Theory
https://csd.cs.cmu.edu/
enThesis Oral Defense - Peter Manohar
https://csd.cs.cmu.edu/calendar/thesis-oral-defense-peter-manohar
<span>Thesis Oral Defense - Peter Manohar</span>
Reddy Conference Room, Gates Hillman 4405 and Zoom
<span><span>Anonymous (not verified)</span></span>
<span><time datetime="2024-07-01T11:00:00-04:00" title="Monday, July 1, 2024 - 11:00">Mon, 07/01/2024 - 11:00</time>
</span>
In Person and Virtual - ET
New Spectral Techniques in Algorithms, Combinatorics, and Coding Theory: The Kikuchi Matrix Method
PETER MANOHAR
<p>In this thesis, we present a new method to solve algorithmic and combinatorial problems by (1) reducing them to bounding the maximum, over <em>x</em> in<em> {-1,1}<sup>n</sup></em>, of homogeneous degree-<em>q</em> multilinear polynomials, and then (2) bounding the maximum value attained by these polynomials by analyzing the spectral properties of appropriately chosen induced subgraphs of Cayley graphs on the hypercube (and related variants) called "Kikuchi matrices". We will present the following applications of this method.</p><ol><li>Designing algorithms for refuting/solving semirandom and smoothed instances of constraint satisfaction problems;</li><li>Proving Feige's conjectured hypergraph Moore bound on the extremal girth vs. density trade-off for hypergraphs;</li><li>Proving a cubic lower bound for 3-query locally decodable codes and an exponential lower bound for 3-query locally correctable codes.</li></ol><p><br><strong>Thesis Committee:</strong> </p><p>Venkatesan Guruswami (Co-Chair, Carnegie Mellon University / University of California, Berkeley)<br>Pravesh K. Kothari (Co-Chair, Carnegie Mellon University / Princeton University)<br>Ryan O’Donnell<br>Uriel Feige (Weizmann Institute)</p><p><em>In Person and </em><a href="https://cmu.zoom.us/j/91890628546?pwd=EbCbqLgDb9UWbARSYSHKapXeXEuG44.1" target="_blank"><em>Zoom</em></a><em> Participation. See announcement.</em></p>
<time datetime="2024-07-01T15:00:00Z">July 1, 2024 11:00am</time>
<time datetime="2024-07-01T17:00:00Z">July 1, 2024 1:00pm</time>
https://www.cs.cmu.edu/~pmanohar/
Ph.D. Candidate, Computer Science Department, Carnegie Mellon University
https://csd.cmu.edu/calendar/thesis-oral-defense-peter-manohar
<a href="mailto:matthewstewart@cmu.edu">matthewstewart@cmu.edu</a>
Thesis Oral
<a href="https://csd.cs.cmu.edu/people/doctoral-student/peter-manohar" hreflang="en">Peter Manohar</a>
<a href="https://csd.cs.cmu.edu/research/research-areas/theory" hreflang="en">Theory</a>
Reddy Conference Room, Gates Hillman 4405 and Zoom
<p>Speaker: PETER MANOHAR, Ph.D. Candidate, Computer Science Department, Carnegie Mellon University</p>
<p>Talk Title: New Spectral Techniques in Algorithms, Combinatorics, and Coding Theory: The Kikuchi Matrix Method</p>
<p>In this thesis, we present a new method to solve algorithmic and combinatorial problems by (1) reducing them to bounding the maximum, over x in {-1,1}n, of homogeneous degree-q multilinear polynomials, and then (2) bounding the maximum value attained by these polynomials by analyzing the spectral properties of appropriately chosen induced subgraphs of Cayley graphs on the hypercube (and related variants) called "Kikuchi matrices". We will present the following applications of this method.</p>
<p>Designing algorithms for refuting/solving semirandom and smoothed instances of constraint satisfaction problems;Proving Feige's conjectured hypergraph Moore bound on the extremal girth vs. density trade-off for hypergraphs;Proving a cubic lower bound for 3-query locally decodable codes and an exponential lower bound for 3-query locally correctable codes.</p>
<p>Thesis Committee: </p>
