BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//הפקולטה למדעי המחשב והמידע - ECPv6.15.15//NONSGML v1.0//EN
CALSCALE:GREGORIAN
METHOD:PUBLISH
X-WR-CALNAME:הפקולטה למדעי המחשב והמידע
X-ORIGINAL-URL:https://cis.haifa.ac.il
X-WR-CALDESC:Events for הפקולטה למדעי המחשב והמידע
REFRESH-INTERVAL;VALUE=DURATION:PT1H
X-Robots-Tag:noindex
X-PUBLISHED-TTL:PT1H
BEGIN:VTIMEZONE
TZID:Asia/Jerusalem
BEGIN:DAYLIGHT
TZOFFSETFROM:+0200
TZOFFSETTO:+0300
TZNAME:IDT
DTSTART:20250328T000000
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0300
TZOFFSETTO:+0200
TZNAME:IST
DTSTART:20251025T230000
END:STANDARD
BEGIN:DAYLIGHT
TZOFFSETFROM:+0200
TZOFFSETTO:+0300
TZNAME:IDT
DTSTART:20260327T000000
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0300
TZOFFSETTO:+0200
TZNAME:IST
DTSTART:20261024T230000
END:STANDARD
BEGIN:DAYLIGHT
TZOFFSETFROM:+0200
TZOFFSETTO:+0300
TZNAME:IDT
DTSTART:20270326T000000
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0300
TZOFFSETTO:+0200
TZNAME:IST
DTSTART:20271030T230000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260108T110000
DTEND;TZID=Asia/Jerusalem:20260108T120000
DTSTAMP:20260416T061820
CREATED:20260105T064430Z
LAST-MODIFIED:20260105T064430Z
UID:5077-1767870000-1767873600@cis.haifa.ac.il
SUMMARY:Jonathan Shafer (MIT) - From Learning Theory to Cryptography: Provable Guarantees for AI
DESCRIPTION:Title: From Learning Theory to Cryptography: Provable Guarantees for AI \nAbstract: \nEnsuring that AI systems behave as intended is a central challenge in contemporary AI. This talk offers an exposition of provable mathematical guarantees for learning and security in AI systems. \nStarting with a classic learning-theoretic perspective on generalization guarantees\, we present two results quantifying the amount of training data that is provably necessary and sufficient for learning: (1) In online learning\, we show that access to unlabeled data can reduce the number of prediction mistakes quadratically\, but no more than quadratically [NeurIPS23\, NeurIPS25 Best Paper Runner-Up]. (2) In statistical learning\, we discuss how much labeled data is actually necessary for learning—resolving a long-standing gap left open by the celebrated VC theorem [COLT23]. \nProvable guarantees are especially valuable in settings that require security in the face of malicious adversaries. The main part of the talk adopts a cryptographic perspective\,  showing how to: (1) Utilize interactive proof systems to delegate data collection and AI training tasks to an untrusted party [ITCS21\, COLT23\, NeurIPS25]. (2) Leverage random self-reducibility to provably remove backdoors from AI models\, even when those backdoors are themselves provably undetectable [STOC25]. \nThe talk concludes with an exploration of future directions concerning generalization in generative models\, and AI alignment against malicious and deceptive AI.
URL:https://cis.haifa.ac.il/event/jonathan-shafer-mit-from-learning-theory-to-cryptography-provable-guarantees-for-ai/
CATEGORIES:סמינרים מדעי המחשב
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260106T120000
DTEND;TZID=Asia/Jerusalem:20260106T130000
DTSTAMP:20260416T061820
CREATED:20260105T065546Z
LAST-MODIFIED:20260105T065855Z
UID:5084-1767700800-1767704400@cis.haifa.ac.il
SUMMARY:Omrit Filtser (Open University) -  Robustly Guarding Polygons
DESCRIPTION:Title: Robustly Guarding Polygons \nAbstract: A fundamental set cover problem that arises in geometric domains is the classic Art Gallery Problem: given a geometric domain (e.g.\, a polygon)\, place a set of points within the domain\, such that every point in it is seen by at least one of the guards. This problem has many variants and has been studied extensively from many perspectives\, including combinatorics\, complexity\, approximation algorithms\, and algorithm engineering.\nIn this talk I will propose precise notions of what it means to robustly guard a domain\, under a variety of models. While approximation algorithms for minimizing the number of (precise) point guards in a polygon is a notoriously challenging area of investigation\, we show that imposing various degrees of robustness on the notion of visibility coverage leads to a more tractable (and realistic) problem for which we can provide approximation algorithms with constant factor guarantees. \nBased on joint work with Rathish Das\, Matya Katz\, and Joe Mitchell.
URL:https://cis.haifa.ac.il/event/omrit-filtser-open-university-robustly-guarding-polygons/
LOCATION:Hanamal 65 St.\, Amir Building\, Seminar Room 413\, Hanamal 65 St.\, Amir Building\, Seminar Room 413\, Haifa\, Israel
CATEGORIES:סמינרים מדעי המחשב
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260105T110000
DTEND;TZID=Asia/Jerusalem:20260105T120000
DTSTAMP:20260416T061820
CREATED:20260105T063751Z
LAST-MODIFIED:20260105T063751Z
UID:5029-1767610800-1767614400@cis.haifa.ac.il
SUMMARY:Tamer Mour (Bocconi University) - Computing on Encrypted Data; Extremely Fast and Simple
DESCRIPTION:Title: Computing on Encrypted Data; Extremely Fast and Simple \n\nAbstract: \nConsider the following setting for computing on private data. A client uploads an encryption of a big input X to an untrusted server\, and then wishes to make an unbounded number of queries f(X) while hiding f and X from the server and using only its secret key. How efficiently can this be done?\nI will present a new framework for achieving the above for two useful special cases:\n1- Data access (secret-key private information retrieval; sk-PIR)\, where X is a database and f(X)=X[i] for a secret index i.\n2- Linear computations (encrypted matrix-vector product; EMVP)\, where X is a matrix and f(X)=Xv for a secret vector v.\nOur framework yields extremely practical solutions\, often approaching the concrete efficiency of cleartext computation across relevant metrics (down to a x1.25 overhead). This is enabled by a non-traditional approach to cryptographic design that prioritizes real-world considerations and from which new sources of hardness emerge.\nIn addition\, we obtain new feasibility results for sk-PIR and EMVP based on the standard Learning Parity with Noise assumption (LPN)\, in a parameter regime not known to imply public-key encryption. \nBased on joint works with Fabrice Benhamouda\, Caicai Chen\, Yuval Ishai\, Shai Halevi\, Hugo Krawczyk\, Tal Rabin and Alon Rosen.
URL:https://cis.haifa.ac.il/event/cs-seminar-computing-on-encrypted-data/
CATEGORIES:סמינרים מדעי המחשב
END:VEVENT
END:VCALENDAR