Search

Icosahedral gold clusters with high-symmetry geometry and magic electronic shells are potential candidates for cluster-assembling, while their assembling rules are still awaiting further investigation. In this work, we use the all-metal icosahedral M@Au12as a building block to assemble a series of bi-, tri-, tetra-, and penta-superatomic molecules with diverse superatomic bonding patterns via face-fusion, aiming to systemically explore the bonding rule of superatoms. Chemical bonding analyses indicate that these bi-, tri-, tetra-, and penta-superatomic molecules [M@Au12]n+/0(M = Re, W, Ta, Ti, Hf, Ir, and Pt) can be considered electronic analogues to Cl2, O2, N2, CO, O3, CO2, NCl3, and CF4molecules with single, double, triple, and multicenter bonds, respectively. In multi-superatomic molecules, the central superatom undergoes super S–P orbital hybridizations to form super σ bonds with each peripheral superatom, mimicking the bonding rules of molecules. Due to the large size of superatoms, superatomic molecules also present some distinct bonding characters in structure and relative energy compared to their analogues. This paper systemically investigated the superatomic bonding rules, giving references to further design and synthesis of superatom-assembled materials.

ABSTRACTBackgroundPolymyxins have been regarded as last-line treatment for multidrug-resistant gram-negative bacterial infections. Nonetheless, concerns regarding toxicity persist. This study aimed to explore and compare potential adverse events (AEs) between colistin and polymyxin B (PMB).MethodsPolymyxins-related AEs were retrieved from the U.S. Food and Drug Administration Adverse Event Reporting System between 2004 and 2022. Potential signals were estimated by the reporting odds ratio (ROR), and subgroup analyses were preformed to adjust for potential factors in AEs with significant disproportionality.ResultsAnalysis of 3,915 records involving 718 patients revealed a higher disproportionality of renal and urinary disorders (ROR 1.62, 95% CI 1.01–2.59) and acute kidney injury (ROR 1.75, 95% CI 1.07–2.87) with colistin treatment. Conversely, colistin exhibited a lower risk for neurotoxicity (ROR 0.47, 95% CI 0.30–0.73). Seven cases of skin hyperpigmentation were reported with PMB, whereas none were reported with colistin. Over 80% of cases involving polymyxin-related AEs occurred during the first two weeks of therapies, with a median onset time of 4.5 days.ConclusionsPatients received colistin displayed a higher potential risk of nephrotoxicity but a lower risk of neurotoxicity. Clinicians should be vigilant in monitoring the AEs of hyperpigmentation disorders induced by PMB.

Recent years have witnessed increasing attention being paid to AI-based quantitative investment. Compared to traditional low-frequency data (e.g., daily, weekly), high-frequency data (e.g., minute-level) is often underutilized for low-frequency stock trend prediction, leaving the vast potential for improvement. However, valuable and noisy information coexist in high-frequency data. The learning process of high-frequency factor extractors can easily be overwhelmed by noise, leading to overfitting. Moreover, common techniques used to prevent overfitting often result in poor performance on this task since they usually roughly restrict the model’s capacity, making it challenging to model complex trading signals in high-frequency data. When designing high-frequency factor extractors, we face a tough dilemma. A high-capacity model may easily overfit to noise, while a simple but robust model may not capture complex high-frequency patterns. To address these problems, we propose maintaining model capacity while preventing overfitting by constructing two components that balance information and noise through interactions between them. Specifically, we propose a novel learning framework called Digger-Guider to extract informative stock representations from noisy high-frequency data. We develop a high-capacity model called Digger to extract local and detailed features from the high-frequency data, and we design a robust model called Guider to capture global tendency features and help the Digger overcome the noise. The Digger and Guider enhance each other through mutual distillation during training, serving as data-driven regularizations that work well on this task. Extensive experiments on real-world datasets demonstrate that our framework can produce powerful high-frequency stock factors that significantly improve stock trend prediction performance and our understanding of the finance market.

Self-supervised Learning (SSL) including the mainstream contrastive learning has achieved great success in learning visual representations without data annotations. However, most methods mainly focus on the instance level information (i.e., the different augmented images of the same instance should have the same feature or cluster into the same class), but there is a lack of attention on the relationships between different instances. In this paper, we introduce a novel SSL paradigm, which we term as relational self-supervised learning (ReSSL) framework that learns representations by modeling the relationship between different instances. Specifically, our proposed method employs sharpened distribution of pairwise similarities among different instances as relation metric, which is thus utilized to match the feature embeddings of different augmentations. To boost the performance, we argue that weak augmentations matter to represent a more reliable relation, and leverage momentum strategy for practical efficiency. The designed asymmetric predictor head and an InfoNCE warm-up strategy enhance the robustness to hyper-parameters and benefit the resulting performance. Experimental results show that our proposed ReSSL substantially outperforms the state-of-the-art methods across different network architectures, including various lightweight networks (e.g., EfficientNet and MobileNet).