<p>Venkatesan Guruswami (Co-Chair, Carnegie Mellon University / University of California, Berkeley)</p>
<p>Pravesh K. Kothari (Co-Chair, Carnegie Mellon University / Princeton University)</p>
<p>Ryan O’Donnell</p>
<p>Uriel Feige (Weizmann Institute)</p>
<p>In Person and Zoom Participation. See announcement.</p>
Mon, 01 Jul 2024 15:00:00 +0000Anonymous222335391 at https://csd.cs.cmu.edu6th Learning Theory Alliance Mentorship workshop
https://csd.cs.cmu.edu/calendar/6th-learning-theory-alliance-mentorship-workshop
<span>6th Learning Theory Alliance Mentorship workshop</span>
<span><span>Anonymous (not verified)</span></span>
<span><time datetime="2024-06-04T09:15:00-04:00" title="Tuesday, June 4, 2024 - 09:15">Tue, 06/04/2024 - 09:15</time>
</span>
Virtual Workshop
For upper-level undergraduate and all-level graduate students, and postdoctoral researchers interested in theoretical computer science and machine learning
We are pleased to invite you to the <strong><a href="https://let-all.com/spring24.html" target="_blank">6th Learning Theory Alliance Mentorship workshop</a></strong>, to be held on 4-5 June 2024. <strong>The workshop is free and fully virtual.</strong>
The workshop is intended for upper-level undergraduate and all-level graduate students as well as postdoctoral researchers who are interested in theoretical computer science and machine learning. No prior research experience in the field is expected, and some sessions may be of interest to researchers in adjacent fields. We have several planned events including:<ul> <li>A “<em>how-to</em>” talk on how to be a good collaborator (discussing what healthy collaborations do and don’t look like, setting expectations, transitioning from junior to senior collaborator roles).</li> <li>A “<em>how-to</em>” talk on how to do theory research (covering topics such as formulating research questions and theory problems, breaking a larger problem into smaller toy problems, and day-to-day best practices).</li> <li>A panel discussion on time management (for example, maintaining a balance between learning and solving, work-life balance, deciding how many projects to work on).</li> <li>A social hour with mentoring tables.</li></ul>Our lineup includes <strong>Shuchi Chawla</strong> (UT), <strong>Adam Groce </strong>(Reed College), <strong>Zhiyi Huang</strong> (University of Hong Kong), <strong>Varun Kanade</strong> (Oxford),<strong> Po-Ling Loh</strong> (University of Cambridge), <strong>Audra McMillan</strong> (Apple),<strong> Ankur Moitra</strong> (MIT), <strong>Devi Parikh</strong> (Georgia Tech), <strong>Aaditya Ramdas</strong> (CMU), and <strong>Steven Wu</strong> (CMU).
A short <a href="https://docs.google.com/forms/d/e/1FAIpQLSfrMK78xJOzrqBd3NubkZlxPkZnMZpMCIGdz9AvMGj4MGINMw/viewform" target="_blank">application form</a> is required to participate with an <strong>application deadline of Tuesday, 28 May 2024</strong>.
Students with backgrounds that are underrepresented or underserved in related fields are especially encouraged to apply. We are trying our best to accommodate all time zones. More information (including the schedule) can be found on the <a href="https://let-all.com/spring24.html" target="_blank">event’s website</a>.
This workshop is part of our broader community-building initiative called the <a href="http://let-all.com/" target="_blank">Learning Theory Alliance</a>. <a href="http://let-all.com/" target="_blank"><em>Learn more</em></a>...
To connect with fellow participants and stay in touch for more announcements, we encourage everyone to join the <a href="https://join.slack.com/t/learningtheor-cui5258/shared_invite/zt-2421d3wfl-oNxDwNyAOd50hh1ijAbUEA" target="_blank">LeT-All slack</a>.
