Overseeing AI strategies for all business divisions and core AI R&D at a $30B revenue South Korean conglomerate alongside the group's CDO (Chief Digital Officer) and CAIO (Chief AI Officer), with a current focus on LLM and GenAI efforts in the entertainment and media domains :)▎(12/23) We made the top-10 list of "NeurIPS LLM Efficiency Challenge" (NVIDIA A100, aka industry track)! ► Invited Talks• "Addressing the Cost Disease with GenAI" at Seoul National Univ. and CJ Logistics (11/23)

https://mysc.imweb.me/ENGLISH

https://themiilk.com

► Company News• 03/22 - With Series-C funding, AKASA has become a 🦄 startup• 12/21 - AKASA recognized by #CBInsights on their #DigitalHealth 150 list for the second year in a row!• 03/21 - Thrilled to announce $60M Series B led by BOND and new brand identity, AKASA! • 08/20 - Alpha Health named to the 2020 CB Insights Digital Health 150 List - one of the most promising private digital health companies in the world!• 06/20 - Official company launch and $20M Series A led by a16z!► Publications• 08/21 - "Read, Attend, and Code: Pushing the Limits of Medical Codes Prediction from Clinical Notes by Machines" paper presented at #MLHC2021 (and covered by BusinessInsider, EnterpriseAI, and AIThority) • 07/20 - "Deep Claim: Payer Response Prediction from Claims Data with Deep Learning" paper spotlighted at ICML2020 HSYS workshop (covered in Medical AI Times, Datanami, and Synced)► Invited Talks• "Autonomous Medical Coding: A Step toward the Future of Drug Development" at Stanford K-BioX (07/21) • "Addressing Baumol's Cost Disease with Machine Learning Startups!" at Upstage (07/21) and Samsung (05/21)• "Addressing Baumol's Cost Disease in Healthcare with Machine Learning" at US-Korea Conference (UKC 2021, 12/21) and Healthcare NLP Summit 2021 (04/21, featured on AI-TechPark)• "Imagining the Future with Machine Learning Startups!" at Lotte Ventures (2/22), Yonsei Univ (11/21), Kookmin Univ (08/21), Korea Univ (11/20), K-Tech@Silicon Valley 2020 (10/20), Stanford K-BioX (09/20), and Koreans in Silicon Valley 2020 (09/20)• "Navigating Korean Communication Barriers in Silicon Valley" at Samsung (11/21)• "LumièreNet and After!" at Kookmin Univ (10/19)► Community• Program committees for NAACL2022 "Clinical NLP", ML4H2021, ICCV2021 "Computer Vision for Automated Medical Diagnosis", ICML2021 "Self-Supervised Learning for Reasoning and Perception", and NeurIPS2019 "Fair ML for Health" workshops• Reviewers for NeurIPS, ICLR, ICML, ACL, NAACL, MLHC, ML4H, and CHIL

▎Led the AI research to shape the future of education and make its direct impact on the world (at scale) through Udacity!► Publications• 07/19 - "LumièreNet: Lecture Video Synthesis from Audio" (https://arxiv.org/abs/1907.02253) featured in the press (VentureBeat, ElearningInside, FanaticalFuturist, etc) and on podcasts (SydneyBusinessInsights, BildungAltEntfernen, etc)• 05/19 - "Deep Learning to Predict Student Outcomes" paper accepted as a contributed talk to the AI for Social Good Workshop at ICLR 2019! (http://bit.ly/iclr2019) • 09/18 - "Domain Adaptation for Real-Time Student Performance Prediction" (https://arxiv.org/abs/1809.06686)• 07/18 - "GritNet: Student Performance Prediction with Deep Learning" paper presented at EDM 2018 (http://bit.ly/GritNet, accepted for oral)► Invited Talks• 04/19 - Invited talk on "Deep Learning to Predict Student Outcomes" at Texas A&M Univ. Workshop on AI in Education (AIEdu2019)• 02/19 - Our team's invited talk on "Predicting and Improving Student Performance with Machine Learning" at BayLAN conf., Stanford Univ. (http://bit.ly/BayLAN2019)• 10/18 - Invited talks on "Deep Learning to help student’s Deep Learning" at Naver DeView conf. (https://deview.kr/2018/schedule/265) and Sungkyunkwan Univ.► Community• 08/19 - Organized “2019 KDD Deep Learning Day Workshop on Deep Learning for Education (DL4Ed)” at Alaska (http://ml4ed.cc/2019-kdd-workshop)

▎We achieved the best (conversational) speech recognition accuracy on par with humans, reaching the closest performance so far to "Human Parity"! ► Publication• 05/17 - Our team paper "Deep Learning-Based Telephony Speech Recognition in the Wild" accepted to Interspeech 2017 as an oral presentation (http://bit.ly/interspeech2017)

▎Marvell is the world leader in ICs for mass storage - as a signal processing algorithms architect of Marvell's “Best and Brightest” data storage coding and signal processing group, led developments of state-of-the-art signal processing, machine learning, and communication algorithms for HDD (aka baseband MIMO communication systems) and SSD channels over multiple generations' architectures that have shipped in millions of ASIC chips.

► Research Assistant• Probabilistic Graphical Models and Message-Passing Algorithms for Recommender Systems+ Introduced a novel clustering-based message-passing framework for the recovery of a data matrix from incomplete observations associated with recommender systems.+ Modeled the problem using a generative factor graph and proposed a new algorithm, termed IMP, which outperforms other algorithms on real collaborative filtering (e.g., Netflix) data matrices when the fraction of observed entries is very small (i.e., improves the cold-start problem for collaborative filtering systems in practice).+ IMP algorithm was adopted to a restaurant recommendation engine named allTomato of startup NextVerb. allTomato pipes Facebook user's activities (photos, check-ins, and status), leverages the friends' activities as a trust filter, and makes the fraud-resilient personalized recommendations.• LDPC Codes and Iterative Decoding for Channels with Memory+ Solved the joint iterative decoding problem for finite-state channels (FSCs) and low-density parity-check (LDPC) codes by generalizing the linear programming (LP) decoder for binary linear codes to joint decoding of binary-input FSCs.+ Developed a novel convergent iterative solver for the joint LP decoding problem which provides the error floor predictability of LP decoding by pseudo-codeword analysis and also significant gains over BCJR and SOVA based turbo equalization (TE) with a computational complexity of TE.+ The expected application is coding for magnetic storage where the required block-error rate is extremely low and system performance is difficult to verify by simulation.► Teaching Assistant• Taught and prepared for undergraduate digital communications/computer networks labs and graduate channel coding/digital communications recitation sections; graded papers, exams and homework, kept office hours.