China Allocates $560 Million to Support 13 Transformative Technology Research Projects, Including 4 in Biomedicine and 2 in AI

The Ministry of Science and Technology issued the “Notice on Releasing the 2017 Project Application Guidelines for the Key Special Program on Critical Scientific Issues in Transformative Technologies under the National Key R&D Program” (hereinafter referred to as the “Notice”). According to the Notice, the 2017 Key Special Program on Critical Scientific Issues in Transformative Technologies will deploy 13 research directions, focusing on areas such as precise reconstruction of chemical bonds, metamaterials, precise mesoscale measurement of organs, bionic dexterous prosthetics, artificial intelligence, and novel terahertz radiation sources. Among these, four are related to medicine and two to AI.

The total state-allocated funding is approximately RMB 390 million. Under the same guideline direction, in principle, only one project will be supported; however, if the review results of the proposed projects are similar and their technical approaches are significantly different, two projects may be supported simultaneously. A dynamic adjustment mechanism will be established to selectively continue support based on mid-term evaluation outcomes. Research content already covered under the five categories of national science and technology programs will not receive duplicate support under this special program. The implementation period of the special program is five years (2017–2021).

Four medical research directions are: precise mesoscale measurement of the 3D structure and functional information of intact organs, precise mesoscale measurement of human organ-on-a-chip systems, novel terahertz radiation sources for biomedical applications, and biomimetic dexterous prosthetics with reconstruction of neural information pathways.

1. Precise Mesoscale Measurement of Complete 3D Structural and Functional Information of Organs

Research Content: Addressing frontier scientific questions in biomedicine, we aim to develop new principles and methods for precise mesoscale measurement, overcoming the bottleneck of existing techniques that struggle to achieve high-resolution 3D measurements in large-volume samples. This will enable high-precision acquisition, reconstruction, and visualization of multidimensional big data in life sciences within vital organs. Furthermore, we will visually present structural and functional maps of different cell types throughout intact organs, annotated with auxiliary coordinates or labels representing anatomical structures, tissue characteristics, and physiological or pathological states.

2. Precision Mesoscale Measurement of Human Organ-on-a-Chip

Research Content: To explore new principles and methods for the mesoscale measurement and characterization of the biochemical features of human organ-on-a-chip systems, and to establish high-resolution, online, and precise detection techniques with multi-parameter, multi-dimensional, and multi-modal capabilities across multiple levels—from molecules and cells to tissues, organs, and even entire systems. This aims to achieve real-time monitoring of micro-organs and objective assessment of their biomimetic structural states. Furthermore, this research will investigate the model characteristics of organ chips and validate their similarity to human tissues, thereby providing technical support for drug screening and disease treatment.

3. Novel Terahertz Radiation Sources for Biomedical Application Research

Research Content: Targeting biomedical applications such as the biological effects and detection of terahertz waves, this study explores the physical mechanisms of terahertz radiation generation through the interaction of free electrons with emerging materials and novel structures. It aims to reveal the fundamental principles of transformative terahertz radiation, overcome the technical bottlenecks of traditional terahertz sources, and generate coherent terahertz radiation that is broadband-tunable, high-power, continuous-wave, miniaturized, and possesses a certain non-diffracting length.

4. Biomimetic Dexterous Prosthetics and Reconstruction of Neural Information Pathways

Research Content: Centered on the scientific objective of "restoring hand function," this research explores the design and manufacturing of biomimetic dexterous prostheses with integrated manipulation and perception, neural interface encoding/decoding algorithms, neural interface hardware systems, methods for reconstructing information channels with the nervous system, and the integration and interaction between neural intelligence and artificial intelligence. Key areas of focus include the principles of designing and manufacturing dexterous prosthetic mechanisms based on soft materials; extraction and decoding of neural signals; neural coding patterns of hand movement information and next-generation neural control models; neural afferent mechanisms of sensory signals and methods for restoring natural sensory function in prostheses. The ultimate goal is to achieve a closed-loop bidirectional neural interface that ensures stability and enables continuous learning and functional improvement.

1. Next-Generation Deep Learning Theories and Technologies

Research Content: Targeting application scenarios such as ubiquitous computing (e.g., mobile computing), high-risk domains (e.g., precision medicine), and high-reliability systems (e.g., intelligent transportation), this research aims to overcome bottlenecks including weak theoretical foundations of deep learning, monolithic model architectures, excessive resource consumption, and strong data dependency.Research on the fundamental theories of next-generation deep learning; non-neural network, resource-efficient deep learning models, methods, and high-efficiency optimization techniques; deep learning methods and technologies suitable for small-sample/unsupervised learning and reinforcement/adversarial learning.

