With the development of artificial intelligence (AI), cloud environments are becoming increasingly important. However, cloud environment networks are at risk of various network attacks. Therefore, it is crucial to detect abnormal traffic in cloud environment networks. With the continuous development of network technology, the diversity of cloud environment network traffic continues to increase (intra-class diversity), and the boundary between malicious and benign behaviors becomes more blurred (inter-class similarity), leading to false detection. At the same time, most game theory defense deception methods for cloud environment networks assume that the attacker and defender maintain consistent views under uncertainty. In fact, the attacker and defender have different views on the same game. To address the above issues, we propose an intrusion detection and dynamic defense mechanism for cloud environments. To address the challenges brought by intra-class diversity and inter-class similarity, we propose an intrusion detection system (IDS) based on contrastive learning, which can make correct decisions when classifying samples of different categories. To identify traffic more accurately, this paper proposes an improved lightweight ResNet-34 model (IResNet34). To address the challenge that the attacker and defender have different views on the same game, we propose a hypergame model involving multiple attackers and defenders. The attacker cannot obtain complete game information through defensive deception technology, resulting in attack failure. In addition, we propose an adaptive defense strategy selection method based on machine learning, which automatically selects the best defense strategy based on the game record. The output of dynamic defense will be fed back to the intrusion detection module to reduce the false alarm rate. Finally, experiments verified that the method based on contrastive learning proposed in this paper can achieve high detection accuracy in the real world and benchmark datasets, and the dynamic defense method can effectively reduce the false positive rate (FPR) of IDS.

Microservice systems generate fragmented and misaligned multi-modal data, including metrics, logs, and traces, posing a major challenge for AIOps, which requires coherent system-state representations for downstream analytics. Existing approaches typically model spatio-temporal relations at a coarse granularity or are tailored to specific tasks, limiting their generalizability across different O&M scenarios. To address these limitations, we propose M3RL, a unified pre-training and fine-tuning framework for learning robust and transferable system-state representations. At its core, the spatio-temporal network (STNet) performs fine-grained semantic fusion of multi-modal data along both temporal and spatial dimensions and models evolving communication-driven dependencies among services. We further pre-train STNet with three complementary objectives: denoising reconstruction for robustness, future prediction for temporal evolution modeling, and causal contrastive learning for node-level discriminability. Extensive experiments on two public benchmark datasets demonstrate that M3RL consistently outperforms recent state-of-the-art baselines across key downstream tasks, reducing state prediction MAE by 1.9%, increasing anomaly detection F1 by 4.7%, and improving root cause localization HR@1 by 4.8% on average. These results show that M3RL provides a robust and transferable representation backbone for multiple AIOps tasks.

The rapid development of Artificial Intelligence (AI) has increased the demand for GPU resources, leading to a surge in deep learning tasks on heterogeneous GPUs. As a result, GPU resources in production clusters are often limited. It is essential to execute deep learning tasks efficiently within these resource and time constraints. Reducing task completion time (TCT) and improving GPU utilization are both critical. However, these goals often conflict, and existing solutions remain basic, making joint optimization challenging. We propose BFE, a method that integrates fragmentation awareness and efficiency orientation. BFE improves GPU utilization and reduces TCT by utilizing a cost model that combines average TCT and GPU fragmentation. We introduce a new fragmentation metric that considers both task size and GPU heterogeneity. For heterogeneous environments, we present a mathematical formulation that further enhances GPU efficiency and task scheduling. We also present a sliding window-based strategy to support the online scheduling scenario. We then design a particle swarm optimization-based scheduling algorithm to minimize the combined cost. Experiments on Alibaba's production traces demonstrate that the proposed BFE improves the GPU utilization by up to 51.73% and reduces the average TCT by up to 55.16%, outperforming existing methods.

The consumption of Chinese quince is hindered by the lack of suitable processing method. Vacuum-frying technology, as an efficient processing method, can help address this issue. However, the effect of vacuum frying on the structure and activity of bioactive components in Chinese quince, such as proanthocyanidins, is currently unknown. In this study, Chinese quince was vacuum fried at 70 degrees C-100 degrees C. Proanthocyanidins were extracted, characterized, and analyzed for their bioactivity. Results revealed that increasing the vacuum-frying temperature decreased the total proanthocyanidin content. Vacuum frying reduced samples' degrees of polymerization. Proanthocyanidin extracts vacuum fried at 70 degrees C exhibited the greatest antioxidant capacity. Proanthocyanidin extracts vacuum fried at 80 degrees C demonstrated pronounced hypoglycemic potential. These results show that vacuum frying can reduce the degree of polymerization of Chinese quince proanthocyanidins while enhancing their antioxidant activity and retaining their hypoglycemic capacity. This study offers valuable insights for the development of functional foods using Chinese quince.

