- 목적
마이크로소프트연구소아시아와 공동연구협력 프로젝트* 선정
* 2022년 글로벌 핵심인재 양성지원 사업(과학기술정보통신부)의 ‘글로벌 기업 연계형’과제로 추진되며, 본 프로젝트는 ‘국가연구개발혁신법 시행령 제64조제2항의4(연구개발과제 수의 제한)’에 의거하여 동시에 수행 가능한 최대 과제 수에 해당함.
- 운영방향
마이크로소프트연구소아시아에서 선정한 연구주제에 부합하는 창의적 아이디어 공모를 통해 선정
프로젝트 수행시 공동연구협력을 위한 마이크로소프트연구소아시아의 전문가 매칭
공동 연구원* 마이크로소프트연구소아시아에 6개월 파견
* 공동연구기관(국내 대학)의 연구원(석박사생)
- 과정개요
- 프로젝트 지원
– 지원규모: 총 20-21억 원*, 20개 과제
* 글로벌 핵심인재 양성지원 사업(과학기술정보통신부) 예산: 18.5억 원
– 지원분야:
Low-cost AI
Low-cost AI has drawn increasing attention as a growing number of large AI models are proposed and deployed in practice, which is not only expensive to host to handle a huge number of requests but also detrimental to the environment because of massive CO2 emission. To reduce the cost, we welcome the proposal regarding Low-cost AI which aims to optimize AI model in various aspects (including but not limited to improving the efficiency of model inference algorithm, reducing model size and memory cost, and system/hardware-oriented optimization). Priority will be given to the proposals which can contribute to addressing the cost issues regarding NLP and related pre-training models.
Topics:
- Natural Language Processing
- NLP Pre-training
- Model Compression and Acceleration
- Device-aware Model Adaptation
Advanced AI Technologies for Intelligent Multimedia Systems
Well-designed multimedia systems can effectively improve the efficiency of information workers and creators. The recent advance in artificial intelligence (AI) allows us to design more intelligent multimedia systems than ever before. Our team is dedicated to improving the viewing and creation experience of audio and video for information workers and creators.
Topics:
- Image/video representation learning
- Image/video analysis, including recognition, detection, segmentation, and summarization
- Disentangled representation learning and audio/video generation
- Intelligent speech processing
Container network performance acceleration
Smart devices that perform various tasks such as video streaming and conferencing require real-time communication with high throughput and low latency. In particular, it is necessary to offer an efficient computing environment for smart devices as the devices have limited computing resources compared to powerful datacenter servers.
This research project needs to first explores the computing and communication overhead of network processing in smart devices. Specifically, we focus on containers as they are well-suited for the deployment over heterogeneous nodes. Because containers have a long networking stack that includes multiple firewalls and IP forwarding, they incur additional computation overhead for network processing compared to the native environment. To quantify the overhead, we need to evaluate the network throughput and latency on smart devices while measuring CPU usage. Then, we plan to investigate major bottlenecks in the container networking stack through system-level profiling to design and implement a new enhanced networking stack for containers. The goal of this research is to improve the network performance on smart devices and reduce CPU overhead for network processing.
Topics:
- TCP/IP Kernel Networking Stack Analysis
- System Benchmarking and Diagnosis
- Data-plane Acceleration for RTC applications
Infuse AI to Empower Heterogeneous Devices and New Applications on the Edge
With the recent advances in software and hardware, there is computing paradigm shift from centralized cloud computing to distributed computing on the edge. Together with the breakthroughs in AI, we advocate intelligent edge computing to infuse AI to empower heterogeneous devices and diverse AI applications on the edge. We call for proposals on intelligent edge computing that 1) advance the state-of-the-art research by top publications, prototypes and open-source code; 2) build and deploy real edge systems to solve real problems and learn experience in the wild; and 3) leverage Azure cloud to connect and enable new generation devices and applications.
Topics:
- Affordable AI models tailored for diverse hardware (model compression and optimization, AutoML etc.)
- Efficient software stack best utilizing heterogenous resources (system optimization, resource management and scheduling, etc.)
- Learning on the edge (distributed learning, continuous learning, collaborative learning etc.)
