defect detection accuracy and propose an effective multiscale enhancement network with attention mechanism for aluminium defect detection (AMMENet). First, to capture key features and mitigate interference from the background, a pluggable parallel residual attention module (PRAM) is proposed for the feature extraction network. To compensate for the loss of deep features, a multilevel semantic enhancement module (C2f-MFF) is proposed to fuse multiscale feature maps. Finally, the model was applied to the Tianchi Aluminium Surface Defect Dataset (TC-ASDD) for ablation experiments and comparisons. The experimental results show that the mean average precision (mAP@0.5) of the proposed AMMENet is 73.6% with a real-time detection speed of 66.2 frames per second (FPS). Compared with YOLOv8 baseline network, AMMENet improves mAP@0.5 by 2.8% with only a slight loss in speed. Moreover, AMMENet is superior to the state-of-the-art detection methods in terms of detection accuracy.

T1- and T2-weighted MRI signal and histology findings in suboptimally fixed human brains

Neuroscientific research that requires brain tissue depends on brain banks that provide very small tissue samples fixed by immersion in neutral-buffered formalin (NBF), while anatomy laboratories could provide full brain specimens. However, these brains are generally fixed by perfusion of the full body with solutions other than NBF generally used by brain banks, such as an alcohol-formaldehyde solution (AFS) that is typically used for dissection and teaching. Therefore, fixation quality of these brains needs to be assessed to determine their usefulness in post-mortem investigations through magnetic resonance imaging (MRI) and histology, two common neuroimaging modalities. Here, we report the characteristics of five brains fixed by full body perfusion of an AFS from our Anatomy Laboratory suspected of being poorly fixed, given the altered signal seen on T1w MRI scans in situ. We describe 1- the characteristics of the donors; 2- the fixation procedures applied for each case; 3- the tissue contrast characteristics of the T1w and T2w images; 4- the macroscopic tissue quality after extraction of the brains; 5- the macroscopic arterial characteristics and presence or absence of blood clots; and 6- four histological stains of the areas that we suspected were poorly fixed. We conclude that multiple factors can affect the fixation quality of the brain. Nevertheless, cases in which brain fixation is suboptimal, consequently altering the T1w signal, still have T2w of adequate gray-matter to white-matter contrast and may also be used for histology stains with sufficient quality.

Adapting to change: resolving the dynamic and dual roles of NCK1 and NCK2.

Adaptor proteins play central roles in the assembly of molecular complexes and co-ordinated activation of specific pathways. Through their modular domain structure, the NCK family of adaptor proteins (NCK1 and NCK2) link protein targets via their single SRC Homology (SH) 2 and three SH3 domains. Classically, their SH2 domain binds to phosphotyrosine motif-containing receptors (e.g. receptor tyrosine kinases), while their SH3 domains bind polyproline motif-containing cytoplasmic effectors. Due to these functions being established for both NCK1 and NCK2, their roles were inaccurately assumed to be redundant. However, in contrast with this previously held view, NCK1 and NCK2 now have a growing list of paralog-specific functions, which underscores the need to further explore their differences. Here we review current evidence detailing how these two paralogs are unique, including differences in their gene/protein regulation, binding partners and overall contributions to cellular functions. To help explain these contrasting characteristics, we then discuss SH2/SH3 structural features, disordered interdomain linker regions and post-translational modifications. Together, this review seeks to highlight the importance of distinguishing NCK1 and NCK2 in research and to pave the way for investigations into the origins of their interaction specificity.

Modeling of artificial intelligence-based respiratory motion prediction in MRI-guided radiotherapy: a review.