In recent years, research on ultra-short-term wind power forecasting models has gradually reached a bottleneck. Despite introducing numerous complex algorithms for wind power prediction, effective combinations have not been adequately explored. The useful information within wind power data has not been fully mined, and the selection of hyperparameters has not achieved an optimal combination, resulting in only a marginal improvement in forecasting accuracy. A novel wind power hybrid forecasting model, combining the modified honey badger algorithm, variational mode decomposition, and long short-term memory network, is proposed in this research. Taking the cosine similarity coefficient combined with root mean square error as the objective function, the modified honey badger algorithm is used to optimize the main parameters of the variational mode decomposition method, which improves the wind power data decomposition and denoising ability of variational mode decomposition and enables the deep mining of effective information in wind power data. Meanwhile, the modified honey badger algorithm is used to optimize the hyperparameters of long short-term memory, which solves the problem of difficult configuration of neural network parameters. Two different datasets with two different time intervals from two distinct Chinese wind farms were selected to validate the effectiveness of the proposed model. The experimental results indicate that the proposed model achieves superior forecasting performance metrics compared to other models across simulation experiments with various original datasets and prediction steps. The proposed model achieves high precision and good stability through the effective combination of multiple algorithms.

The maintenance of hematopoietic stem cells (HSCs) is a complex process involving numerous cell-extrinsic and -intrinsic regulators. The first member of the cyclin-dependent kinase family of inhibitors to be identified, p21, has been reported to perform a wide range of critical biological functions, including cell cycle regulation, transcription, differentiation, and so on. Given the previous inconsistent results regarding the functions of p21 in HSCs in a p21-knockout mouse model, we employed p21-tdTomato (tdT) mice to further elucidate its role in HSCs during homeostasis. The results showed that p21-tdT+HSCs exhibited increased self-renewal capacity compared to p21-tdT−HSCs. Zbtb18, a transcriptional repressor, was upregulated in p21-tdT+HSCs, and its knockdown significantly impaired the reconstitution capability of HSCs. Furthermore, p21 interacted with ZBTB18 to co-repress the expression of cKitin HSCs and thus regulated the self-renewal of HSCs. Our data provide novel insights into the physiological role and mechanisms of p21 in HSCs during homeostasis independent of its conventional role as a cell cycle inhibitor.

ABSTRACTThis study aimed to holistically investigate fraction knowledge by considering pupils’ fraction performance on various tasks (i.e. mapping, equivalence, comparison, and arithmetic) and in relation to whole number arithmetic, fraction misconceptions, mathematics anxiety, and confidence in responses. Northern Irish pupils (N = 123; 77 girls; Mage = 11.1 years) demonstrated a strong understanding of fraction magnitudes with few pupils showcasing misconceptions across magnitude tasks. In contrast, fraction arithmetic knowledge, which was uniquely predicted by whole number arithmetic and fraction comparison, was still developing with many pupils incorrectly applying whole number strategies to arithmetic problems. Mathematics anxiety was low, but negatively correlated with performance, and confidence reports were more closely aligned with performance on whole number arithmetic compared to fraction tasks. Overall, researchers and educators need to consider not just fraction performance, but also foundational skills, the presence of misconceptions, and pupils’ feelings and beliefs to better understand fraction learning.

Lung cancer remains the leading cause of death among cancer patients, and the five-year survival rate is less than 25%. However, Methyl-CpG Binding Domain(MBD)4polymorphism rs140693 predicts the prognosis of lung cancer patients still needs further verification. Primary lung cancer patients (n= 839) were collected from two hospitals, genomic DNA was extracted from blood, and genotyping was performed using SNPcan technology. Kaplan-Meier technique and multivariate Cox proportional hazards model were used to analyze the prognosis association between MBD4and clinical characteristics. Significantly conferred a poorer prognosis was associated with the CT genotype (CT vs. CC; adjusted hazard ratio [HR] = 1.21, 95% CI: 1.03–1.43, p= 0.023) and dominant CT + TT genotype (CT + TT vs. CC; HR = 1.19, 95% CI: 1.02–1.39, p= 0.029) of MBD4polymorphism rs140693 for all lung cancer patients, compared with the CC genotype. Stratified analysis showed that polymorphism rs140693 CT and dominant CT + TT genotype conferred a significantly poorer prognosis in female and lung adenocarcinoma (ADC) cancer patients, compared with the CC genotype. Non-small cell lung cancer (NSCLC) patients with the CT genotype had a poorer prognosis than those with the CC genotype. Additionally, the allele T of small cell lung cancer (SCLC) patients compared with the allele C was associated with a poor prognosis, and the CT and recessive TT genotype of SCLC patients conferred a significantly poor prognosis. The MBD4polymorphism rs140693 is a significant prognostic genetic marker for predicting the prognosis of lung cancer patients.