- New applications and scenarios (AIoT, AR, VR, gaming, 5/6G etc.)
- Hardware design for AI and AI for hardware design (AI accelerator design, AI accelerator and model co-design, AI for electronic design etc.)
- Privacy and security (user privacy, data and model protection etc.)
Multilingual Neural Machine Translation
Multilingual neural machine translation (MNMT) draw great attention in machine translation community. A multilingual translation model is capable of translate a single source sentence into any target language. It not only helps address low-resource and zero-shot language translation issues, but also simplifies system deployment in productions and application. There are lots of challenges on how to build super MNMT models. We welcome any proposals related to MNMT research on pre-training & fine-tuning techniques, big model training, low-resource translation, zero-shot translation, model structure innovation, model fusion, model compression, cross-lingual generation, and so on. Priority will be given to the proposals that result in disruptive ideas or solutions to MNMT problem.
Low-cost AI has drawn increasing attention as a growing number of large AI models are proposed and deployed in practice, which is not only expensive to host to handle a huge number of requests but also detrimental to the environment because of massive CO2 emission. To reduce the cost, we welcome the proposal regarding Low-cost AI which aims to optimize AI model in various aspects (including but not limited to improving the efficiency of model inference algorithm, reducing model size and memory cost, and system/hardware-oriented optimization). Priority will be given to the proposals which can contribute to addressing the cost issues regarding NLP and related pre-training models.
Intelligent Software Engineering
Recently, big pre-trained models, such as GPT3 and Codex, have achieved surprisingly good results. For example, Codex can seemingly write computer programs. that our AI system that translates natural language to code. There are many potential research opportunities and applications of intelligent software engineering through exploiting the synergy between new AI (Artificial Intelligence) technology and software engineering. In this theme, we look forward to collaborating with the academic community to tackle important challenges in intelligent software engineering. Potential research topics include but are not limited to intelligent and trustworthy code generation, intelligent code repairing, new code searching algorithms, code to code translation, code summarization.
Topics:
- Intelligent and trustworthy code generation
- Intelligent code repairing
- Code searching algorithms
- Code to code translation
- Code summarization
Responsible AI
The development of AI is creating new opportunities to improve the lives of people around the world, from business to healthcare to education. In order to guide the responsible development and use of AI, in 2018, Microsoft proposed a set of human-centered principles, i.e., fairness, reliability and safety, privacy and security, inclusiveness, transparency, and accountability. These principles are the cornerstone of a responsible and trustworthy approach to AI, especially as intelligent technology becomes more prevalent in the services we use every day. In order to support these principles, we not only need to review current AI solutions, but we also need to invest in developing novel responsible machine learning technologies and bring them into reality. In this theme, we look forward to collaborating with the academic community to tackle important challenges in responsible AI. Potential research topics include but are not limited to federated learning, privacy preserving machine learning, robust deep learning, recommendation diversity and fairness, interpretability and causality in machine learning.
Topics:
- Explainable machine learning
- Federated learning
- Privacy preserving machine learning
- Recommendation diversity and fairness
Human-like Visual Learning and Reasoning
Big data driven deep learning has helped significantly improve the performance of visual tasks in the past few years, but it has also exhibited its limitations in scalability and adaptation to real-world scenarios due to its heavy dependency on human annotation, data heterogeneity, and its “black-box” nature. It becomes extremely important to develop architectures and algorithms that can help address the above limitations, by, e.g., leveraging human prior knowledge, knowledge learned in pre-trained models, and approaches that can capture the fundamentals of how human learn, infer, and reason. Our researchers in Microsoft Research Asia are also willing to partner closely with collaborators in academia to more effectively towards these fields.
