Backdoor defense, which aims to detect or mitigate the effect of malicious triggers introduced by attackers, is becoming increasingly critical for machine learning security and integrity. Fine-tuning based on benign data is a natural defense to erase the backdoor effect in a backdoored model. However, recent studies show that, given limited benign data, vanilla fine-tuning has poor defense performance. In this work, we firstly investigate the vanilla fine-tuning process for backdoor mitigation from the neuron weight perspective, and find that backdoor-related neurons are only slightly perturbed in the vanilla fine-tuning process, which explains its poor backdoor defense performance. To enhance the fine-tuning based defense, inspired by the observation that the backdoor-related neurons often have larger weight norms, we propose FT-SAM, a novel backdoor defense paradigm that aims to shrink the norms of backdoor-related neurons by incorporating sharpness-aware minimization with fine-tuning. We demonstrate the effectiveness of our method on several benchmark datasets and network architectures, where it achieves state-of-the-art defense performance, and provide extensive analysis to reveal the FT-SAM’s mechanism. Overall, our work provides a promising avenue for improving the robustness of machine learning models against backdoor attacks. Codes are available at https://github.com/SCLBD/BackdoorBench.

Background: Klotho is a single transmembrane protein originally identified as an ‘aging suppressor'. Emerging evidence reveals that soluble Klotho (sKl) in the circulation plays important roles in anti-aging, anti-oxidation, anti-apoptosis and Wnt signaling. However, the role of serum sKl in the vascular calcification in hemodialysis patients is not clear. The aim of this study was to determine the associations of sKl with abdominal aortic calcification in patients on maintenance hemodialysis (MHD). Methods: 129 MHD patients were enrolled prospectively. Serum sKl level was detected by ELISA. Abdominal aortic calcification was measured by abdomen lateral plain radiograph, and the abdominal aorta calcification (AAC) score was calculated. The sKl levels were observed in patients with different degrees of calcification. Logistic regression analysis was used to determine the risk factor of abdominal aortic calcification in MHD patients. The diagnostic value of sKl for abdominal aortic calcification was assessed using receiver operator characteristic (ROC). Results: Abdominal aortic calcification was seen in 87 of 129 patients. The median AAC score was 4.0 (0.00, 11.00) and the median sKl level was 616.29 (378.19, 821.61) pg/ml. Serum sKl levels were inversely associated with AAC. When evaluated as AAC categories (15) with ordinal logistic regression, each SD higher sKl was associated with 37.1% lower odds of AAC severity (proportional odds ratio: 0.629; 95% confidence interval: 0.413-0.959, p = 0.031) in models adjusted for demographic data, lifestyle factors, traditional CVD risk factors and uremic risk factors. Multivariate logistic regression analysis showed that serum sKl levels and smoking were independent risk factors for severe AAC. The area under the receiver-operating characteristic curve (AUC) of serum sKl for severe abdominal aortic calcification was 0.746 (0.612-0.880, p = 0.001), sensitivity was 0.885, and specificity was 0.562 for a cutoff value of 265.39 pg/ml. Conclusions: Lower serum sKl levels are independently associated with severe AAC. Serum sKl might have a diagnostic value for the severe AAC in MHD patients.

Objective. To compare the safety and efficacy of transnasal high-flow oxygen therapy (HFNT) and noninvasive positive pressure ventilation (NIV) in the treatment of chronic obstructive pulmonary disease (COPD) with type II respiratory failure. Methods. PubMed, the Cochrane Library, Embase, CBM, CNKI, and other databases were searched for randomized controlled trials (RCTS) on the efficacy of HFNT and NIV in the treatment of COPD. Meta-analysis was conducted using RevMan 5.3 software after two researchers screened literatures, extracted data, and evaluated the methodological quality of the included studies according to inclusion and exclusion criteria. Results. A total of 948 patients were included in 12 RCTS. Comprehensive analysis results showed that the HFNC group had higher levels of 12 h-PAO₂, 48 h-PACO₂ and, 48 h-pH than the NIV group, and the differences were statistically significant ( ). There were no significant differences in 24 h-PAO₂ and 72 h-PAO₂, 12 h-PACO₂, 24 h-PACO₂ and 72 h-PACO₂, 24 h-pH, 48 h-pH, and 72 h-pH between the two groups after treatment ( ). Conclusions. Compared with NIV, HFNC does not increase the treatment failure rate in COPD patients with type II respiratory failure, and HFNC has better comfort and tolerance, which is a new potential respiratory support treatment for COPD patients with type II respiratory failure.

Postoperative delirium (POD) is one of the most common and severe complications among older adults, and is closely associated with poor outcomes, increased healthcare costs, and long-term cognitive decline. With the rapid growth of the global aging population and the continuous emergence of new geriatric surgical procedures, the prevention and management of POD have become a major challenge. Traditional risk assessment tools have inherent limitations in clinical practice and are insufficient to meet the growing demands of precision medicine. In recent years, digital technologies have shown tremendous potential in perioperative management; innovations such as machine learning–based predictive models, wearable sensors, electroencephalogram (EEG) monitoring, and large language models are increasingly being adopted for POD risk identification. This review comprehensively summarizes research advances in digital technology for POD risk identification and their role in clinical decision support for older adults, with the aim of providing evidence-based insights and implementation guidance for clinical practice.