With the increasing reliance on digital communication, email has become an essential tool for personal and professional correspondence. However, despite its numerous benefits, digital communication faces significant challenges, particularly the prevalence of spam emails. Effective spam email classification systems are crucial to mitigate these issues by automatically identifying and filtering out unwanted messages, enhancing the efficiency of email communication. We compare five traditional machine-learning and five deep-learning spam classifiers against a novel meta-learner, evaluating how different word embeddings, vectorization schemes, and model architectures affect performance on the Enron-Spam and TREC 2007 datasets. The primary aim is to show how the meta-learner's combined predictions stack up against individual ML and DL approaches. Our meta-learner outperforms all state-of-the-art models, achieving an accuracy of 0.9905 and an AUC score of 0.9991 on a hybrid dataset that combines Enron-Spam and TREC 2007. To the best of our knowledge, our model also surpasses the only other meta-learning-based spam detection model reported in recent literature, with higher accuracy, better generalization from a significantly larger dataset, and lower computational complexity. We also evaluated our meta-learner in a zero-shot setting on an unseen real-world dataset, achieving a spam sensitivity rate of 0.8970 and an AUC score of 0.7605. These results demonstrate that meta-learning can yield more robust, bias-resistant spam filters suited for real-world deployment. By combining complementary model strengths, the meta-learner also offers improved resilience against evolving spam tactics.
Tauopathies, including Alzheimer's disease and frontotemporal dementia with Parkinsonism linked to chromosome 17 (FTDP-17), are characterized by the aberrant aggregation of tau protein into neurofibrillary tangles. Despite extensive studies on tau aggregation, the mechanisms of tau misfolding and propagation remain incompletely understood. In this study, we utilize the SPAM (S320F/P301S) tau transgenic mouse model, which expresses 0N4R human tau with two FTDP-17 mutations, to investigate the biochemical properties and seeding potential of misfolded tau from these mice. Sarkosyl extraction and ultracentrifugation were employed to isolate detergent-insoluble tau aggregates (SPAM-tau) from aged SPAM mice. These aggregates were then tested for their prion-type seeding activity in an established HEK293T cell model comparing the induced aggregation of wild-type and mutant forms of human and murine tau. Our results show that SPAM-tau exhibits distinct and vigorous prion-like seeding properties, inducing the aggregation of human and murine tau homologues with the formation of amyloidogenic (Thioflavin S-positive) inclusions. Importantly, SPAM-tau aggregates can facilitate the prion-type misfolding of wild-type and mutant forms of human and mouse tau. We demonstrated that these induced tau aggregates are able to be further transmitted in passaging studies. Furthermore, SPAM-tau preferentially templated 4R tau isoforms, sharing strain-like seeding properties with insoluble tau derived from the brains of individuals with progressive supranuclear palsy (PSP-tau). In summary, these findings enhance our understanding of tau aggregation and propagation, suggesting that SPAM-tau may serve as a valuable tool for studying tauopathies and evaluating potential therapeutic strategies aimed at halting tau misfolding and propagation.