The stable operation of edge networks relies heavily on the robustness of the underlying infrastructure, whose operational status can be monitored in real time via edge sensors. With the growing number of sensors and increasing monitoring frequency, the dimensionality and volume of time series data have surged dramatically, leading to the problem of catastrophic forgetting. Moreover, due to limitations in infrastructure and environmental factors, edge networks are more prone to data missing issues, posing significant challenges to data processing and analysis. To address these challenges, we propose a time series anomaly detection model tailored for edge network infrastructure. Specifically, to handle missing data, a multi-scale graph-based node imputation method is introduced to recover the missing values in the sequences. In the spatial dimension, a replay-based incremental learning strategy is designed, which leverages graph compression to capture critical features in the sequences, enabling the model to learn new features while alleviating the forgetting problem dynamically. In the temporal dimension, we adopt a single-layer WaveNet architecture that utilizes dilated convolutions and depthwise separable convolutions to efficiently extract features from multivariate time series, thereby reducing computational overhead. Extensive experiments conducted on the SMD and SWaT datasets demonstrate that the proposed model, i.e., ADEI, achieves superior performance.

The goal of time series anomaly detection in Cyber-Physical Systems (CPS) is to maintain the reliability and security of environmental sensors. Due to the lack of labeled data in real-world CPS scenarios, most existing anomaly detection methods utilize unsupervised learning to capture the normal behavior of unlabeled time series. However, constructing reliable anomaly detection models for CPS is challenging, as they must operate accurately in the context of complex system dynamics and uncertain sensor noise levels. To address this issue, we propose a robust attention-based contrastive representation learning method designed to capture discriminative and robust representations, specifically tailored to solve challenges posed by noisy and dynamic sensor data in CPS. In particular, we introduce the graph convolutional network (GCN) to learn embeddings for all variables, effectively capturing the time-varying correlations between variables in MTS and enhancing the representations' robustness. Furthermore, we leverage the feature graphs of variables constructed by GCN to perform multi-scale attention-based contrastive representation learning. Finally, through minimizing the discrepancies between original and GCN-enhanced samples across multiple attention scales, our model learns a robust and discriminative representation.

The exponential growth of smart devices and network scale has led to a rapid increase in network traffic, posing severe challenges to network resource utilisation and transmission efficiency. Intent-based networking (IBN) provides a high-level, automated method for network management. It dramatically simplifies network operations and enhances network flexibility and manageability. However, existing studies mainly focus on applying IBN in certain stages of network management without fully leveraging IBN’s network awareness and automated deployment features to comprehensively optimise network management and traffic forwarding. We propose an architecture for optimising network traffic based on user intents, i.e., UINT, which aims to simplify network management and optimise network traffic forwarding to enhance the Quality of Service (QoS) for end users. The proposed UINT leverages IBN’s automated sensing capabilities to perceive end device users’ network intents, proactively formulating adaptive network traffic forwarding strategies and deploying these strategies to switches before the user’s requested network traffic arrives. When the traffic user requests arrive, it directly matches the flow table for forwarding, eliminating any waiting time. UINT considers the differences in QoS requirements of various traffic and adjusts the traffic forwarding strategy based on network conditions, providing new perspectives and methods for formulating network traffic forwarding strategies. We verify the effectiveness and reliability of UINT in various network environments through experiments. Extensive evaluation experiments indicate the UINT predictor’s effectiveness and the efficacy of its adaptive routing algorithm and dynamic adjustment mechanism in optimising network traffic latency, throughput, and bandwidth.