<time datetime="2024-06-04T13:15:00Z">June 4, 2024 9:15am</time>
<time datetime="2024-06-04T13:15:00Z">June 4, 2024 9:15am</time>
https://www.cs.cmu.edu/scs-ece-career-center/career-center-events
Conference/Workshop
<a href="https://csd.cs.cmu.edu/research/research-areas/theory" hreflang="en">Theory</a>
<p>Speaker: For upper-level undergraduate and all-level graduate students, and postdoctoral researchers interested in theoretical computer science and machine learningWe are pleased to invite you to the 6th Learning Theory Alliance Mentorship workshop, to be held on 4-5 June 2024. The workshop is free and fully virtual.<br>
The workshop is intended for upper-level undergraduate and all-level graduate students as well as postdoctoral researchers who are interested in theoretical computer science and machine learning. No prior research experience in the field is expected, and some sessions may be of interest to researchers in adjacent fields. We have several planned events including: A “how-to” talk on how to be a good collaborator (discussing what healthy collaborations do and don’t look like, setting expectations, transitioning from junior to senior collaborator roles). A “how-to” talk on how to do theory research (covering topics such as formulating research questions and theory problems, breaking a larger problem into smaller toy problems, and day-to-day best practices). A panel discussion on time management (for example, maintaining a balance between learning and solving, work-life balance, deciding how many projects to work on). A social hour with mentoring tables.Our lineup includes Shuchi Chawla (UT), Adam Groce (Reed College), Zhiyi Huang (University of Hong Kong), Varun Kanade (Oxford), Po-Ling Loh (University of Cambridge), Audra McMillan (Apple), Ankur Moitra (MIT), Devi Parikh (Georgia Tech), Aaditya Ramdas (CMU), and Steven Wu (CMU).<br>
A short application form is required to participate with an application deadline of Tuesday, 28 May 2024.<br>
Students with backgrounds that are underrepresented or underserved in related fields are especially encouraged to apply. We are trying our best to accommodate all time zones. More information (including the schedule) can be found on the event’s website.<br>
This workshop is part of our broader community-building initiative called the Learning Theory Alliance. Learn more...<br>
To connect with fellow participants and stay in touch for more announcements, we encourage everyone to join the LeT-All slack.</p>
Tue, 04 Jun 2024 13:15:00 +0000Anonymous222335053 at https://csd.cs.cmu.eduThesis Oral Defense - Praneeth Kacham
https://csd.cs.cmu.edu/calendar/thesis-oral-defense-praneeth-kacham
<span>Thesis Oral Defense - Praneeth Kacham</span>
Reddy Conference Room, Gates Hillman 4405 and Zoom
<span><span>Anonymous (not verified)</span></span>
<span><time datetime="2024-05-30T15:00:00-04:00" title="Thursday, May 30, 2024 - 15:00">Thu, 05/30/2024 - 15:00</time>
</span>
In Person and Virtual - ET
On Efficient Sketching Algorithms
PRANEETH KACHAM
<p>Sketching refers to a wide variety of techniques to compress large datasets into much smaller forms that can be efficiently processed to answer questions about the original dataset. Over the past few decades, sketching has emerged as a key tool to efficiently handle large datasets in majorly three settings: (i) the Classic setting, in which the dataset is given to us and we want to solve a problem as quickly as possible, (ii) the Streaming setting, in which the underlying dataset is defined by a large stream of updates and we want to compute interesting properties of the dataset using a small amount of space, and (iii) the Distributed setting, in which the dataset of interest is split among multiple servers and we want protocols that use a small amount of communication among servers to solve problems of interest on the underlying dataset. Each of the above settings presents a different challenge with regard to the measure of efficiency we are interested in. </p><p>In this thesis, we study sketching algorithms in these three settings for a variety of problems. While the techniques required to obtain our algorithms differ across problems and settings, the underlying idea of (possibly randomized) data compression to convert the original large dataset into a much smaller form is a key ingredient behind all of the results in this thesis. </p><p><strong>Thesis Committee:</strong> </p><p>David P. Woodruff (Chair)<br>Pravesh K. Kothari<br>Richard Peng<br>Rasmus Pagh (University of Copenhagen)<br> </p><p><em>In Person and </em><a href="https://cmu.zoom.us/j/97294654625?pwd=OG05aHJxckZKNWpCRENvc1F2L1RiUT09" target="_blank"><em>Zoom</em></a><em> Participation. See announcement.</em></p>
<time datetime="2024-05-30T19:00:00Z">May 30, 2024 3:00pm</time>
<time datetime="2024-05-30T21:00:00Z">May 30, 2024 5:00pm</time>
https://www.praneethkacham.com/
Ph.D. Candidate, Computer Science Department, Carnegie Mellon University
<a href="mailto:matthewstewart@cmu.edu">matthewstewart@cmu.edu</a>
Thesis Oral
<a href="https://csd.cs.cmu.edu/people/doctoral-student/praneeth-kacham" hreflang="en">Praneeth Kacham</a>
<a href="https://csd.cs.cmu.edu/research/research-areas/theory" hreflang="en">Theory</a>
Reddy Conference Room, Gates HIllman 4405 and Zoom
Thu, 30 May 2024 19:00:00 +0000Anonymous222335292 at https://csd.cs.cmu.eduComputer Science Thesis Oral
https://csd.cs.cmu.edu/calendar/thesis-oral-WILLIAMS-2024-05-02