SMS spam detection remains a critical challenge in mobile communication security, particularly when addressing the inherent class imbalance present in real-world datasets, where spam messages constitute only 13–15% of total communications. This study presents a comprehensive framework integrating advanced word embeddings, deep learning architectures, and Generative Adversarial Networks (GANs) for synthetic data augmentation to enhance SMS spam classification performance. A systematic evaluation is conducted across six machine learning algorithms (Support Vector Machine (SVM), Logistic Regression (LR), K-Nearest Neighbors (KNN), Decision Tree (DT), Stochastic Gradient Descent (SGD), Random Forest (RF)) and two deep learning models (Long Short-Term Memory (LSTM), Bidirectional LSTM (Bi-LSTM)), combined with five embedding techniques (Term Frequency–Inverse Document Frequency (TF-IDF), Bag of Words (BoW), Word2Vec, GloVe, Bidirectional Encoder Representations from Transformers (BERT)), resulting in 120 experimental configurations tested both with and without data augmentation. A novel GAN-based approach is employed to generate synthetic word embeddings rather than raw text, preserving semantic coherence while addressing dataset imbalance more effectively than traditional oversampling methods (Synthetic Minority Over-sampling Technique (SMOTE), Adaptive Synthetic Sampling (ADASYN)). Experimental validation on both the monolingual UCI SMS Spam Collection and multilingual datasets demonstrates that the optimal BERT+Bi-LSTM+GAN configuration achieves exceptional performance, with F1-scores of 97.61% (monolingual) and 94.44% (multilingual), surpassing existing state-of-the-art approaches. The comprehensive evaluation framework, incorporating Matthews Correlation Coefficient (MCC) and Cohen’s Kappa (CK), provides robust assessment for imbalanced classification scenarios. Results reveal that contextual embeddings consistently outperform traditional frequency-based methods, with BERT achieving perfect precision of 100% in baseline configurations. The study establishes strategic deployment guidelines: BERT configurations for maximum accuracy scenarios, Word2Vec approaches for balanced performance–efficiency requirements, and traditional methods for resource-constrained environments. Cross-linguistic validation confirms the universality of the approach, demonstrating only a 3.25% performance degradation in multilingual contexts. This research advances both theoretical understanding of imbalanced text classification and practical implementation of robust SMS spam detection systems, providing a methodological foundation applicable to broader cybersecurity and natural language processing challenges.
Social networking platforms like Twitter, Instagram, Youtube, Facebook, Whatsapp have completely changed people's daily routine. Users of these social media networks have total freedom to upload anything that has political, commercial, or entertainment value. The data collected from these sources can be genuine or fake. There are no concerns or problems if the data published is true and relevant. The main difficulty arises while we deal with the spam data. So, this problem of spam data should be properly handled. In order to achieve a spam free environment, researchers have proposed numerous methods and algorithms for spam detection. Out of them few algorithms are implemented to detect the spam data in twitter.•We compare the outcomes in each scenario using various state-of-the-art word embedding techniques, such as Word2Vecv, GloVe, and FastText.•To account for the restrictions, two deep learning hybrid fusion classifier techniques-Text-based classifier and Combined classifier-are used in this work. These classifiers are built using deep learning techniques including GRU, LSTM, and CNN.•These methods will be evaluated using a range of measures, including F1-score, accuracy, recall, and precision. These actions could enhance the performance of the hybrid fusion approach.
Novel spam with rapidly evolving content faces a scarcity of labeled data in its early stages. Yet, current detection models rely heavily on large datasets and high-dimensional features, leading to poor generalization and opaque decisions when data is scarce. This opacity hinders error tracing and limits their use in early threat detection and response. The belief rule base (BRB), as an expert system, demonstrates effective learning under small-sample conditions, and its rule-based reasoning mechanism provides decision interpretability. However, high-dimensional features may cause combination explosion. To address these issues, a BRB spam detection model based on the Discriminative term frequency-inverse document frequency (TF-IDF) method (DTI-BRB) is proposed in this paper. By discriminating whether terms are more indicative of ham or spam, the Discriminative TF-IDF method converts raw text into low-dimensional features, thereby effectively resolving the combination explosion problem inherent in the traditional BRB model. Through two case studies under small-sample conditions, the effectiveness of the proposed model is validated. With only 200 samples, it achieves accuracies of 91.5% and 95.5% in the two cases, respectively, exhibiting excellent predictive performance and interpretability.
As email usage expands, spam has become a critical challenge, threatening network security and reducing communication efficiency. Conventional detection methods face persistent limitations: traditional machine learning models often struggle with high-dimensional sparse data, while deep learning requires substantial computational resources. This study introduces a Van der Waerden rank score feature attention-enhanced Support Vector Machine (VWR-Attn-SVM) to address these issues. The method applies Van der Waerden rank transformation to normalize text features, improving robustness against outliers and preserving ordinal relationships. An enhanced attention mechanism further optimizes feature selection through non-linear processing with regularization, highlighting the features most relevant to spam detection. Experiments on the UCI Spambase and Indonesian Spam datasets show that VWR-Attn-SVM outperforms traditional classifiers in accuracy, precision, recall, F1-score, and AUC. By combining high performance with reduced computational cost, the method provides an efficient and interpretable solution for spam classification, with potential extension to other text-based platforms such as messaging and social media.