Assessment Indicators: In light of the complex characteristics of deep learning models, such as high nonlinearity and a hierarchical, vast parameter space, establish a theoretical framework to elucidate the working mechanisms of deep learning and develop a suite of analytical tools and methods for deep learning models; devise a series of novel machine learning models, methods, and techniques based on non-neural network architectures, achieving breakthroughs in the interpretability, high scalability, and ease of configuration of deep learning models; propose various deep learning models and methods with low storage and computational resource consumption, and design novel gradient-based and gradient-free optimization techniques that are fast, efficient, and suitable for non-convex deep learning training, thereby significantly enhancing the deployment capability of deep learning technologies; develop deep learning methods and techniques tailored for small-sample, unsupervised, weakly labeled, and multi-label scenarios, reducing the heavy reliance of deep learning on large-scale, high-quality annotated data; create multi-event-triggered deep learning models and technologies to adapt to the open environment of the information society and rapidly emerging new phenomena; expand the application domains of deep learning by proposing methods and techniques suitable for online learning, reinforcement learning, and game-theoretic learning.

2. New Principles, Architectures, and Methods for Deep Neural Network Processors

Research Content: Deep neural networks have played a critical supporting role in various cloud and edge applications. However,Existing chips fall far short of meeting the speed and energy efficiency demands of deep neural networks. It is therefore necessary to explore the design principles, architectures, instruction sets, and programming languages for novel processors capable of efficiently handling large-scale deep neural networks; to investigate the impact of deep sub-micron processes (≤16 nm) and emerging devices on the design methodologies of deep neural network processors; to develop new deep neural network processor chips; and to explore ultra-low-power neural network architectures with fully asynchronous features.。

Performance Metrics: Develop a prototype deep neural network (DNN) processor capable of handling large-scale DNNs comprising 100 million neurons and one billion synapses. The prototype shall support a domestically developed DNN instruction set, integrate hardware neurons and synapses as its computational units, and support time-division multiplexing of hardware neurons. It shall be compatible with mainstream DNN programming frameworks such as Caffe, TensorFlow, and MXNet, and enable the deployment of mainstream DNN architectures including Multilayer Perceptrons (MLP), Convolutional Neural Networks (CNN), Long Short-Term Memory networks (LSTM), Recurrent Neural Networks (RNN), Generative Adversarial Networks (GAN), and Faster Region-based Convolutional Neural Networks (Faster R-CNN). Measured energy efficiency and performance shall exceed those of the NVIDIA M40 GPU by more than 20-fold. Design a benchmark suite for DNN processors covering applications in speech, image, and natural language understanding. Design efficient DNN processor cores and on-chip interconnect architectures. Develop programming languages, compilers, and assemblers tailored for DNN processors. Develop drivers and system software for DNN processors. Deploy the DNN processor in over one million mobile terminals, enabling intelligent tasks that previously required cloud computing to be processed locally on mobile devices.

“The Notice” specifies detailed eligibility criteria for application, such as:Government agencies shall not lead or participate in the application; the project (topic) leader must hold a senior professional title or a doctoral degree.Foreign scientists employed by mainland institutions, as well as scientists from the Hong Kong, Macao, and Taiwan regions, may serve as project (topic) leaders for key special projects. Full-time employees must provide valid proof of full-time employment issued by the mainland employing institution, while part-time employees must provide valid proof of employment issued jointly by the mainland employing institution and the overseas institution. These documents shall be submitted together with the printed preliminary project application.

The project (topic) leader is limited to applying for only one project (topic); ongoing projects (including tasks or topics) under the National Key Basic Research Development Program (973 Program, including the Major Scientific Research Program), the National High-Tech Research and Development Program (863 Program), the National Science and Technology Support Program, the National Special Program for International Scientific and Technological Cooperation, the National Special Program for Development of Major Scientific Instruments and Equipment, the Special Program for Scientific Research in Public Welfare Industries, the National Major Science and Technology Projects, and the Key Special Projects of the National Key R&D Program.Project leaders are not permitted to lead the application for projects (topics). Furthermore, project leaders of ongoing key special projects under the National Key R&D Program (excluding task or topic leaders) are also prohibited from participating in project (topic) applications.Wait.

This application is to be submitted online, with the acceptance period from 8:00 on October 12, 2017, to 17:00 on November 13, 2017. For detailed application instructions, please visit the official website of the Ministry of Science and Technology of China.

Link to the Official Website of the Ministry of Science and Technology:http://www.most.gov.cn/