Insufficient drug accumulation in tumors severely limits the antitumor efficiency of hyaluronic acid (HA) nanomedicine in solid tumors due to superficial penetration depth, low cell uptake, and nonspecific drug release. Hence, we constructed a dual NO prodrug (alkynyl-JSK) and doxorubicin prodrug (cis-DOX)-conjugated HA nanoparticle (HA-DOX-JSK NPs), which achieved cascade-boosted drug delivery efficiency based on a relay strategy of NO-mediated deep tumor penetration─HA target CD44 tumor cell uptake─tumor microenvironment (TME)-responsive drug release. The nanoparticle demonstrated sustained and locoregionally GSH/GST-triggered NO release and GSH/pH-responsive DOX release in the tumor. The released NO first mediated collagen degradation, causing deep tumor penetration of nanoparticles in the dense extracellular matrix. Immediately, HA was relayed to enhance CD44-targeted tumor cell uptake, and then, the nanoparticles were finally triggered by specific TME to release DOX and NO in the deep tumor. Relying on the relayed delivery strategy, a significant improvement of DOX accumulation in tumors was realized. Moreover, NO depleted GSH-induced intracellular reactive oxygen species, enhancing DOX chemotherapy. Based on this strategy, the tumor inhibition rate in breast cancer was up to 87.8% in vivo. The relay drug-delivery HA system would greatly cascade-boost drug accumulation in deep tumors for efficient solid tumor therapy.

Users have become accustomed to storing data on the cloud using ciphertext policy attribute-based encryption (CP-ABE) for fine-grained access control. However, this encryption method does not consider the ability of malicious users to launch thousands of file download requests when launching an economic denial of sustainability attack (EDoS), which may be more expensive for data owners. Existing solutions typically use a cloud server to verify the download permissions of the data users. However, cloud servers are not completely trusted and cloud server providers and colluding data users can still launch an EDoS attack. With our scheme, using CP-ABE, a blockchain is introduced for verifying the download permission of data users. In addition, we propose a new mechanism to solve the problem of malicious user revocations under EDoS attacks by updating the ciphertext and symmetric encryption technology. A formal security proof has demonstrated that the proposed scheme is suitable for plaintext attack security. Theoretical and experimental analyses show that our scheme performs more efficiently than previous methods.

To meet the requirements of long-range electric vehicles and aviation, the high-mass-loading electrode with high areal capacity is a promising solution to realize ultrahigh-energy lithium-metal batteries (LMBs). However, enabling the operation of high mass loading with a long cycling life is still a challenge without in-depth investigation. Herein, we figured out that the polarization appearing in the cycled lithium-metal anodes (LMAs) is responsible for the poor cycling of LMBs with high mass loading. Moreover, the origin of fast degradation of LMAs is affected by mass loading through the Li plating process, which is decided by the Li plating morphology. Hence, manipulating the mass loading can directly promote lithium reversibility and further mitigate cell polarization in LMBs, endowing high-mass-loading LMBs with excellent cycling stability. Consequently, we achieved an ultrahigh energy density (605 W h kg–1) of a 10.1 A h pouch cell with an excellent retention of 91.7% capacity and 86% energy after 50 cycles. The feasible strategy points out a promising approach for designing high-energy-density LMBs in the future.