Analysis of longitudinal data often relies on models which can be prone to statistical artifacts. We have previously shown that several published prospective associations can be explained by a combination of a general association between constructs, imperfect measurement reliability, and regression to the mean. Here, we formalize our analysis of this type of statistical artifact and introduce the Spurious Prospective Associations Model (SPAM). We show that the SPAM performs better than adjusted cross-lagged effects models to explain several observed prospective associations, including new examples involving loneliness and social anxiety and resilience and depressive symptoms, without assuming any true increasing or decreasing effects between constructs over time. Moreover, unlike the models we challenge, the SPAM is consistent with seemingly paradoxical findings indicating simultaneous increasing and decreasing effects between constructs. We conclude that the SPAM agrees well with observed data and is better supported than competing adjusted cross-lagged effects models in the cases investigated here.
This study aimed to prepare Pickering emulsions stabilized by the complexes of Spanish mackerel meat (SMM) and hyaluronic acid (HA) for the encapsulation of curcumin (Cur), and explore its application as a fat replacer in Spam. The SMM-HA complex was formed through hydrogen bonding. When the HA concentration was 0.75%, the three-phase contact angle of SMM-HA reached 88.2°, which was close to 90°, indicating its excellent emulsifying capability. Pickering emulsion prepared by SMM-HA (0.75%) with an oil phase fraction of 60% exhibited superior rheological properties, and a cross-linked network structure was formed between oil droplets. The emulsion encapsulation significantly improved the retention rate of Cur under various environmental conditions and demonstrated strong ABTS and DPPH free radical scavenging abilities. After in vitro simulated digestion, the Pickering emulsion stabilized by SMM-HA (0.75%) showed significantly higher free fatty acids (FFAs) release rate, bioaccessibility, and Cur release rate compared to the Pickering emulsion stabilized by SMM and free Cur in corn oil. Complete replacement of fat in Spam with the emulsion stabilized by SMM-HA (0.75%) significantly improved the product's textural properties, reduced cooking loss by 68.50%, decreased the pH from 4.93 to 4.62, and enhanced the water-holding capacity (WHC) by 8.50% compared to the non-replacement group. Therefore, the Pickering emulsion stabilized by SMM-HA provides a theoretical foundation for Cur protection and supports the development of low-fat healthy food products.
With the widespread adoption of internet technologies and email communication systems, the exponential growth in email usage has precipitated a corresponding surge in spam proliferation. These unsolicited messages not only consume users' valuable time through information overload but also pose significant cybersecurity threats through malware distribution and phishing schemes, thereby jeopardizing both digital security and user experience. This emerging challenge underscores the critical importance of developing effective spam detection mechanisms as a cornerstone of modern cybersecurity infrastructure. Through empirical analysis of machine learning (ML) performance on publicly available spam datasets, we established that algorithmic ensemble methods consistently outperform individual models in detection accuracy. We propose an optimized stacking ensemble framework that strategically combines predictions from four heterogeneous base models (NBC, k-NN, LR, XGBoost) through meta-learner integration. Our methodology incorporates grid search cross-validation with hyperparameter space optimization, enabling systematic identification of parameter configurations that maximize detection performance. The enhanced model was rigorously evaluated using comprehensive metrics including accuracy (99.79%), precision, recall, and F1-score, demonstrating statistically significant improvements over both baseline models and existing solutions documented in the literature.
[This corrects the article DOI: 10.1371/journal.pone.0307112.].
