Medical misinformation propagates rapidly across digital platforms, posing substantial risks to public health decision-making. Traditional content-based detection approaches exhibit limited effectiveness when confronting the nuanced language and domain-specific terminology inherent in health-related claims. While foundation language models demonstrate remarkable semantic understanding capabilities, their direct application to medical misinformation detection encounters two fundamental obstacles: hallucination-induced evidence unreliability and insufficient explainability in high-stakes health contexts. Rather than treating foundation models as monolithic end-to-end predictors, we reconceptualize them as modular evidence generators and verifiers within a collaborative reasoning architecture. We propose Medical misinformation GUard via multi-model Adversarial Reasoning and Dimensional evidence (MedGUARD), a two-stage framework that leverages foundation model collaboration for explainable detection. The evidence generation stage employs adaptive in-context learning and multi-model adversarial negotiation to synthesize reliable multidimensional evidence, effectively mitigating hallucination artifacts. The semantic fusion stage integrates claim representations with evidence-augmented features through multimodal factorized bilinear pooling. Experiments across four medical misinformation benchmarks demonstrate that MedGUARD achieves improvements of 3.2 to 4.5 F1 points over 25 baseline methods, with all advantages confirmed by McNemar’s significance test. Adversarial negotiation reduces hallucination rates by 66% to 70% while generating transparent reasoning trails across seven evidential dimensions. Computational efficiency analysis quantifies MedGUARD’s cost-performance trade-off against all baselines, and cross-domain generalization experiments confirm that its accuracy advantage holds when training and test distributions differ, enabling medical professionals to verify automated assessments before clinical deployment.
Sentiment analysis is important for understanding affective states in human conversations and has potential applications in intelligent healthcare, such as remote emotion monitoring and patient support. While early methods rely on uni-modal textual features, they often fail to capture the full spectrum of emotional cues present in speech and facial expressions. Multi-modal approaches attempt to address this limitation but still suffer from insufficient data and sub-optimal fusion strategies. To overcome these challenges, we propose a novel framework called LLM-LRA for multi-modal sentiment analysis. It integrates large language model (LLM)-guided weak supervision with main-modality-guided low-rank attention. Specifically, modality-specific encoders (LLM-RoBERTa, Wav2Vec2.0, and contrastive language-image pre-training (CLIP)) are used to extract rich representations. This work then develops a main-modality-driven low-rank attention to efficiently fuse audio and visual features under the guidance of LLM-enhanced textual representations. A stacked Transformer with residual connections is further employed to capture dialogue-level temporal dependencies. Finally, we employ ReLU activation and a softmax layer to output emotion labels. Experiments on IEMOCAP and MELD demonstrate the effectiveness of LLM-LRA on public benchmarks, which indicate its potential for healthcare related emotion monitoring.
Many U.S. states have adopted the goal of achieving carbon neutrality in the coming decades. Pathways for achieving these targets would likely yield a reduction of air pollutant emissions as well. We assess several state-level decarbonization pathways for Massachusetts, including electrification of buildings and transportation and decarbonization of the electric grid, estimating their impacts on in- and out-of-state emissions of carbon dioxide (CO2) and several air pollutants. The analysis applies the GCAM-USA integrated assessment model with state-level resolution. We consider interactions between existing and proposed state and regional policies. The most aggressive state decarbonization pathway that was examined would achieve only a 65% reduction in CO2 emissions by 2050 relative to 2015, indicating the need for further decarbonization options within the industry and non-road transportation sectors to achieve net-zero emission goal stated by the current administration. The air pollution co-benefits under these decarbonization pathways are substantial, especially for primary fine particulate matter. While policies focusing only on Massachusetts are shown to deliver in-state reductions of CO2 and air pollutant emissions, these reductions were partially countered by emissions increases in other states, indicating the need for coordinated regional planning to prevent emissions leakage.
Owing to the outstanding strength under casting conditions, eutectic high/medium entropy alloys with the body-centered cubic structure show great application potential and have drawn extensive attention. However, high-strength materials typically have limited ductility. In this study, a novel Al17Cr25Fe25Ni33 eutectic medium entropy alloy (EMEA) with the phase-separated structure is prepared for the first time. The primary FCC1 phase is successfully separated into alpha' phase and the nanoscale stripe-like FCC2 phase, while newly-formed spherical B2 particles with an average size of similar to 129 nm are formed in the alpha' phase. This method significantly improves the balance between strength and ductility of the alloy by delivering sustained and effective strain hardening capacity through a sequentially activated strain hardening (SASH) mechanism. This mainly stems from the inherent heterogeneity of the primary eutectic lamellar structure, the substantial shearing of spherical B2 precipitates, and the heterogeneity of the alpha' and FCC2 phases resulting from the separation of the FCC1 phase. Notably, the alloy achieves a yield strength of 1002 +/- 10 MPa and an ultrahigh tensile strength of 1560 +/- 15 MPa, while maintaining a good ductility of 12.4% +/- 1%. This work sheds light on the SASH effect assisted by multiple hetero-deformation induced strengthening combined with a significant shear mechanism in EMEA, thereby providing an easily implementable procedure to improve the ductility of high-strength materials. (c) 2026 Published by Elsevier Ltd on behalf of The editorial office of Journal of Materials Science & Technology.
Islanded microgrids play a vital role in renewable energy integration and distributed energy management, yet their operational stability is challenged by renewable fluctuations, communication delays, and external disturbances. To address these issues, this paper proposes an event-triggered adaptive voltage regulation strategy for microgrids modeled as multi-agent systems. The proposed scheme integrates a prescribed performance control framework, a switching-threshold event-triggering mechanism, and a full-state constraint handling method. Specifically, each distributed generation unit is modeled as an agent that exchanges local information with its neighbors to achieve global voltage coordination. A switching-threshold event-triggered mechanism is designed to reduce unnecessary control updates and communication, thereby improving energy efficiency and response speed. To ensure system stability under all operating conditions, a barrier Lyapunov function is employed to handle full-state constraints, and a prescribed performance controller is constructed to guarantee that voltage tracking errors remain within predefined bounds during both transient and steady-state phases. Furthermore, an enhanced Northern Goshawk Optimization algorithm is employed to tune the controller parameters. Comparative simulations against five representative population-based optimizers, conducted with the same population size of 30 and maximum iteration number of 50, show that the enhanced NGO attains the lowest final fitness value. Experimental results demonstrate that the proposed strategy significantly outperforms conventional control methods in maintaining voltage stability, enhancing operational efficiency, and improving system robustness.