The advancement of precision radiotherapy techniques, such as volumetric modulated arc therapy (VMAT), stereotactic body radiotherapy (SBRT), and particle therapy, highlights the importance of radiotherapy in the treatment of cancer, while also posing challenges for respiratory motion management in thoracic and abdominal tumors. MRI-guided radiotherapy (MRIgRT) stands out as state-of-art real-time respiratory motion management approach owing to the non-ionizing radiation nature and superior soft-tissue contrast characteristic of MR imaging. In clinical practice, MR imaging often operates at a frequency of 4Hz, resulting in approximately a 300 ms system latency of MRIgRT. This system latency decreases the accuracy of respiratory motion management in MRIgRT. Artificial intelligence (AI)-based respiratory motion prediction has recently emerged as a promising solution to address the system latency issues in MRIgRT, particularly for advanced contour prediction and volumetric prediction. However, implementing AI-based respiratory motion prediction faces several challenges including the collection of training datasets, the selection of prediction methods, and the formulation of complex contour and volumetric prediction problems. This review presents modeling approaches of AI-based respiratory motion prediction in MRIgRT, and provides recommendations for achieving consistent and generalizable results in this field.

Contrastive Functional Connectivity Defines Neurophysiology-informed Symptom Dimensions in Major Depression.

Major depressive disorder (MDD) is a prevalent psychiatric disorder characterized by substantial clinical and neurobiological heterogeneity. Conventional studies that solely focus on clinical symptoms or neuroimaging metrics often fail to capture the intricate relationship between these modalities, limiting their ability to disentangle the complexity in MDD. Moreover, patient neuroimaging data typically contains normal sources of variance shared with healthy controls, which can obscure disorder-specific variance and complicate the delineation of disease heterogeneity. We employed contrastive principal component analysis to extract disorder-specific variations in fMRI-based resting-state functional connectivity (RSFC) by contrasting MDD patients (N=233) with age-matched healthy controls (N=285). We then applied sparse canonical correlation analysis to identify latent dimensions in the disorder variations by linking the extracted contrastive connectivity features to clinical symptoms in MDD patients. Two significant and generalizable dimensions linking distinct brain circuits and clinical profiles were discovered. The first dimension, associated with an apparent "internalizing-externalizing" symptom dimension, was characterized by self-connections within the visual network and also associated with choice reaction times of cognitive tasks. The second dimension, associated with personality facets such as extraversion and conscientiousness typically inversely associated with depression symptoms, is primarily driven by self-connections within the dorsal attention network. This "depression-protective personality" dimension is also associated with multiple cognitive task performances related to psychomotor slowing and cognitive control. Our contrastive RSFC-based dimensional approach offers a new avenue to dissect clinical heterogeneity underlying MDD. By identifying two stable, neurophysiology-informed symptom dimensions in MDD patients, our findings may enhance disease mechanism insights and facilitate precision phenotyping, thus advancing the development of targeted therapeutics for precision mental health.

Enhancing ophthalmology medical record management with multi-modal knowledge graphs

The electronic medical record management system plays a crucial role in clinical practice, optimizing the recording and management of healthcare data. To enhance the functionality of the medical record management system, this paper develops a customized schema designed for ophthalmic diseases. A multi-modal knowledge graph is constructed, which is built upon expert-reviewed and de-identified real-world ophthalmology medical data. Based on this data, we propose an auxiliary diagnostic model based on a contrastive graph attention network (CGAT-ADM), which uses the patient’s diagnostic results as anchor points and achieves auxiliary medical record diagnosis services through graph clustering. By implementing contrastive methods and feature fusion of node types, text, and numerical information in medical records, the CGAT-ADM model achieved an average precision of 0.8563 for the top 20 similar case retrievals, indicating high performance in identifying analogous diagnoses. Our research findings suggest that medical record management systems underpinned by multimodal knowledge graphs significantly enhance the development of AI services. These systems offer a range of benefits, from facilitating assisted diagnosis and addressing similar patient inquiries to delving into potential case connections and disease patterns. This comprehensive approach empowers healthcare professionals to garner deeper insights and make well-informed decisions.

Classifying High-Frequency Oscillations by Morphologic Contrast to Background, With Surgical Outcome Correlates.