Lithium metal batteries with ultrahigh-specific energy densities can be realized by combining lithium metal anodes with high-voltage and high-capacity lithium-rich manganese oxide cathodes. Their practical application is largely hindered by the notorious dendrite growth of lithium metal anode, fast structure degradation of cathode, and insufficient electrode–electrolyte interphase kinetics that arise from severe side reactions between conventional electrolytes and very active lithium metals. Here, the use of a high-concentration fluorinated carbonate electrolyte consisted of methyl 2,2,2-trifluoroethyl carbonate as solvents and lithium bis(fluorosulfonyl)imide as salts can significantly improve the electrochemical stability of the electrolytes and prohibit the structural failure of lithium metal. In Li||LR-NCM114 pouch cells, the cycling performance of the pouch cell was greatly improved (92.3% capacity retention over 100 cycles). Furthermore, a pouch cell with a capacity of 6.2 Ah is fabricated and achieves an ultrahigh energy density over 500 Wh kg–1. Our work has explored the application of high-concentration fluorinated carbonate electrolytes in high-voltage lithium metal batteries with a high energy density.

Superoxide anion (O2·–) and peroxynitrite (ONOO–), two important oxidants under oxidative stress, coexist in complex cell and organism systems, playing crucial roles in various physiological and pathological processes, particularly in neurodegenerative diseases. Despite the absence of robust molecular tools capable of simultaneously visualizing O2·–and ONOO–in biosystems, the relationship between these two species remains understudied. Herein, we present sequentially activated fluorescent probe, DHX-SP, which exhibits exceptional selectivity and sensitivity toward O2·–and ONOO–. This probe enables precise imaging of these species in living PC12 cells under oxidative stress conditions using distinct fluorescence signal combinations. Furthermore, the probe DHX-SPhas the ability to visualize changes in O2·–and ONOO–levels during ferroptosis of PC12 cells and in the Parkinson’s disease model. These findings establish a connection between the crosstalk of the phosphorus group of O2·–and ONOO–in PC12 cells under oxidative stress.

The research on fluorescent rotors for viscosity has attracted extensive interest to better comprehend the close relationships of microviscosity variations with related diseases. Although scientists have made great efforts, fluorescent probes for cellular viscosity with both aggregation-induced emissions (AIEs) and large Stokes shifts to improve sensing properties have rarely been reported. Herein, we first report four new meso-C═N-substituted BODIPY-based rotors with large Stokes shifts, investigate their viscosity/AIE characteristics, and perform cellular imaging of the viscosity in subcellular organelles. Interestingly, the meso-C═N-phenyl group-substituted probe 6showed an obvious 594 nm fluorescence enhancement in glycerol and a moderate 650 nm red AIE emission in water. Further, on attaching CF3to the phenyl group, a similar phenomenon was observed for 7with red-shifted emissions, attributed to the introduction of a phenyl group, which plays a key role in the red AIE emissions and large Stokes shifts. Comparatively, for phenyl-group-free probes, both the meso-C═N-trifluoroethyl group and thiazole-substituted probes (8and 9) exhibited good viscosity-responsive properties, while no AIE was observed due to the absence of phenyl groups. For cellular experiments, 6and 9showed good lysosomal and mitochondrial targeting properties, respectively, and were further successfully used for imaging viscosity through the preincubation of monensin and lipopolysaccharide (LPS), indicating that C═N polar groups potentially work as rotatable moieties and organelle-targeting groups, and the targeting difference might be ascribed to increased charges of thiazole. Therefore, in this study, we investigated the structural relationships of four meso-C═N BODIPY-based rotors with respect to their viscosity/AIE characteristics, subcellular-targeting ability, and cellular imaging for viscosity, potentially serving as AIE fluorescent probes with large Stokes shifts for subcellular viscosity imaging.

The use of radiation therapy to treat pelvic and abdominal cancers can lead to the development of either acute or chronic radiation enteropathy. Radiation-induced chronic colonic fibrosis is a common gastrointestinal disorder resulting from the above radiation therapy. In this study, we establish the efficacy of inulin supplements in safeguarding against colonic fibrosis caused by irradiation therapy. Studies have demonstrated that inulin supplements enhance the proliferation of bacteria responsible to produce short-chain fatty acids (SCFAs) and elevate the levels of SCFAs in feces. In a mouse model of chronic radiation enteropathy, the transplantation of gut microbiota and its metabolites from feces of inulin-treated mice were found to reduce colonic fibrosis in validation experiments. Administering inulin-derived metabolites from gut microbiota led to a notable decrease in the expression of genes linked to fibrosis and collagen production in mouse embryonic fibroblast cell line NIH/3T3. In the cell line, inulin-derived metabolites also suppressed the expression of genes linked to the extracellular matrix synthesis pathway. The results indicate a novel and practical approach to safeguarding against chronic radiation-induced colonic fibrosis.