Academics, including researchers and scholars, might receive undesired/unsolicited emails, including spam. This volume might differ depending on whether they use a web-based or institutional email, since filters for each may differ. In the author's experience, most unsolicited emails have mainly been related to publishing, such as requests for submissions to lesser-known or academically suspect journals, and have become the norm. In addition, in the COVID-19 pandemic era (2020-2023), unsolicited emails related to the virus or the pandemic were received, as were some emails related to the Russo-Ukrainian war in 2022-2024. To gain an appreciation of the daily and monthly volumes of emails received by the author in 2018-2024, emails were stored in email folders over these 7 years on a monthly basis. A total of 130,941 unsolicited emails (sensu lato) were received in this 7-year period (14,514; 17,438; 15,668; 20,458; 19,845; 21,321; 21,697 in 2018, 2019, 2020, 2021, 2022, 2023, and 2024, respectively). The volume per month for each of these 7 years was 1613, 1938, 1741, 2273, 2205, 2369, and 2411 while the daily volume was 54, 64, 58, 75, 73, 79, and 80, respectively. Practical solutions are needed for academics to manage such unsustainable volumes of unsolicited emails. This brief assessment has limitations.
Real-world data suggest that the early use of highly active therapies (HAT) may reduce the risk of transition to secondary progressive MS (SPMS). However, current knowledge about predictive factors of outcomes needs to be improved. The primary objective of this study was to determine factors associated with the occurrence of SPMS in patients treated early after MS onset with an HAT. Retrospective, multicentric study based on the French MS database. Patients who initiated a HAT within 5 years after MS onset, EDSS ⩽4, and had a follow-up >5 years were included. The association of each covariate at baseline with time to the occurrence of SPMS was quantified by hazard ratios (HRs) in unadjusted and adjusted Cox proportional hazards models. Two thousand two hundred and thirty-seven patients were included in the analysis: mean age 31.6 years, female/male sex ratio 2.3, and median EDSS 2.0. The estimated probability of reaching SPMS, progression independent of relapse activity (PIRA) and progression independent of activity (PIA) at 10 years was 8%, 22%, and 11%, respectively. After adjustment, we found that female patients (HR 0.64, p = 0.036) had a lower risk of developing SPMS. Older age, EDSS >0 (HR 7.44, p < 0.001), and oral versus intravenous HAT (HR 1.97, p = 0.003) were significantly associated with an increased SPMS risk. Early PIRA and PIA predicted conversion to SPMS. Early HAT use resulted in a low risk of developing SPMS over 10 years. Introducing the HAT before any residual disability was associated with a lower risk of progression.
暂无摘要（点击查看详情）
暂无摘要（点击查看详情）
Periodontitis is a bacterial infectious disease. Photodynamic therapy (PDT) offers high selectivity, drug-resistance-free treatment, and immune regulation. The second-generation porphyrin photosensitizer Ce6 excels in reactive oxygen species (ROS) production. However, periodontitis pathogens' negative charge limits Ce6's interaction with them. This study prepared a modified cationic cyclodextrin (sPAM) and encapsulated Ce6 in an aqueous medium to create a nano-photodynamic system (Ce6@sPAM), which was characterized. In vitro evaluations assessed Ce6@sPAM's photodynamic performance, safety, antibacterial properties, and effects on immunoregulation. TEM images revealed Ce6@sPAM's irregular spherical shape, with a size of 236 nm by DLS and a Zeta potential of +16.4 mV. Ce6@sPAM exhibits a notably brief light half-life of merely 13 min, facilitating its swift in vivo clearance. SOSG and DCFH-DA fluorescence experiments showed Ce6@sPAM had stronger ROS generation (p < 0.05) and better bacterial penetration (p > 0.05) than Ce6. Co-incubation with Ce6@sPAM reversed bacterial surface potential from negative to positive. Bio-safety tests confirmed its excellent biocompatibility. In antibacterial tests, sPAM showed antibacterial properties, and Ce6@sPAM had a stronger effect than Ce6 under light (p < 0.001). Ce6@sPAM also exhibited high macrophage killing rates (> 90%) without specificity (p > 0.05) and can induce M1 macrophages to M2 polarization. Ce6-loaded modified cyclodextrin nanoparticles hold great promise for synergistic PDT in periodontitis treatment, especially in early stages for optimal immunomodulation.