Tigernut is a potential source of valuable edible oil; however, current oil extraction techniques are inefficient. We assessed high temperature-induced variations in oil absorption and enzymatic hydrolysis of tigernut starch (TS) in the presence of protein to explore the intrinsic reasons for the low oil extraction from tigernut. The results showed that, due to high temperature and the presence of protein, an increase in the volume mean diameters and agglomeration of TS granules occurred. As the temperature increased (80-140 degrees C), the relative crystallinity (19.09 %-24.40 %) of the long-range ordered structure and the orderliness of the short-range ordered structure increased, the total oil absorption (TOA: 0.25-0.19 g oil/g sample) decreased, and the starch-lipid complex index (2.56 %-24.61 %) increased. With increasing temperature in the range of 170-200 degrees C, the short-range ordered structure of TS became more compact, and the TOA (0.18-0.14 g oil/g sample) and the starch-lipid complex index (24.61 %-5.64 %) decreased. Changes in the structure of TS led to an increase and then a decrease in its thermal stability, an enhancement of the gel network structure, and a weakening of enzymatic hydrolysis. Results can help reveal the oil absorption mechanism of TS and regulate its physicochemical properties for the efficient extraction of tigernut oil.

Encrypted network traffic classification plays an important role in enhancing network security and improving network performance. However, the imbalanced nature of traffic data makes the classification of encrypted network traffic challenging and may result in poor classification performance. Existing encrypted network traffic classification studies attempt to rebalance the data distribution through resampling strategies, which suffer from information loss, overfitting, and increased model complexity. Motivated by this, we propose an improved supervised contrastive learning approach to improve the classification performance of supervised contrastive learning classifiers for the traffic class imbalance problem in encrypted network traffic classification. Our method consists of two parts: data processing and traffic classification. In the data processing stage, we transform the raw network traffic data into grayscale images. In the traffic classification stage, we design optimized class-complement and class-averaging schemes in supervised contrastive learning. The construction of contrastive tasks is a critical link in contrastive learning. However, when constructing the set of positive and negative samples of network traffic, the samples generated by traditional methods do not conform to the salient features of network traffic. Traditional methods typically involve color modification, cropping, rotation, noise injection, and random erasure. When these traditional methods are applied to images generated from network traffic data, they may alter significant features of the network traffic data, such as changing the distribution of packet sizes. This is detrimental to maintaining the characteristics of traffic classes and does not aid the learning process. Therefore, we preprocess the traffic into images in a particular format suitable for contrastive learning, and then design a novel contrastive task construction method. The evaluation results on public datasets show that the proposed method can significantly improve the classification performance of encrypted traffic classification on imbalanced datasets.

With the exponential growth of user-generated on-line videos, multimodal sarcasm detection has recently attracted widespread attention. Despite making significant progress, there are still two main challenges: 1) previous works primarily relied on word-level feature interactions to establish relationships between inter-modality and intra-modality, which could potentially lead to the loss of fundamental emotional information. 2) they obtained the incongruity information only interacted with textual modality, which may lead to the neglect of incongruities. To address these challenges, we propose a novel utterance-level incongruity learning network (ULIL) for multimodal sarcasm detection, where the multimodal utterance-level attention (M-ULA) and incongruity learning network (ILN) are the two core modules. First, we present M-ULA to interact with utterance-level multimodal information, complementing word-level features. Furthermore, ILN selects primary modality and auxiliary modality automatically, and leverages cross-attention and self-attention to learning incongruity representations. We conduct extensive experiments on public datasets, and the results indicate that our proposed model achieves state-of-the-art performance in multimodal sarcasm detection.

This study investigated the efficacy of using plasma-activated water (PAW) as a novel and additive-free pretreatment in reducing the formation of free and bound heterocyclic aromatic amines (HAAs) and advanced glycation end products (AGEs) in roasted fish patties. PAW activated for different durations (50 s, 100 s, and 150 s), significantly decreased the levels of HAAs and AGEs. The highest inhibition rates were observed at 35.44% and 19.82% for free and bound HAAs, respectively, and 42.76% and 23.23% for free and bound AGEs, respectively. PAW pretreatment reduced the formation of HAAs and AGEs and mitigated their increase during storage. Analysis using electron paramagnetic resonance and UPLC-MS/MS showed a decrease in free radicals and reactive carbonyls (phenylglyoxal, glyoxal, and methylglyoxal), which can serve as intermediates for HAAs and AGEs. Correlation analysis indicated a significant positive correlation between HAAs/AGEs and phenylacetaldehyde, glyoxal, methylglyoxal, and the total spin number (P < 0.05). This suggests that PAW suppresses the production of reactive carbonyls by quenching free radicals, thereby inhibiting the formation of HAAs and AGEs. In conclusion, with its additive-free approach, PAW pretreatment holds promise for reducing HAAs and AGEs and improving the safety of roasted fish products.

