This scientific research institute is a one of the TOP 4 scientific research institutions in Germany. Well-known as European Research Institute, which is famous for its continuous research and high-quality output in the field of modern life science and medicine, it is committed to integrating the knowledge of chemistry, biology, physics and informatics through advanced analytical technology to provide multi parameter analysis methods for biomaterials and to form innovative researches in comprehensive disciplines, so as to further promote the solution of modern life science and medical problems, disease prevention and early diagnosis for more accurate treatment.
1) Data collection and processing
In the Scientific Research Institute the preprocessing of all kinds of medical sample data is IO intensive. The first challenge users faced with is how to quickly obtain the amount of labeled data and how to improve the efficiency of data preprocessing. The second big challenge is that the medical image, gene map and other image data in the original data occupy a large proportion, the screening and analysis of images is high time-consuming process.
2) Difficulties of Data analysis
To build mathematical model for simulation and processing is a common research method, and the Institute is no exception. Only through the principles and methods of deep learning data processing, can the model be effectively evaluated to ensure the effectiveness of model training. Every training process of deep learning is essentially a process of matrix multiplication and with millions of parameters and millions of iterations, which means a process of huge amount of computation. If the computing ability is not enough, the deep learning training may takes more than a few months, and a training model will lose its timeliness.
In the face of the huge demand for computing, it needs a machine that can provide super computing power, image processing and deep learning abilities to optimize the whole process of scientific research and exploration.
1) Super computing power improves efficiency of image analysis
The images needed for the analysis and research include anatomical structure imaging, pathological images, molecular structure, gene map, and other medical reports. Processing and analyzing these images is a severe test for the cooperation and performance of GPU. NF5488A5 optimizes the topology, adoptes NUMA, supports the full link of PCIe 4.0, using the specified communication between CPU and the nearest GPU, and greatly improves the data transmission efficiency among CPUs. Therefore, when analyzing the related medical and biological images, users can recognize the main features of the image faster and extract the image information, such as the relevant pathological stage, body structure data, molecular structure, gene arrangement, etc. Through Lasso analysis, univariate analysis and other commonly used data analysis methods, these data are further screened and classified.
2) Training and evaluation of deep learning ability assistance model
After the preliminary analysis of the data, NF5488A5 can deeply learn the modeling principles, further evaluate the model, and improve the accuracy of the model by self optimization through a large number of calculations. For example, after inputting the relevant data of gene map into the model, the model will continuously adjust and optimize itself according to its own calculation results and compared with the labeling results of the original data, so as to generate a more accurate gene map analysis model. At the same time, after learning random forest, KNN, NLP and other models, the institute further built its own protein mass spectrometry analysis model based on its own research needs, realized customized analysis and research, and greatly improved the efficiency of the model.
Moreover, NF5488A5 also fully considers the user's needs in terms of energy efficiency and heat dissipation: the 4U size design is suitable for a wider range of data center deployment environment; the optimized power supply strategy can improve the power supply stability and reduce TCO to meet the customer's consideration of energy saving; the advanced heat dissipation system ensures the heat dissipation stability and reliability in real time, and perfectly meets the working temperature of 35 ℃.
After NF5488A5 was put into use, its performance in computing power, deep learning and reasoning greatly exceeded the customer's expectation. In the research of protein molecular structure analysis, gene mapping analysis and other scientific research, the time of data processing and image analysis is greatly saved by more than 30%. Moreover, with the support of powerful computing power and deep learning ability, the learning speed of KNN, xgboost and other models is faster, and it can help researchers to develop a variety of calculation programs and carry out calculation and analysis at the same time, so as to improve the accuracy and reliability of evaluation model. As a new generation of AI server that breaks the test record repeatedly, NF5488A5 is widely used in typical AI application scenarios such as image and video, modeling and analysis, to help customers' scientific researches.