Topics:
- Video analytics
- Few/zero-shot visual learning
- Learning for interpretability
- Generalization and adaptation
- Multi-modality video understanding
AI-powered Rich Media Communication
Modern work increasingly relies on online collaboration with real-time media communication. In Microsoft, we have a vision to provide real-time, intelligent, and immersive media services to get more fun and productive discussions out of online collaborations. Multimedia and AI technologies are the two pillars of the strategy to realize this vision. We’ve been aggressively adopted computer vision and speech techniques to improve user experiences in Teams. Besides the new user experiences enabled by computer vision and speech analytics, AI also represents a paradigm shift of software development, e.g., the coding and streaming techniques learnt with machines from large-scale data. In this theme, we encourage researchers to invest on core media (audio, video, and screen) technologies and leverage the latest advances in AI to empower media experiences. Our researchers in Microsoft Research Asia are also willing to partner closely with collaborators in universities to move towards the evolution of AI-powered rich media communication.
Topics:
- Real-time computer vision
- Speech analytics
- Augmented reality
- Neural media codec
- AI-based RTC optimization
3D Content Creation for Learning and from Learning
As deep learning demonstrates its capability for 2D image/video generation, deep learning based 3D content creation still faces several challenges. One challenge is the limited amount of 3D data in the existing dataset, which limits research and development of deep learning algorithm in this field. Another challenge is how to fuse deep learning-based 3D content creation methods with existing interactive modeling approaches for variant 3D modeling/animation tasks. In this research theme, we aim to develop a set of dataset and labeling tools, as well as algorithms to tackle these two challenges.
Topics:
- High quality 3D object database acquisition
- Data labeling tools and system
- Learning based 3D content creation algorithms/network design
Efficient Multi-Modal Pre-Training
Recent advances of multi-modal pre-training explored the power of pre-training on large scale of data to accelerate performances on many multi-modal tasks (e.g., visual question and answer, image-text retrieval, video localization, speech recognition). Despite the great success of pre-training models, the large scale of parameters and heavy computation requirement has limited its application in real-world scenarios and exploration of more researchers. To address this challenge and further advance multi-modal research, we welcome research proposals which aim to design fast, lightweight and efficient pre-training models for multi-modality. We believe the efficient multi-modal pre-training models will advance the real scenarios in both research and industry community with the latest pre-training works.
Topics:
- Knowledge distillation for multi-modal pre-training
- Model suppression in multi-modal pre-training
- Quantization in multi-modal pre-training
- Multi-modal pre-training model optimization
Spoken Language Processing
Spoken language processing draws on the latest advances and techniques from multiple fields: computer science, electrical engineering, acoustics, linguistics, mathematics, psychology, and beyond. Given the raw speech signal, spoken language processing tries to enhance it by removing the noise, separate the speech signal for each speaker, recognize it into text, analyzes the semantic meaning, generates the proper responding text, and synthesize it into speech output. With the rapid development of deep learning and the more and more available data for model training, remarkable progress is being made in spoken language processing, such as automatic speech recognition achieves human parity for the specific domain. However, there are still big challenges for further exploration, such as how to leverage large-scale unlabeled speech data, how to improve model accuracy for low resource scenarios, how to deal with the cocktail party problem in a complex scenario.
Topics:
- Pre-training for Speech
- Automatic Speech Recognition
- Multilingual ASR
- Low Resource ASR
- Speech Separation
- Speech Enhancement
- Speech Translation
Multimodal NLP
Learning joint representations of vision and language could lead to the next AI breakthroughs. Motivated by this, we propose multimodal NLP as a research theme and call for collaborations with professors/researchers from academic community. The goal is to develop cutting-edge language-centered multimodality models for various tasks, such as commonsense knowledge learning from visual contents, text-to-image/video retrieval and generation, image/video-based QA, reasoning and captioning etc. From research perspective, we hope the collaborations can lead to impactful research papers or achieve state-of-the-art results on latest research-driven leaderboards.
Topics:
- Language-centered Vision-Language Pre-training
- Commonsense Knowledge Learning from Visual Contents
- Text-to-Image/Video Retrieval and Generation
Reinforcement learning (RL)
Topics:
- Structured reinforcement learning for real-world applications
Trustworthy Artificial Intelligence for Health
In recent years, advancements have been made in applying AI in healthcare and several AI based solutions have been approved for usage around the world. However, to further increase the impact of AI in healthcare, one must further AI technologies in many dimensions, such as transparency, fairness, robustness, and reliability. To give healthcare professionals confidence in using AI based systems, they must be able to understand the basis for the AI systems’ decisions. Such a need is becoming acknowledged by multiple stakeholders in healthcare. For example, recently the Department of Health and Human Services in the US have issued a Trustworthy AI (TAI) Playbook (HHS Trustworthy Artificial Intelligence (AI) Playbook (09/30/2021)).