Ictal high-frequency oscillations (HFOs) are a reliable indicator of a seizure onset zone for intracranial EEG recordings. Interictal HFOs often are also observed and may be a useful biomarker to supplement ictal data, but distinguishing pathologic from physiologic HFOs continues to be a challenging task. We present a method of classifying HFOs based on morphologic contrast to the background. We retrospectively screened 31 consecutive patients who underwent intracranial recordings for epilepsy at Stanford Medical Center during a 2-year period, and 13 patients met the criteria for inclusion. Interictal EEG data were analyzed using an automated event detector followed by morphologic feature extraction and k-means clustering. Instead of only using event features, the algorithm also incorporated features of the background adjacent to the events. High-frequency oscillations with higher morphologic contrast to the background were labeled as pathologic, and "hotspots" with the most active pathologic HFOs were identified and compared with clinically determined seizure onset zones. Clustering with contrast features produced groups with better separation and more consistent boundaries. Eleven of the 13 patients proceeded to surgery, and patients whose hotspots matched seizure onset zones had better outcomes, with 4 out of 5 "match" patients having no disabling seizures at 1+ year postoperatively (Engel I or International League Against Epilepsy Class 1-2), while all "mismatch" patients continued to have disabling seizures (Fisher exact test P-value = 0.015). High-frequency oscillations with higher contrast to background more likely represent paroxysmal bursts of pathologic activity. Patients with HFO hotspots outside of identified seizure onset zones may not respond as well to surgery.

2D23D: 3D images from 2D images using statistical filters and color shading

Below is the development of an algorithm that allows generating 3D models from 2D images for analysis and visualization, with an application focused on images from the health and space sectors, domains chosen for their contrast characteristics. The characteristics used in the project are presented and determined, as well as 3D programming and visualization environment tools that allow the generation of voxel depth, as well as the management of image formats from the medical or spatial sector, including the results obtained with images. from both sectors, from which the 3D models were created.

Automatic medical report generation combining contrastive learning and feature difference

The automatic medical report generation is a challenging task because it requires accurate capture and description of abnormal regions, especially for those discrepancies between patient and normal. In most cases, normal region descriptions dominate the entire medical report, and existing methods may fail to focus on abnormal regions due to data bias. Medical reports can be automatically generated by combining contrastive learning with feature difference in order to capture and describe abnormal regions effectively. By capturing discrepancy attributes between the input image and normal images, this method can provide more accurate diagnostic reports and better represent the visual features of abnormal regions. Specifically, we propose the feature difference approach to make the model focus more on abnormal regions, and on the other hand, we propose the combination of contrastive learning for enhancing the visual representation of feature difference based on the feature difference approach, thus improving the performance of the model. Experimental results on the IU-Xray and MIMIC-CXR datasets demonstrate the effectiveness of our approach.

Soil Carbon Accumulation Under Afforestation Is Driven by Contrasting Responses of Particulate and Mineral‐Associated Organic Carbon

AbstractAfforestation is widely believed to sequester carbon (C) in soil. However, the effect of afforestation on soil organic C (SOC) accumulation is still debated due to the contrasting features of particulate and mineral‐associated organic C (POC and MAOC). We conducted a field investigation of 144 paired sampling sites by comparing afforested and non‐afforested lands to investigate the POC and MAOC dynamics after afforestation across the Danjiangkou basin in subtropical China, where forests are dominated by Platycladus orientalis, Quercus variabilis and Pinus massoniana. The average contents of SOC, POC, and MAOC were significantly increased by afforestation; however, POC and MAOC responded differently to afforestation type. All afforestation types promoted the POC content, and MAOC also showed positive responses to afforestation except that afforestation with P. massoniana from shrubland significantly reduced the MAOC content. With increasing SOC content, the POC grew at a faster rate than MAOC at high SOC levels. Afforestation hindered the growth rate of POC, while it promoted the growth rate of MAOC as SOC accrued, which potentially obscured the distinct patterns of C accumulation triggered by afforestation. The variation partitioning suggests that, under afforestation, microbial traits had a higher contribution to both POC and MAOM variations compared with non‐afforested land. These results suggest that the robust buildup of microbial biomass due to increased plant C input following afforestation could contribute to soil C accumulation by promoting microbial necromass.