As the capacity of lithium-ion batteries gradually reaches its limit, the high-capacity characteristics of lithium metal batteries (LMBs) make them one of the most promising electrochemical energy storage devices currently. However, uncontrolled lithium dendrite growth can cause poor cell performance and severe safety issues, seriously slowing down the commercialization of LMBs. Transition metals (TMs) are lithiophilic materials that effectively stabilize lithium metal anode (LMA) and inhibit dendrite growth. Herein, the growth of lithium dendrites and the performance of TMs in stabilizing lithium metals are comprehensively discussed in this review. First, we elucidated the formation mechanisms of lithium dendrite, and we put forward several strategies to suppress lithium dendrite formation based on the understanding of these mechanisms. Then, the application of TMs and their compounds to stabilize LMA and their corresponding mechanisms are comprehensively clarified. Last, further research perspectives of TMs in stabilizing LMA are emphasized in detail.

The genetic and genomic basis of sex differences in blood pressure (BP) traits remain unstudied at scale. Here, we conducted sex-stratified and combined-sex genome-wide association studies of BP traits using the UK Biobank resource, identifying 1,346 previously reported and 29 new BP trait-associated loci. Among associated loci, 412 were female-specific (Pfemale≤ 5 × 10−8; Pmale> 5 × 10−8) and 142 were male-specific (Pmale≤ 5 × 10−8; Pfemale> 5 × 10−8); these sex-specific loci were enriched for hormone-related transcription factors, in particular, estrogen receptor 1. Analyses of gene-by-sex interactions and sexually dimorphic effects identified four genomic regions, showing female-specific associations with diastolic BP or pulse pressure, including the chromosome 13q34-COL4A1/COL4A2locus. Notably, female-specific pulse pressure-associated loci exhibited enriched acetylated histone H3 Lys27 modifications in arterial tissues and a female-specific association with fibromuscular dysplasia, a female-biased vascular disease; colocalization signals included Chr13q34: COL4A1/COL4A2, Chr9p21: CDKN2B-AS1and Chr4q32.1: MAP9regions. Sex-specific and sex-biased polygenic associations of BP traits were associated with multiple cardiovascular traits. These findings suggest potentially clinically significant and BP sex-specific pleiotropic effects on cardiovascular diseases.

Vanilla aerial trackers exhibit sensitivity to low-light conditions (e.g., nighttime aerial tracking scenario). To mitigate this, existing methods incorporate the light enhancement method as a preprocessing for aerial tracking. Despite the advancements, these approaches are restricted to the disparity in task objectives between the enhancer and tracker. Motivated by the observation that feature channels exhibit varying sensitivity to illumination, we propose to decouple the feature representation into two distinct parts: 1) illumination-invariant feature embedding and 2) illumination-sensitive feature embedding. The former, realized by the illumination invariant embedding (IIE) module, enhances features that remain invariant to illumination changes. Meanwhile, the latter, facilitated by the illumination sensitive embedding (ISE) module, aims to mitigate the negative impact of illumination-sensitive features on tracking performance. Building upon this decoupling strategy, we introduce NiDR, a simple yet effective nighttime aerial tracker. The proposed NiDR exhibits strong performance on three nighttime aerial tracking benchmarks (i.e., UAVDark135, NAT2021, and DarkTrack2021). Notably, it outperforms previous competitors by large margins, e.g., 3.1 points on the UAVDark135 and 2.0 points on the Darktrack2021 in terms of precision for nighttime scenarios.

The grab bucket plays an important role in the construction process of dredging, and analyzing the excavation characteristics of the grab is helpful for the mechanical design of the grab bucket. In this investigation, the digging device and the digging mudstone are detailed as a system dynamic process, and a theoretical model for calculating digging resistance is established based on the Rankine theory. Furthermore, the particle shear test is performed to analyze the influence of changes in sand physical parameters on its shear characteristics. The DEM-MBD coupling simulation is adopted to calculate the particle distribution and the digging trajectory, and the influence of digging depth and particle cohesion strength is analyzed on the digging resistance. Moreover, an excavator digging experimental platform is established to validate the feasibility of the experimental results. The results show that the 10% moisture content is the inflection point of the sand particles physical and mechanical properties, and its shear strength decreases with the augment of moisture content; the variation law of the simulation results is consistent with the theoretical calculation, and the digging resistance reduction is proportional to the cohesion strength when the particles are completely sheared; the same key point parameter error of the digging curve obtained by the test and the simulation are consistent.