Subacute post-traumatic ascending myelopathy (SPAM) is a rare but devastating complication of spinal cord injury (SCI). It is character-ized by progressive neurological deterioration extending several segments above the primary lesion within days to weeks after trauma. The underlying pathophysiology remains uncertain, and treatment strategies are not standardized. A 38-year-old man sustained trau-matic C6-7 spondylolisthesis with bilateral facet dislocation following a motorcycle accident. Initial magnetic resonance imaging (MRI) demonstrated cord contusion and edema extending from C5 to C7. After traction and reduction, the patient underwent anterior C6 corpectomy with placement of an expandable cage and C5-7 plating, followed by C5-6 total laminectomy and C4-7 posterior in-strumentation. Postoperatively, partial neurological recovery was observed. However, on postoperative day 10, the patient developed quadriparesis rapidly progressing to quadriplegia, accompanied by spinal shock and respiratory failure requiring mechanical ventilation. Imaging studies excluded hematoma and implant failure, although postoperative MRI was limited by metallic artifacts. Differential diagnoses, including pulmonary embolism, cardiac dysfunction, and sepsis, were ruled out. Based on the clinical progression and exclu-sion of alternative causes, a diagnosis of ascending myelopathy was established. Despite intensive supportive care, the patient died on the fourth day of mechanical ventilation. SPAM remains an unpredictable and fatal complication of SCI. Limitations in postoperative imaging, particularly metal-related artifacts, may hinder diagnosis, underscoring the importance of correlating clinical and radiological findings. Vigilant monitoring and continued reporting of cases are essential to improve recognition, refine diagnostic strategies, and guide management of this rare entity. Subakut posttravmatik asendan myelopati (SPAM), omurilik yaralanmalarının nadir fakat yıkıcı komplikasyonlarından biridir. Travmadan sonraki günler veya haftalar içinde, başlangıç lezyonunun birkaç segment üzerinde nörolojik kötüleşme ile kendini gösterir. Patofizyolojisi tam olarak bilinmemekte olup standart bir tedavi yaklaşımı yoktur. Otuz sekiz yaşındaki erkek hasta motosiklet kazası sonrası acil servise getirildi. Başlangıç radyolojik incelemelerde C6–7 düzeyinde travmatik spondilolistezis, bilateral faset dislokasyonu ve C5–7 düzeyinde medulla spinalis kontüzyonu ile ödem saptandı. Redüksiyon sonrası anterior C6 korpektomi, ekspandibl kafes ve C5–7 plak yerleştirilmesi yapıldı, ardından C5–6 total laminektomi ve C4–7 posterior enstrümantasyon uygulandı. Postoperatif dönemde kısmi nörolojik düzelme izlendi. Ancak 10. günde kuadriparezi hızla kuadriplejiye ilerledi; spinal şok bulguları ve solunum depresyonu gelişti. Kontrol görüntülemelerinde hematom veya implant malpozisyonu saptanmadı, ancak manyetik artefakt nedeniyle medulla intrensek patolojisi net değerlendirilemedi. Pulmoner emboli, kardiyak disfonksiyon ve sepsis gibi ayırıcı tanılar ekarte edildi. Klinik seyir asendan myelopati ile uyumlu değerlendirildi ve yoğun destek tedavisine rağmen hasta kaybedildi. SPAM, servikal omurga travmalarının nadir fakat ölümcül bir komplikasyonudur. Postoperatif dönemde metalik artefakt nedeniyle radyolojik tanı sınırlı olabilir. Bu nedenle klinik-radyolojik korelasyon, dikkatli nörolojik takip ve yeni vaka bildirimleri bu nadir sendromun daha iyi anlaşılması ve yönetimi açısından kritik öneme sahiptir.
