I'm an experimental computer scientist, software engineering lead, and system architect. I'm solving challenging problems at Google. My current research interests are in on-device machine learning technologies for privacy and safety critical applications.
CV: Updated in March 2020
External Links: Google search, LinkedIn, Google Research, Github.IO, Google Scholar, DBLP, Research Gate, IEEE (1 & 2), ACM (1 & 2), ORCiD, Math Genealogy, USPTO
Emails: my lastname (at) {google, android} (dot) com or my lastname (at) chromium (dot) org
Office: Sunnyvale (CA), Home: San Jose (CA)
English: Yim, Keun Soo (You can call me KS.)
한글: 임 근수, 汉语: 任 根洙
Software Engineering Manager / Tech Lead Manager (current rank), Google, Mountain View, California, USA [09/10/2012-Present]
Has been working on diverse projects (such as Ads, Cloud, Android, Voice Assistant, and YouTube), while taking various roles including software engineer, tech lead and eng manager.
R&D Staff Member / Project Leader (final rank), Samsung Advanced Institute of Technology (SAIT), Republic of Korea
Research Intern, IBM T. J. Watson Research Center, Hawthorne, NY, USA
Software Engineering Intern, Google, Inc., Chicago, IL, USA
Research Intern, National Institute of Information and Communication Technology (NICT), Yokosuka, Japan
RA, Information Trust Institute, UIUC, Illinois, USA
RA, School of Computer Science and Engineering, Seoul National University
Software Engineer (Freelancer), venture companies
GenAI/ML
Robustness of AI/ML Systems Against Failures and Security Threats: My work consistently focused on building reliable, resilient, and secure systems. This is evident in the papers on "forecasting sporadic or spiky production outages," "predicting likely-vulnerable code changes," and "assessment of security defense of native programs." The research combines time-series forecasting, machine learning for anomaly detection, and software reliability engineering to prevent system failures and security vulnerabilities. This area directly addresses the critical challenge of ensuring the dependability of large-scale, complex software systems, especially those powered by AI.
Foundational Models for System Monitoring and Optimization: A significant theme is the application of advanced models, specifically foundational models, to solve complex system problems. The paper on forecasting outages uses a foundational model to predict rare, spiky events—a departure from traditional use cases. This demonstrates a clear interest in pushing the boundaries of these powerful models beyond conventional applications. A future research direction would be to apply these models to other system-level challenges like predicting resource utilization spikes, optimizing energy consumption, or detecting performance bottlenecks in cloud-native environments.
Human-in-the-Loop AI and Intelligent User Interfaces: Several patents, such as "Human-in-the-Loop Voice Automation System" and those on "actionable suggestions," highlight a strong focus on the interaction between humans and intelligent systems. This research area explores how AI can assist users in a context-aware and proactive manner. The work on the Task-oriented Queries Benchmark (ToQB) further underscores this interest by creating a standardized way to evaluate AI assistants and LLMs for user-centric tasks. Future research could extend this to creating more personalized, adaptive, and efficient user experiences through sophisticated AI-driven interfaces.
Machine Learning: ML Infrastructure, Voice Assistant, and Recommendation
Mobile and embedded Software: Modular Android, RTOS, Linux kernel, Open Source Ecosystem
Security & Reliability: Experimental Validation and System Security
Industrial Software Engineering and DevOps - Agile Development for Mobile and Big Data Applications: My research has a consistent thread of automating software development and operations. The paper on "predicting likely-vulnerable code changes" is a prime example of automating security reviews. Similarly, the work on "Norming to Performing" and "Hybrid Testing Automation Framework" focuses on automating deployment and testing procedures. This area is about using data and AI to make the software engineering lifecycle more efficient and reliable. Research could expand into AI-driven code generation, automated testing for complex systems, or proactive defect prevention in continuous integration/continuous deployment (CI/CD) pipelines.
Hardware-Software Co-design: Error Detection, Flash Memory, Sensor Networks
High-Performance, Parallel Computing - Cache/Memory Compression, TCP Offload, Jitters: Early work on "fault characterization and detection in parallel computer systems" and "lightweight silent data corruption error detector for GPGPU" establishes a solid foundation in the field of high-performance computing (HPC) and system dependability. While my more recent work has shifted towards AI and ML, this background is highly relevant for building reliable and fault-tolerant AI/ML platforms. Modern AI models are often trained on large clusters of GPUs, which are susceptible to the very transient faults and silent data corruptions my early research addressed. Therefore, a promising research area would be to merge the expertise in fault-tolerant HPC with the challenges of large-scale AI model training and deployment.
International Journals (all as the first author except for one)
SCIE (Science Citation Index Expanded): 6 papers (LNCS 2006 x 2 papers, LNCS 2004 x 4 papers)
IEEE Transactions: 1 paper (IEEE TCE 2004 - was not SCI in 2004 and is SCI 2020)
ACM Transactions: 1 paper (ACM TECS 2019 - SCIE)
Other: 2 papers (JSTS 2005 - was not SCI in 2005 and is SCI in 2020, LNCS 2011 - no longer SCIE since 2007)
International Conferences (CORE Ranking; all as the first author)
Rank A: 7 papers (IPDPS 2014/2013/2011, ISSRE 2016/2014, SRDS 2016, DSN 2010)
Rank B: 4 papers (SAC 2005, PRDC 2009, RTCSA 2006, PDPTA 2003)
Other: 3+ papers (IEEE Aerospace 2012, OLS 2008, ISOCC 2004)
US Patents: 30+ issued (as of November 2020)
"Few, but ripe."
"It is not knowledge, but the act of learning, not possession but the act of getting there, which grants the greatest enjoyment. When I have clarified and exhausted a subject, then I turn away from it, in order to go into darkness again. ..."
By C. F. Gauss