We revisit Byzantine-robust distributed estimation for high-dimensional sparse linear models. By combining local l(1)-regularized robust estimation with robust aggregation at the server, the resulting framework applies to pseudo-Huber regression, quantile regression, and sparse SVM. We show that the resulting estimators yield non-asymptotic guarantees and attain near-optimal statistical rates under mild conditions while remaining communication-efficient. Simulations confirm robust estimation, support recovery, and classification accuracy under various Byzantine attacks.
This study investigated the influence of flaxseed oil addition on high-moisture textured vegetable proteins (TVPs) made of soy protein isolate (SPI) and wheat gluten. The effects of oil form (non-emulsified vs. emulsified) and emulsifier type on structure, oxidation stability and digestion were evaluated. Three TVPs were prepared by adding emulsions stabilised with SPI, octenyl succinic anhydride-modified starch and lecithin during extrusion. These samples were named TVP-E-SPI, TVP-E-OSA starch and TVP-E-Lecithin, respectively. Two controls, TVP-Bulk oil and TVP-N, were prepared with added bulk oil and without added oil, respectively. Oil/emulsion addition reduced fibrous layer separation and hardness, increased porosity and enhanced protein secondary structure order. Compared with bulk oil addition, emulsion addition promoted finer filamentous fibre formation with less influence on texture and higher protein secondary structure order. The oxidation stability of emulsion-added TVPs depended on emulsifier type. TVP-E-SPI showed oxidation stability comparable to TVP-Bulk oil, whereas TVP-E-OSA starch showed increased secondary oxidation and TVP-E-Lecithin exhibited increased primary and secondary oxidation. Among emulsion-added TVPs, TVP-E-SPI displayed the most abundant filamentous fibres and highest fibre density, suggesting the strongest oil droplet-matrix interaction. Hardness, chewiness, protein secondary structure order and oxidation stability generally followed the trend: TVP-E-SPI > TVP-E-OSA starch > TVP-E-Lecithin, positively correlating with interaction strength. In vitro digestion demonstrated that the addition of bulk oil or emulsions improved protein digestibility. Emulsion-added TVPs exhibited higher protein digestibility and oil bioaccessibility than TVP-Bulk oil. However, emulsifier type did not significantly affect protein digestion or oil bioaccessibility, likely due to the low emulsion addition level.
Current research on dynamic spectrum access (DSA) has approached the limits of performance and resource utilization at the individual or single-organization level. Specifically, acquiring comprehensive data under strict privacy constraints is challenging, while incorporating external data often incurs prohibitive computing costs. This paper proposes a new swarm learning (SL) architecture for DSA in edge computing, which improves spectrum sensing accuracy and resource utilization efficiency by aggregating group data under strict privacy protection, facilitating deterministic spectrum fair scheduling to suppress access conflicts and ensure the orderly access of massive IoT devices in highly dynamic environments. To address challenges such as community instability induced by edge device mobility, aggressive spectrum contention, communication overhead surges, and interference of similar models in dense areas on global aggregation, a dynamic tree-hierarchical graph architecture is designed. This architecture can adapt to the heterogeneous characteristics of different geographic communities and solve model failure problems caused by edge device mobility without increasing the model parameters or computing power burden of individual devices. Furthermore, to ensure long-term efficiency, we propose a cross-cycle adjustment mechanism based on model groups under the new architecture, enabling SL communities to dynamically select the most suitable models over long cycles, thereby continuously improving overall system efficiency.
Abstract Land finance (LF) in China reflects a short-term, growth-oriented development model. Given the systematic and complex impact of LF on urbanization efficiency (UE), this paper decomposes UE into urbanization technical efficiency (UTE) and urbanization scale efficiency (USE). Using panel data from 281 prefecture-level cities in China from 2004 to 2017, we employ a dynamic spatial Durbin model for empirical analysis. The results show that (1) LF and UTE exhibit a positive U-shaped relationship. LF exerts a direct and spatial spillover effect on UTE that shifts from inhibition to promotion, with stronger long-term effects. (2) LF and USE display an inverted U-shaped relationship. LF’s direct and spatial spillover effects on USE change from promotion to inhibition, and the long-term effects are more pronounced. (3) Real estate investment amplifies LF’s positive impact on UTE and strengthens its spatial spillover on USE but fails to mitigate LF’s negative direct effect on USE. (4) Industrial structure enhances LF’s promotion of USE but temporarily weakens its spatial spillover on UTE—a negative effect that turns positive in the long run. These findings provide empirical support for optimizing LF’s revenue and expenditure structure, improving the long-term assessment system for officials, accelerating industrial upgrading, and alleviating development pressure from the overheated real estate sector to promote sustainable urbanization.
Many natural food proteins with desirable functional properties are limited by poor water dispersibility, which restricts their practical application. In this study, egg yolk granules (EYG) and zein, two intrinsically water-insoluble proteins with distinct origins of insolubility, were used to construct soluble protein complexes through a simple pH-driven co-assembly process. Stable EYG-Zein complexes were obtained within a defined compositional range, and the optimal mass ratio of 2:3 showed the smallest hydrodynamic diameter and the highest colloidal stability. Multiscale characterization indicated that the mutual solubilization of EYG and zein was governed primarily by non-covalent interactions and accompanied by structural reconstruction. Fluorescence quenching and thermodynamic analysis further suggested that hydrophobic interactions were the dominant driving force for complex formation. Compared with dissociated EYG, the EYG-Zein complexes exhibited higher surface hydrophobicity and showed enhanced ice recrystallization inhibition activity, which became stronger with an increasing proportion of zein. In an ice cream system, incorporation of EYG-Zein effectively suppressed ice crystal growth while causing minor changes in color and no significant adverse effects on melting behavior, texture, or aroma-related properties. These results demonstrate that pH-driven all-protein co-assembly can serve as an effective strategy for converting EYG and zein into functional colloidal complexes and highlight the potential of EYG-Zein complexes as antifreeze ingredients in frozen foods.