The viscosity stability of polymer solution is one of the challenges in enhancing oil recovery, and zwitterionic copolymers present excellent viscosity stability and emulsification performance, enabling effective control of the oil/water interface mobility and enhancing oil recovery. Herein, a zwitterionic copolymer (P(AM/AMBS/MAPTAC)) containing sulfonic acid groups and quaternary amine groups was synthesized by segmentation initiation with AM, AMBS and MAPTAC as monomers. The chemical structure of P(AM/AMBS/MAPTAC) was confirmed by FTIR and 1H NMR. The Mw value of P(AM/AMBS/MAPTAC) was 9.91 × 106 g/mol, and the apparent viscosity of the 2000 mg/L solution was 24.92 mPa·s at 60 °C at a salinity of 5000 mg/L. P(AM/AMBS/MAPTAC) with sulfonic acid groups and quaternary amine groups exhibited outstanding salt tolerance and shear resistance. When the salinity was 10,000 mg/L and the shear rate was 300 s-1, the apparent viscosity for the P(AM/AMBS/MAPTAC) solution was 23.45 mPa·s and the viscosity reduction rate was 69.23% at 60 °C for 30 d. Moreover, P(AM/AMBS/MAPTAC) exhibited an improved emulsifying property and a greater oil-water interface thickness than HPAM and SPAM due to the synergistic effect of the sulfonic acid and quaternary amine groups in the P(AM/AMBS/MAPTAC) molecule. The polymer flooding and alkali-surfactant-polymer flooding formed by P(AM/AMBS/MAPTAC) had high chemical oil recovery, and the oil displacement efficiency of P(AM/AMBS/MAPTAC) was higher than that of HPAM and SPAM in the polymer flooding and alkali-surfactant-polymer flooding systems.
The DBLMSP gene family in Plasmodium falciparum encodes surface antigens involved in immune evasion and red blood cell invasion, yet its extensive polymorphism has long defied straightforward classification. While analyzing DBLMSP1 sequences from samples collected along the China-Myanmar border, we found that haplotypes could not be readily explained by standard population genetic models. Instead, comparative alignment and BLAST analysis revealed that DBLMSP1 and DBLMSP2 consist of discrete, recombinable sequence modules, flanked by conserved upstream and downstream regions. This led us to propose a modular framework that redefines allele structure as combinations of well-defined building blocks with consistent boundaries and positional constraints. Through global mining of DBLMSP sequences, we identified nine genotypes each for DBLMSP1 and DBLMSP2, with modules labeled sequentially (e.g., 1M2a, 1M3c, and 2M3b). Some modules were shared across paralogs, notably the identical sequence of DBLMSP1 module 1M3c and DBLMSP2 module 2M3a, suggesting historical inter-locus recombination. In the dominant genotype DBLMSP1-1, nucleotide diversity and Tajima's D peaked within variable modules, whereas conserved structural elements, including the receptor-binding cleft and SPAM domain, were under purifying selection. Patterns of long-range linkage disequilibrium aligned with module junctions, suggesting that modular structure may shape recombination patterns independently of selection. Modular recombination has been widely recognized in viral systems and multigene families such as var, but its relevance in DBLMSPs has been underappreciated. By applying this framework to P. falciparum DBLMSPs, we aim to provide a useful perspective for understanding their structural diversity and evolutionary dynamics, with implications for immunogen design and parasite surveillance. La recombinaison au niveau des modules est à l’origine du polymorphisme et de la conservation fonctionnelle de DBLMSP chez Plasmodium falciparum. La famille de gènes DBLMSP chez Plasmodium falciparum code des antigènes de surface impliqués dans l’échappement immunitaire et l’invasion des globules rouges. Cependant, son polymorphisme étendu a longtemps résisté à une classification simple. L’analyse des séquences DBLMSP1 d’échantillons prélevés le long de la frontière sino-birmane a révélé que les haplotypes ne pouvaient être facilement expliqués par les modèles de génétique des populations classiques. En revanche, l’alignement comparatif et l’analyse BLAST ont montré que DBLMSP1 et DBLMSP2 sont constitués de modules de séquences discrets et recombinables, flanqués de régions conservées en amont et en aval. Ceci nous a conduits à proposer un cadre modulaire qui redéfinit la structure allélique comme une combinaison d’éléments constitutifs bien définis, présentant des limites et des contraintes positionnelles cohérentes. Grâce à l’analyse globale des séquences DBLMSP, nous avons identifié neuf génotypes pour DBLMSP1 et neuf pour DBLMSP2, avec des modules numérotés séquentiellement (par exemple, 1M2a, 1M3c, 2M3b). Certains modules étaient partagés entre les paralogues, notamment la séquence identique des modules 1M3c de DBLMSP1 et 2M3a de DBLMSP2, suggérant une recombinaison inter-locus ancienne. Dans le génotype dominant DBLMSP1-1, la diversité nucléotidique et le D de Tajima atteignaient leur maximum au sein des modules variables, tandis que les éléments structuraux conservés, tels que le site de liaison au récepteur et le domaine SPAM, étaient soumis à une sélection purificatrice. Les profils de déséquilibre de liaison à longue distance s’alignaient avec les jonctions des modules, suggérant que la structure modulaire pourrait influencer les profils de recombinaison indépendamment de la sélection. La recombinaison modulaire est largement reconnue dans les systèmes viraux et les familles multigéniques comme var, mais son importance dans les DBLMSP a été sous-estimée. En appliquant ce cadre aux DBLMSP de P. falciparum, nous visons à fournir une perspective utile pour comprendre leur diversité structurelle et leur dynamique évolutive, avec des implications pour la conception d’immunogènes et la surveillance des parasites.