At MSRA, we have built significant expertise in machine learning, natural language processing, and computer vision. We are integrating our knowledge in these areas to create explainable medical image analysis systems. To go beyond black box systems that can only generate probabilities of classification output, we seek to build systems that can conduct question and answer sessions with radiologists and explain the basis of decisions. We plan to undertake this research in collaboration with partners in academia and industry and look forward to working with Korean professors and students on this important and exciting endeavor.
Deep Learning for Drug Discovery and Design
Drug discovery and design (DDD) is an extremely complex, time-consuming, and expensive process. Incentivized by recent successes in various applications such as computer vision, natural language understanding and generation, speech synthesis and recognition, and game playing, deep learning has attracted more and more attention in DDD, revolutionizing the design-make-test-analyze (DMTA) cycle, accelerating the drug design process and therefore reducing cost. In this theme, we encourage researchers to collaborate and tackle important challenges in DDD, including data sparsity, data quality, data heterogeneity, and lack of interpretability. We welcome research proposals which aim to improve the DDD process by leveraging recent deep learning techniques.
Topics:
- DL for structure-based drug design
- DL for lead optimization
- DL for retrosynthesis
- DL for anti-infective drug design
- DL for clinical drug development
- DL for natural product synthesis
– 지원기간: 2022.5.1 – 2023.4.30
– 지원내용: 프로젝트경비 (기업부문: USD10K와 정부부문: Korea Won 80-100M) – 총 9천 – 1억 2천만원
* 정부부문의 프로젝트비 산정. 사용 등은 정보통신방송 연구개발 관리규정에 따름 (IITP의 추후 안내)
* 기업부문은 기업과제 별도 계약에 따름.
– 선정심사: 마이크로소프트연구소아시아의 전문성심사(서면)
* 심사결과는 선정된 과제에 한하여 개별 통보하며 공개되지 않음
* 과제 공동연구기관으로 선정된 국내 대학은 IITP와 협약 등 정부 과제 수행을 위한 절차를 추진해야 함(대학별로 공동연구기관으로 협약 체결)
- 공동연구원 파견
– 공동 연구원: 별도 심사를 통해 선발
– 파견기간: 6개월(2022. 9월 – 2023. 3월 예정)
– 파견기관: 마이크로소프트연구소아시아 (중국, 북경)
- 신청자격
프로젝트당 학생(2-5명) 및 지도교수로 팀을 구성
학생: 국내 ICT 분야의 연구 및 ICT 융합연구분야 대학원에 재학중인 전일제 석박사과정 대학원생
* 한국 국적의 내국인 (휴학생 또는 박사후 과정은 제외)
교수: 국내 ICT 분야의 연구 및 ICT 융합연구분야 전임교원으로서 지원기간 동안 프로젝트 총괄 및 학생 연구 지도가 가능한 자
- 지원절차
프로젝트 선정 공고 -> 제안서 제출(온라인 접수, 지원양식, 100% 영문제안) -> 선정심사 -> 지원대상 선정통보 -> 협약체결 및 프로젝트 경비 지급
* 지원양식에서 예산작성은 기업부문 USD10K 기준으로 작성. 정부부문의 프로젝트비는 선정통보후 별도 안내 예정
- 신청 유의사항
프로젝트팀은 총1개 분야에 한해 신청할 수 있음
신청자격에 부합하지 않을 경우 선정심사 대상에서 제외될 수 있음
* 국가연구개발사업에 참여제한 중인 자와 기관은 신청 불가
- 신청요령
신청방법: 이메일 신청([email protected])
신청접수마감: 2022년 3월 31일(목) 17:00
* 제출된 서류는 일제 반환되지 않음
- 문의처
사업담당: 마이크로소프트연구소 이미란 전무 (010-3600-4226, [email protected])