Association networks are widely applied for the prediction of bacterial interactions in studies of human gut microbiomes. However, the experimental validation of the predicted interactions is challenging due to the complexity of gut microbiomes and the limited number of cultivated bacteria. In this study, we addressed this challenge by integrating in vitrotime series network (TSN) associations and co-cultivation of TSN taxon pairs. Fecal samples were collected and used for cultivation and enrichment of gut microbiome on YCFA agar plates for 13 days. Enriched cells were harvested for DNA extraction and metagenomic sequencing. A total of 198 metagenome-assembled genomes (MAGs) were recovered. Temporal dynamics of bacteria growing on the YCFA agar were used to infer microbial association networks. To experimentally validate the interactions of taxon pairs in networks, we selected 24 and 19 bacterial strains from this study and from the previously established human gut microbial biobank, respectively, for pairwise co-cultures. The co-culture experiments revealed that most of the interactions between taxa in networks were identified as neutralism (51.67%), followed by commensalism (21.67%), amensalism (18.33%), competition (5%) and exploitation (3.33%). Genome-centric analysis further revealed that the commensal gut bacteria (helpers and beneficiaries) might interact with each other via the exchanges of amino acids with high biosynthetic costs, short-chain fatty acids, and/or vitamins. We also validated 12 beneficiaries by adding 16 additives into the basic YCFA medium and found that the growth of 66.7% of these strains was significantly promoted. This approach provides new insights into the gut microbiome complexity and microbial interactions in association networks. Our work highlights that the positive relationships in gut microbial communities tend to be overestimated, and that amino acids, short-chain fatty acids, and vitamins are contributed to the positive relationships.

Thermal infrared (TIR) object detection plays a crucial role in diverse around-the-clock applications, such as search-and-rescue operations and wildlife protection. Achieving rapid and robust detection of small objects from an aerial perspective is particularly significant in these scenarios. However, the task is compounded by two interrelated challenges, rendering it even more tricky. For one, small objects only occupy a few pixels and contain limited information. For another, TIR sensors are typically of low resolution (LR) due to inherent challenges associated with the imaging mechanism of the TIR spectrum. In contrast, high-resolution (HR) RGB sensors are readily available due to their cost-effectiveness and widespread application. Recognizing the importance of HR information, especially in the context of small-object detection, we propose a cross-modality HR knowledge distillation framework (CMHRD), which leverages knowledge from the HR-RGB modality and provides a novel strategy for TIR small-object detection. The proposed framework introduces three key components: a super-resolution generative (SRG) distillation loss for cross-modal HR representation learning, a cross-modality affinity (CMA) distillation loss to extract scene-level cross-modality information, and a response distillation loss aimed at mimicking the HR prediction. To facilitate research on small-object detection with HR-RGB and LR-TIR data, we have curated and annotated two datasets, namely NOAA-Seal and VTUAV-det-small. Experimental results on the NOAA-Seal demonstrate that CMHRD yields significant improvements, achieving a remarkable 6.39 mAP50 increase over a strong baseline without introducing additional computational cost during inference. Experiments on the single-category dataset VTUAV-det-small and the multicategory dataset RTDOD also show consistent improvements brought by CMHRD. The project is available at https://github.com/NNNNerd/CMHRD.

Cross-view geo-localization (CVGL) involves determining the geographical location of a query image by matching it with a corresponding GPS-tagged reference image. Current state-of-the-art methods predominantly rely on training models with labeled paired images, incurring substantial annotation costs and training burdens. In this study, we investigate the adaptation of frozen models for CVGL without requiring ground-truth pair labels. We observe that training on unlabeled cross-view images presents significant challenges, including establishing relationships within unlabeled data and reconciling view discrepancies between uncertain queries and references. To address these challenges, we propose a self-supervised learning framework to train a learnable adapter for a frozen foundation model (FM). This adapter is designed to map feature distributions from diverse views into a uniform space using unlabeled data exclusively. To establish relationships within unlabeled data, we introduce an expectation-maximization (EM)-based pseudolabeling module, which iteratively estimates matching between cross-view features and optimizes the adapter. To maintain the robustness of the FM’s representation, we incorporate an information consistency module with a reconstruction loss, ensuring that adapted features retain strong discriminative ability across views. Experimental results demonstrate that our proposed method achieves significant improvements over vanilla FMs and competitive accuracy compared to supervised methods while necessitating fewer training parameters and relying solely on unlabeled data. Evaluation of our adaptation for task-specific models further highlights its broad applicability. Particularly, on the University-1652 dataset, our method outperforms the FM baseline by a substantial margin, achieving about 39 points improvement in Recall@1 and more than 34 points increase in average precision (AP). The project is available at https://collebt.github.io/EM-CVGL.