Recent studies have shown that 2D convolution and self-attention exhibit distinct spectral behaviors, and optimizing their spectral properties can enhance vision model performance. However, theoretical analyses remain limited in explaining why 2D convolution is more effective in high-pass filtering than self-attention and why larger kernels favor shape bias, akin to self-attention. In this paper, we employ graph spectral analysis to theoretically simulate and compare the frequency responses of 2D convolution and self-attention within a unified framework. Our results corroborate previous empirical findings and reveal that node connectivity, modulated by window size, is a key factor in shaping spectral functions. Leveraging this insight, we introduce a spectral-adaptive modulation (SPAM) mixer, which processes visual features in a spectral-adaptive manner using multi-scale convolutional kernels and a spectral re-scaling mechanism to refine spectral components. Based on SPAM, we develop SPANetV2 as a novel vision backbone. Extensive experiments demonstrate that SPANetV2 outperforms state-of-the-art models across multiple vision tasks, including ImageNet-1K classification, COCO object detection, and ADE20K semantic segmentation.
During oil production, crude oil is commonly recovered as water-in-oil (w/o) emulsions, which require the chemical destabilization of the emulsion for phase separation. Despite the extensive use of production chemicals, their combined effects on emulsion stability and produced water quality in extra-heavy oils are still poorly understood. In this work, the influence of a corrosion inhibitor (I cor), a scale inhibitor (I sca), and two enhanced oil recovery (EOR) polymers (SPAM and HPAM) on the performance of three commercial emulsion destabilizers (D 1, D 2, and D 3) was systematically investigated using synthetic emulsions prepared with asphaltic crude oil and synthetic brine (∼90,000 ppm). Bottle tests showed that, in the absence of EOR polymers, D1, D2, and D3 achieved water separation of ∼55%, 58%, and 78%, respectively, at high concentrations. The individual or combined addition of Icor and Isca did not induce phase separation (0.0% water separated) and did not significantly affect the demulsifiers' performance. In contrast, the presence of EOR polymers reduced demulsification efficiency, with HPAM showing the most pronounced effect, decreasing the maximum water separation to ∼39%, 45%, and 59% for D 1, D 2, and D 3, respectively. The method of additive addition did not significantly alter the separation efficiency. Interfacial tension measurements decreased from ∼27.6 mN/m (without additives) to ∼13.2-11.6 mN/m in the presence of demulsifiers, while smaller reductions were observed when EOR polymers were present, indicating competitive adsorption at the oil-water interface. Droplet size analysis revealed a direct correlation between increased water separation and larger droplet values, with D90 values reaching ∼1900 μm at high demulsifier concentrations. Although inhibitors did not affect separation efficiency, enhanced oil recovery (EOR) polymers significantly reduced the quality of the produced water, increasing the total oil and grease (TOG) content from ∼36 to ∼410 ppm (HPAM) and ∼600 ppm (SPAM). These results demonstrate that EOR polymers impair both the efficiency of emulsion destabilization and the quality of produced water in extra-heavy oil systems, while inhibitors showed no effect on the emulsion destabilization performance.