The abilities of Chinese quince free proanthocyanidins (FP) and bound proanthocyanidins (BP) at different levels (0.1%, 0.15%, and 0.3%) to mitigate heterocyclic aromatic amine (HAA) formation in fried chicken patties were investigated for the first time and compared with vitamin C (Vc). FP and BP reduced HAAs in a dose-dependent manner. Significantly, high concentrations of FP (0.3%) resulted in a reduction of PhIP, harman, and norharman levels by 59.84%, 22.91%, and 38.21%, respectively, in chicken patties. The addition of proanthocyanidins significantly (p < 0.05) reduced the weight loss of fried chicken patties. Furthermore, a positive correlation was observed among pH, weight loss, and total HAA formation in all three groups (FP, BP, and Vc). Multivariate analysis showed that FP had a more pronounced effect than BP from the perspective of enhancing the quality of fried chicken patties and reducing the formation of HAAs. These results indicate that proanthocyanidins, both BP and FP, but especially FP, from Chinese quince can inhibit the formation of carcinogenic HAAs when added to protein-rich foods that are subsequently fried.

Flos Sophorae Immaturus extract (FSIE) is rich in flavonoids and phenolics. The effects of various concentrations (0.2, 0.4, and 0.6 g/100 g of meat) of FSIE on the quality and heterocyclic amine (HCA) formation of chicken patties fried at 170 degrees C and 190 degrees C were investigated for the first time. The total HCA content of chicken patties fried at 190 degrees C (7.23-13.73 ng/g) was remarkably higher than that of patties fried at 170 degrees C (2.34-2.57 ng/g). It is noteworthy that most HCAs formed on the surface of the patties. FSIE at 0.2-0.6 g/100 g significantly reduced the formation of HCAs in chicken patties fried at 190 degrees C by 27.09%-47.34%, which may be related to the enhanced total phenolics, total flavonoids, and antioxidant activities of FSIE after heating. With the addition of FSIE, the L-& lowast; and a(& lowast;) values of meat patties decreased and b(& lowast;) values increased. However, the addition of FSIE to chicken did not negatively affect the sensory attributes of fried chicken patties. Also, FSIE somewhat reduced the TBARS and carbonyl values of fried chicken patties. Correlation analysis revealed that the protein carbonyl value of meat patties was significantly and positively correlated with the total HCA formation (r = 0.90, p < 0.01). These results provide evidence that the addition of FSIE to high-temperature fried products, especially on the product surface or in the frying medium, can effectively reduce the formation of hazardous HCAs.

Side-scan sonar (SSS) images have a wide range of applications in underwater target detection and recognition. However, due to the complexity of the underwater environment, the classification performance of sonar images is usually constrained by issues such as noise and inconspicuous texture features, including speckle noise, sensor noise, and interference from other sources. These noises can degrade the image quality, making it challenging to extract meaningful features and affecting the performance of classification algorithms. To address these challenges, we propose a novel classification model named Shuffle-RDSNet for SSS images. Specifically, we design the residual dual-path shrinkage network (RDSNet), which utilizes a soft thresholding function and determines the threshold by combining two paths to extract features from varying scales. The RDSNet is then integrated with the ShuffleNet V2 network to construct the proposed Shuffle-RDSNet model. This approach effectively mitigates the effect of noise in the feature extraction process and enhances the classification performance of the model. Furthermore, we employ additional techniques, including dilated convolution and depthwise separable convolution (DSC), to optimize the model and further enhance the classification accuracy and performance of the proposed method. Experimental results show that our model outperforms other classification models with a classification accuracy of up to 96.74

This work systematically analyzed the microstructure and physicochemical properties of amylose (AM) and amylopectin (AP) isolated from tigernut starch (TS). The yields of AM and AP were 66.50

Sesame oil may contain the heterocyclic aromatic amines（HAAs） 9H-pyrido[3,4-b]indole（norharman） and 1-methyl-9H-pyrido[3,4-b]indole（harman）. A high performance liquid chromatography-fluorescence detection（HPLC-FLD） method was established for effectively detecting HAAs content in sesame oil and used to compare the HAAs content difference among sesame oils extracted from sesame of different producing areas. The results showed that HPLC-FLD could complete determination for HAAs in sesame oils within 15 min. The content of HAAs in 14 sesame oils from different producing areas were significantly different（p＜0.05） and the HAAs in sesame oil from Africa（421.87 μg/kg） was higher than that from Asia（329.32 μg/kg） under the same roasting conditions. Among the 14 sesame oils, the highest HAAs（720.81 μg/kg） was found in sesame oil from Ethiopia, and the lowest HAAs（236.01 μg/kg and 239.63 μg/kg） was found in sesame oil from Sudan and Zhumadian, Henan province, China, respectively. In conclusion, the content of HAAs in sesame oil is related to the sesame raw materials, and the selection of suitable sesame raw materials is important for the reduction and control of HAAs in sesame oil and for the safety of sesame oil consumption.

