Nuowei Technology is a professional privacy computing technology service provider. With the Nuoweixin privacy protection computing platform as the core, the company has developed a complete set of independent, secure and controllable privacy computing infrastructure platform, which is the realization of "data availability is invisible". ”, “data immovable value”, and the new computing paradigm of “controllable and measurable” data usage, empowering medical, insurance, government affairs, finance and other industries, and promoting multi-party data collaboration and computing.
The amount of private computing data is huge, and it is urgent to improve computing performance
Privacy computing generally uses cryptography to realize multi-party collaborative computing, involving multiple cryptographic protocols and algorithms. In order to ensure the security of the computing process, ciphertext computing often requires larger computing and communication loads than plaintext computing, resulting in performance bottlenecks. question. For example, ciphertext computations are orders of magnitude more expensive than plaintext computations in homomorphic encryption.
In the engineering application of privacy computing platform, higher requirements for computing performance are put forward, and it is necessary to support large-scale and high-concurrency computing under the premise of ensuring data security and acceptable accuracy, especially for some involving super-multi-party and super-large data. application scenarios of volume, complex scenarios, and complex algorithms.
Take the medical field as an example. In terms of data structure, there are many types of medical data. In addition to structured data, there are unstructured data, genetic data, and image data. For example, a doctor's medical order is a very common medical unstructured data, which needs to be semantically recognized by using natural language processing technology, and then converted into structured data for processing. There are many technical problems that need to be solved; For example, the processing of image data involves the delineation and identification of lesions, which requires the support of professional privacy computing tools in practice.
From the perspective of data volume, the amount of data that needs to be processed in the medical field is huge. For example, the amount of genetic data is very large. The whole genome data of a single individual is about 300GB. If a disease study is to be done, even if the sample is only 1000 patients, the amount of data that needs to be processed is 300TB, not to mention that some studies require more Samples, larger data volumes.
GPU Acceleration Solution
Resolve computing performance bottlenecks in complex scenarios
Professor Wang Shuang, founder and chairman of Nuowei Technology, said that in order to improve computing performance, in addition to algorithm optimization such as algorithm acceleration and code acceleration, there is also a type of solution that is hardware acceleration. Through new cryptography technologies and algorithm protocols, combined with GPU and other hardware acceleration technologies and proprietary algorithms have greatly increased the amount of computation, meeting the performance requirements of high throughput and low latency, as well as the real-time requirements of some specific scenarios.
The GPU computing chip is a general-purpose chip that is suitable for deep learning training model fields such as computationally intensive and highly parallel computing.
Combined with NVIDIA V100 GPU, Nuowei Technology's privacy computing platform has launched a GPU acceleration solution to resolve the performance bottleneck of ciphertext computing in complex scenarios, improve ciphertext computing speed, expand privacy computing application scenarios and scale, and further accelerate the business of privacy computing. Falling speed.
In medical scenarios, with the help of NVIDIA V100 GPU, Nuowei Technology's privacy computing platform can ensure good computing performance in medical scenarios of large-scale participants, and can also solve the problem of communication performance degradation caused by large-scale multi-party data. Nuowei Technology integrates various analysis methods in the medical field, such as feature selection, statistical matching, gene alignment, genome-wide association analysis, medical image segmentation and classification, with privacy computing technology, and forms functional modules, which can realize the analysis of medical images, genes, etc. Calculation of unstructured data such as data.
For example, relying on the underlying technology and platform of Nuowei Technology, a set of AI medical imaging analysis system with privacy protection has been developed to realize rapid auxiliary diagnosis of pneumonia caused by novel coronavirus infection. Its applications cover intelligent auxiliary diagnosis of diseases, intelligent delineation of target areas, intelligent judgment of pathological slices, image reconstruction of imaging equipment, and other intelligent auxiliary diagnosis solutions. In addition, labor costs can be saved, and a fast, reliable and accurate auxiliary diagnosis reference can be provided for the doctor's diagnosis.
The above AI medical imaging research content is mainly based on the medical image segmentation algorithm, feature extraction algorithm, quantitative analysis and comparison algorithm based on secure federated deep learning. With the help of V100 GPU, the computing time has been greatly optimized, so it can run complex algorithms, complex Data and research projects with large sample sizes.
Reviewing Editor: Tang Zihong