A Multi-Feature Fusion Approach for Dialect Identification using 1D CNN

The phonological variety of Kurdish, a language with several dialects, poses a distinct problem in automatically identifying dialects. This study examines and evaluates several sound criteria for identifying Kurdish dialects: Badini, Hawrami, and Sorani. We deployed a dataset including 6,000 samples and utilized a mix of 1D convolutional neural networks (CNN) and fully connected layers to conduct the identification job. Our study aimed to assess the efficacy of different sound characteristics in accurately identifying dialects. We employed the Mel-frequency Cepstral Coefficients (MFCC) and other features such as the Mel spectrogram, spectral contrast, and polynomial features to extract the sound characteristics. We conducted training and testing of our models utilizing both individual characteristics and a composite of all features. Our analysis revealed that the identification task achieved excellent accuracy rates, suggesting a promising potential for success. We achieved 95.75% accuracy using MFCC combined with a Mel spectrogram. The accuracy improved by including contrast in the MFCC feature extraction process to 91.42%. Similarly, using poly_features resulted in an accuracy of 90.83%. Remarkably, accuracy reached a maximum of 96.5% when all the attributes were combined.

A Novel Framework for Multimodal Brain Tumor Detection with Scarce Labels.

Brain tumor detection has advanced significantly with the development of deep learning technology. Although multimodal data, such as Magnetic Resonance Imaging (MRI) and Computed Tomography (CT), has potential advantages in diagnostics, most existing studies rely solely on a single modality. This is because common fusion methods may lead to the loss of critical information when attempting multimodal fusion. Therefore, effectively integrating multimodal data has become a significant challenge. Additionally, medical image analysis requires large amounts of annotated data, and labeling images is a resourceintensive task that demands experienced professionals to spend a considerable amount of time. To address these challenges, this paper introduces a new unsupervised learning framework named Double-SimCLR. This framework builds on the foundation of contrastive learning and features a dual-branch structure, enabling direct and simultaneous processing of MRI and CT images for multimodal feature fusion. Given the "weak feature" characteristics of CT images (e.g., low soft tissue contrast and low resolution), we incorporated adaptive weight masking technology to enhance CT feature extraction. Moreover, we introduced a multimodal attention mechanism, which ensures that the model focuses on salient information, thereby elevating the precision and robustness of brain tumor detection. Even without substantial labeled data, experimental results demonstrate that Double-SimCLR achieves 93.458% accuracy, 92.463% precision, and a 93.058% F1-score, outperforming state-of-the-art (SOTA) models by 2.871%, 2.643%, and 3.098%, respectively.

How Did 19th-Century Alphorns Sound? A Reconstruction Based on Written Accounts of Its Musical Timbre

This paper reconstructs the sound of 19th-century alphorns based on contemporary written descriptions, which allows for a better understanding of literature and compositions that quoted and imitated the alphorn throughout the 19th century. In the absence of sound recordings, historical documents and literary sources provide valuable insights into the timbre of these traditional Alpine instruments. The research examines descriptions from 19th-century texts, comparing them with modern understandings of musical timbre. By analyzing the language used to describe the alphorn’s sound, the study identifies recurring descriptors and contextualizes them within the broader acoustic environment, including the influence of natural sounds like waterfalls and echoes. Historical sources reveal a complex perception of the alphorn’s timbre, described in terms of its resemblance to muted trumpets and a blend of brass and woodwind qualities. Authors such as Hermann Alexander von Berlepsch and François-Joseph Fétis provided detailed accounts, noting contrasting characteristics like “rough”, “soft”, “sharp”, and “melodious”, which varied with the listener’s distance from the instrument. These descriptions highlight the alphorn’s unique sound profile, distinct from modern perceptions that emphasize a warmer, fuller timbre. The findings underscore the importance of considering ecological and psychoacoustic contexts in the study of historical musical instruments.

Ultra-dense Green InGaN/GaN Nanoscale Pixels with High Luminescence Stability and Uniformity for Near-Eye Displays.