The world has observed an increasing trend in the development of Privacy-Preserving Machine Learning (PPML) for cross-silo collaborative model training over sensitive data. As the first essential step of cross-silo PPML, it is critical that the parties can align their dataset with privacy assurance, i.e., private data join . However, the existing private data join methods typically leak the ID information in the dataset intersection, which often raises privacy concerns. In this work, we propose iPrivJoin : a novel framework of ID-private data join for PPML. Compared with naively using circuit-based Private Set Intersection (circuit-PSI) for data join, the proposed framework has two advantages: 1) data volume reduction. iPrivJoin utilizes oblivious shuffle to securely trim off the redundant data that is outside the intersection, while the entire dataset needs to be carried to further process in the circuit-PSI based approach. 2) efficiency improvement. iPrivJoin introduces a new private encoding technique to avoid the expensive circuit evaluation that is needed in circuit-PSI. As a result, compared with directly using circuit-PSI, PPML with iPrivJoin enjoys approximately $3\times $ of speedup. Moreover, we propose a new oblivious shuffle protocol, which may be of independent interest. It achieves $1.44\times $ of speedup to the state-of-the-art in the real-world WAN network setting.

Packaging is a food preservation technology widely used in the world. Naturally-sourced, biodegradable polymers are becoming increasingly popular in the food packaging sector. Packaging films prepared using cellulose as raw material would contribute to resource sustainability, but the difficulty of cellulose solubilization limits their further development. In view of this, a series of novel solvent systems (LiCl/DMAc, ILs, TBAH/DMSO, NMMO, alkali/urea solutions, metal-complex solutions) were used to prepare high-strength and high-performance cellulose-based films; their characteristics and the mechanisms involved were investigated. Composite films prepared by blending cellulose with various polymers (synthetic polymers, natural polymers, proteins and enzymes, metal particles, and plant phenols) and their functional properties were also investigated. Problems with cellulose-based films and the outlook for future research are discussed. It is hoped that in the near future high-performance, biodegradable and sustainably-produced cellulose-based films will lead the food packaging industry into a new era and further promote the sustainable use of resources.

Diethylene glycol has been used as a reaction solvent in model systems because of its high boiling point and solubility. This study was to evaluate the influences of diethylene glycol on the model systems of 9H-pyrido[3,4-b]-indole (norharman) and 1-methyl-9H-pyrido[3,4-b]-indole (harman) formation. The model systems containing diethylene glycol had much more norharman and harman than those without (p<0.05). Verification experiments proved that the reactant tryptophan combined with formaldehyde to favor norharman formation, and with acetaldehyde to favor harman formation. Formaldehyde and acetaldehyde provided the means for diethylene glycol to promote the formation of norharman and harman, respectively. Results of this study are useful for making informed decisions about construction of model systems simulating the formation of norharman and harman in food.

Surface modification of nanostructured lipid carriers (NLCs) can be an effective way to improve their oral delivery for active ingredients. In this study, four type of guar gum series modified NLCs for the delivery of phytosterols (PS) were constructed and the effects of the polysaccharides on their structure and physicochemical properties were studied. DLS and AFM results revealed that positively charged polysaccharides could bind to PS-NLCs through electrostatic attraction and made the complexes finally take positive charges, while negatively charged polysaccharides were more likely to fill in the gaps of NLC systems to achieve a balance between electrostatic repulsion and intermolecular forces. Although all four polysaccharides exhibited good storage stability and controlled release of PS in simulated intestinal digestion, PS-NLCs modified with partially hydrolyzed cationic guar gum (PHCG) at medium or high concentrations exhibited better gastric stability, mucoadhesion, and cellular uptake, which had considerable significance for improving the oral bioavailability of PS. This might be related to the coating structure of PHCG-PS-NLCs confirmed by AFM, FTIR, and Raman characterization. This study provide a reference value for designing suitable PS-NLC complexes without synthetic surfactants.