DS+, BK 2024년도 제 3회 통계세미나 개최 안내(2/26(월))
DS플러스
2024-02-22
2024년도 제 3회 BK 통계 세미나 개최를 안내드립니다.
고려대학교 통계학과 통계연구소, BK21 통계학교육연구팀과 DS+ 사업단 주최로 이루어지는 세미나입니다.
일시 : 2024년 02월 26일 (월) 오전 11시
장소 : 온라인(ZOOM)
연사 : 송민재 박사 (University of Washington)
주제 :
Computational Hardness of Statistical Inference
Abstract :
Efficiently extracting signals from high-dimensional data presents a major challenge in modern statistics and machine learning. Oftentimes, the obstacle lies not in the lack of information within the dataset but rather in computational constraints. Recent research suggests that our ability to extract meaningful information from datasets is fundamentally limited by computational constraints. Thus, when a statistical problem defies efficient algorithms despite being theoretically solvable with unlimited computation, one must ponder: is the failure due to a lack of algorithmic ingenuity or inherent complexity of the task?
To answer such questions, we establish connections between problems from lattice-based cryptography and statistical inference, demonstrating that computational hardness is inherent in certain problems. The Continuous Learning with Errors (CLWE) problem lies at the center of this fruitful connection. This talk will introduce the CLWE problem and explore implications of its computational hardness. One notable application is the construction of "backdoored" Gaussian distributions which have been used to plant undetectable backdoors in certain machine learning models.
홍보 자료 : 첨부파일 확인 부탁드립니다
Zoom 링크는 아래와 같습니다.
참가 Zoom 회의
https://korea-ac-kr.zoom.us/j/4145766503?pwd=FMeFToJRalvz6UDl8xOB9g1uQOyufg.1&omn=89053988454
회의 ID: 414 576 6503
암호: Kustat123@
많은 관심 부탁드립니다.
감사합니다.
고려대학교 통계학과 통계연구소, BK21 통계학교육연구팀과 DS+ 사업단 주최로 이루어지는 세미나입니다.
일시 : 2024년 02월 26일 (월) 오전 11시
장소 : 온라인(ZOOM)
연사 : 송민재 박사 (University of Washington)
주제 :
Computational Hardness of Statistical Inference
Abstract :
Efficiently extracting signals from high-dimensional data presents a major challenge in modern statistics and machine learning. Oftentimes, the obstacle lies not in the lack of information within the dataset but rather in computational constraints. Recent research suggests that our ability to extract meaningful information from datasets is fundamentally limited by computational constraints. Thus, when a statistical problem defies efficient algorithms despite being theoretically solvable with unlimited computation, one must ponder: is the failure due to a lack of algorithmic ingenuity or inherent complexity of the task?
To answer such questions, we establish connections between problems from lattice-based cryptography and statistical inference, demonstrating that computational hardness is inherent in certain problems. The Continuous Learning with Errors (CLWE) problem lies at the center of this fruitful connection. This talk will introduce the CLWE problem and explore implications of its computational hardness. One notable application is the construction of "backdoored" Gaussian distributions which have been used to plant undetectable backdoors in certain machine learning models.
홍보 자료 : 첨부파일 확인 부탁드립니다
Zoom 링크는 아래와 같습니다.
참가 Zoom 회의
https://korea-ac-kr.zoom.us/j/4145766503?pwd=FMeFToJRalvz6UDl8xOB9g1uQOyufg.1&omn=89053988454
회의 ID: 414 576 6503
암호: Kustat123@
많은 관심 부탁드립니다.
감사합니다.
로그인한 회원만 댓글 등록이 가능합니다.