Ultra-dense (>4,000 pixels per inch) and highly stable full-color III-nitride nanoscale pixels are crucial for near-eye display technologies like virtual and augmented-reality glasses. In this context, InGaN-based long wavelength green microscale light-emitting diodes face major bottlenecks, such as low efficiency and inadequate wavelength stability. These challenges are associated with the presence of both nonradiative surface defects and the strain induced quantum-confined Stark effect. Herein, we report nanoscale pixelation of green InGaN/GaN LEDs incorporating strain-engineered ultra-dense nanowire (NW) arrays, corresponding to ∼36,000 pixels per inch. The NW pixel arrays exhibit a stable peak wavelength emission at ∼500 nm for over 3 orders of magnitude of injection current densities (from ∼4 A/cm2 to ∼1 kA/cm2). The observed wavelength stability enhancement (a reduced blue-shift of just ∼4 nm) directly results from the suppressed built-in electric field achieved by strain relaxation of the axial multi quantum wells in the NWs. Finite difference time domain simulations show that emission of the pixel array is significantly increased owing to the enhanced spontaneous emission rate (characterized by a high Purcell factor of ≈2) of the ultra-dense NWs. We have demonstrated top-down NWs, where each NW (diameter ranging down to 200 nm) shows excellent uniformity and light output characteristics in direct contrast to bottom-up grown NW heterostructures. The results of this study establish a viable route for realizing nanoscale pixels with high luminescence stability and wafer-scale uniformity with high (>20%) indium composition InGaN/GaN LED heterostructures, for next-generation near-eye displays.

C2IENet: Multi-branch medical image fusion based on contrastive constraint features and information exchange

C2IENet: Multi-branch medical image fusion based on contrastive constraint features and information exchange

SAAMBE-MEM: a sequence-based method for predicting binding free energy change upon mutation in membrane protein-protein complexes.

Mutations in protein-protein interactions can affect the corresponding complexes, impacting function and potentially leading to disease. Given the abundance of membrane proteins, it is crucial to assess the impact of mutations on the binding affinity of these proteins. Although several methods exist to predict the binding free energy change due to mutations in protein-protein complexes, most require structural information of the protein complex and are primarily trained on the SKEMPI database, which is composed mainly of soluble proteins. A novel sequence-based method (SAAMBE-MEM) for predicting binding free energy changes (ΔΔG) in membrane protein-protein complexes due to mutations has been developed. This method utilized the MPAD database, which contains binding affinities for wild-type and mutant membrane protein complexes. A machine learning model was developed to predict ΔΔG by leveraging features such as amino acid indices and position-specific scoring matrices (PSSM). Through extensive dataset curation and feature extraction, SAAMBE-MEM was trained and validated using the XGBoost regression algorithm. The optimal feature set, including PSSM-related features, achieved a Pearson correlation coefficient of 0.64, outperforming existing methods trained on the SKEMPI database. Furthermore, it was demonstrated that SAAMBE-MEM performs much better when utilizing evolution-based features in contrast to physicochemical features. The method is accessible via a web server and standalone code at http://compbio.clemson.edu/SAAMBE-MEM/. The cleaned MPAD database is available at the website.

ECCT: Efficient Contrastive Clustering via Pseudo-Siamese Vision Transformer and Multi-view Augmentation

Image clustering aims to divide a set of unlabeled images into multiple clusters. Recently, clustering methods based on contrastive learning have attracted much attention due to their ability to learn discriminative feature representations. Nevertheless, existing clustering algorithms face challenges in capturing global information and preserving semantic continuity. Additionally, these methods often exhibit relatively singular feature distributions, limiting the full potential of contrastive learning in clustering. These problems can have a negative impact on the performance of image clustering. To address the above problems, we propose a deep clustering framework termed Efficient Contrastive Clustering via Pseudo-Siamese Vision Transformer and Multi-view Augmentation (ECCT). The core idea is to introduce Vision Transformer (ViT) to provide the global view, and improve it with Hilbert Patch Embedding (HPE) module to construct a new ViT branch. Finally, we fuse the features extracted from the two ViT branches to obtain both global view and semantic coherence. In addition, we employ multi-view random aggressive augmentation to broaden the feature distribution, enabling the model to learn more comprehensive and richer contrastive features. Our results on five datasets demonstrate that ECCT outperforms previous clustering methods. In particular, the ARI metric of ECCT on the STL-10 (ImageNet-Dogs) dataset is 0.852 (0.424), which is 10.3% (4.8%) higher than the best baseline.