Quantization via Distillation and Contrastive Learning.

Quantization is a critical technique employed across various research fields for compressing deep neural networks (DNNs) to facilitate deployment within resource-limited environments. This process necessitates a delicate balance between model size and performance. In this work, we explore knowledge distillation (KD) as a promising approach for improving quantization performance by transferring knowledge from high-precision networks to low-precision counterparts. We specifically investigate feature-level information loss during distillation and emphasize the importance of feature-level network quantization perception. We propose a novel quantization method that combines feature-level distillation and contrastive learning to extract and preserve more valuable information during the quantization process. Furthermore, we utilize the hyperbolic tangent function to estimate gradients with respect to the rounding function, which smoothens the training procedure. Our extensive experimental results demonstrate that the proposed approach achieves competitive model performance with the quantized network compared to its full-precision counterpart, thus validating its efficacy and potential for real-world applications.

Associations of environmental factors with total cholesterol level of middle-aged and elderly people in China.

BACKGROUND: Dyslipidemia is a key factor causing cardio cerebrovascular diseases, and the total cholesterol (TC) is an important lipid indicator among them. Studies have shown that environmental factors have a strong association with TC levels. Previous studies only focused on the seasonal variation of TC level and the short-term effects of some environmental factors on TC level over time, and few studies explored the geographical distribution of TC level and quantified the impact of environmental factors in space. METHODS: Based on blood test data which was from China Health and Retirement Longitudinal Study (Charls) database, this study selected the TC level test data of middle-aged and elderly people in China in 2011 and 2015, and collected data from 665 meteorological stations and 1496 air pollutant monitoring stations in China. After pretreatment, the spatial distribution map of TC level was prepared and the regional statistics were made. GeoDetector and geographically weighted regression (GWR) were used to measure the relationship between environmental factors and TC level. RESULTS: The TC level of middle-aged and elderly in China was higher in females than in males, and higher in urban areas than in rural areas, showing a clustered distribution. The high values were mainly in South China, Southwest China and North China. Temperature, humidity, PM10 and PM2.5 were significant environmental factors affecting TC level of middle-aged and elderly people. The impact of pollutants was more severe in northern China, and TC level in southern China was mainly affected by meteorological factors. CONCLUSIONS: There were gender and urban-rural differences in TC levels among the middle-aged and elderly population in China, showing aggregation in geographical distribution. Meteorological factors and air pollutants may be very important control factors, and their influencing mechanism needs further study.

Nonlinear associations between environmental factors and lipid levels in middle-aged and elderly population in China: A national cross-sectional study.

BACKGROUND: Blood lipid is an important factor affecting cardiovascular disease in middle-aged and elderly people. At present, the associations between environmental factors and blood lipid level in elderly people has been controversial, and the nonlinear effect of their relationship is lack of research. METHODS: This study used data from a national cross-sectional survey of blood lipid levels in 13,354 subjects and data from environmental monitoring sites. Logistic regression was used to measure the relationship between the basic characteristics of the study population and blood lipid levels. After controlling the confounding factors, the nonlinear associations between environmental factors and blood lipid levels of middle-aged and elderly people in different geographical regions were studied by random forest model. RESULTS: The risk of dyslipidemia is significantly higher in middle-aged women, obese people, elderly people, and urban people. Smoking and alcohol consumption increase the risk. The associations between environmental factors and lipid levels of middle-aged and elderly people are nonlinear, the correlation effect between air pollutants and blood lipid level is mainly shown in northern China, and the correlation between meteorological factors and blood lipid level is more obvious in southern China. CONCLUSIONS: This study shows that the associations between environmental factors and lipid levels in middle-aged and elderly population are nonlinear and have regional differences. Therefore it should be considered in optimizing the allocation of public health resources and preventing and controlling environmental exposure of middle-aged and elderly population.