<span>Computer Science Thesis Oral</span>
Reddy Conference Room, Gates Hillman 4405 and Zoom
<span><span>Anonymous (not verified)</span></span>
<span><time datetime="2024-05-02T15:00:00-04:00" title="Thursday, May 2, 2024 - 15:00">Thu, 05/02/2024 - 15:00</time>
</span>
In Person and Virtual - ET
Setup Times in Multiserver Systems
JALANI K. WILLIAMS
<p>In many systems, servers do not turn on instantly; instead, a setup time must pass before a server can begin work. These “setup times” can wreak havoc on a system's queueing; this is especially true in modern systems, where servers are regularly turned on and off as a way to reduce operating costs (energy, labor, CO2, etc.). To design modern systems which are both efficient and performant, we need to understand how setup times affect queues. </p><p>Unfortunately, despite successes in understanding setup in the single server setting, setup in the multiserver setting remains poorly understood. To circumvent the main difficulty in analyzing multiserver setup, all existing results assume that setup times are memoryless, i.e. distributed Exponentially. However, in most practical settings, setup times are close to Deterministic, and the widely used Exponential-setup assumption leads to unrealistic model behavior and a dramatic underestimation of the true harm caused by setup times. </p><p>This thesis represents a comprehensive characterization of the average waiting time in a multiserver system with Deterministic setup times, the M/M/k/Setup-Deterministic. In particular, we derive multiplicatively-tight lower and upper bounds on the average waiting time, demonstrating that setup times, along with their distributions, can not be ignored; setup times can cause profound increases in waiting time, especially when the distribution of setup time has low variability. Our bounds are the first closed-form bounds on waiting time in any finite-server system with setup times, including the extensively-studied Exponential setup system. Furthermore, we use our bounds to derive a highly-accurate approximation, which we evaluate in a variety of settings. These results are made possible via our new method for bounding the expectation of a random time integral, called the Method of Intervening Stopping Times or MIST. </p><p><strong>Thesis Committee:</strong></p><p>Weina Wang (Chair)<br>Mor Harchol-Balter<br>Alan Scheller-Wolf<br>Jamol Pender (Cornell University)<br>Bill Massey (Princeton University)</p><p><em>In Person and </em><a href="https://cmu.zoom.us/j/97486384732?pwd=Uk13dzhIY1VIclFwam95Tko2R3pkQT09" target="_blank"><em>Zoom</em></a><em> Participation. See announcement.</em></p>
<time datetime="2024-05-02T19:00:00Z">May 2, 2024 3:00pm</time>
<time datetime="2024-05-02T21:00:00Z">May 2, 2024 5:00pm</time>
https://jalaniw.github.io/
Ph.D. Candidate, Computer Science Department, Carnegie Mellon University
<a href="mailto:matthewstewart@cmu.edu">matthewstewart@cmu.edu</a>
Thesis Oral
<a href="https://csd.cs.cmu.edu/research/research-areas/theory" hreflang="en">Theory</a>
Reddy Conference Room, Gates Hillman 4405 and Zoom
Thu, 02 May 2024 19:00:00 +0000Anonymous222334971 at https://csd.cs.cmu.eduComputer Science Thesis Oral
https://csd.cs.cmu.edu/calendar/thesis-oral-FLANIGAN-2024-04-30
<span>Computer Science Thesis Oral</span>
Reddy Conference Room, Gates Hillman 4405 and Zoom
<span><span>Anonymous (not verified)</span></span>
<span><time datetime="2024-04-30T13:30:00-04:00" title="Tuesday, April 30, 2024 - 13:30">Tue, 04/30/2024 - 13:30</time>
</span>
In Person and Virtual - ET
Strengthening our Participatory Democracy Toolkit using Algorithms, Social Choice, and Social Science
BAILEY FLANIGAN
<p>In most of the world's democracies, policy decisions are primarily made by elected political officials. However, under mounting dissatisfaction with representative government due to issues ranging from social inequality to public distrust, a new proposal is taking off: to augment representative democracy with mechanisms by which the public can <em>directly participate in policymaking</em>. </p><p>The guiding application of this thesis will be one particular model of participation, <em>deliberative minipublics</em> (DMs), though we will argue that our contributions may apply to many models of direct participation. In a DM, a panel of citizens is selected by lottery from the population; then, this panel convenes around a particular policy issue to study background information, deliberate amongst themselves, and then weigh in on the issue. DMs have been gaining momentum over the past decade, and they are now being used at national and supranational levels, and integrated into representative governments.</p><p>Motivated by this application domain, we make the following main contributions: In <strong>Part I</strong>, we design algorithms for performing the random selection of DM participants, a process known as <em>sortition</em>. Our sortition algorithms permit users to make optimal trade-offs between descriptive representation and other desirable properties conferred by randomness, and we characterize these tradeoffs using game theory, optimization, and empirics. In <strong>Part II</strong>, we use a novel social choice theory framework to investigate a notion of representation that departs from descriptive representation in a key way: it accounts for the political reality that people may be affected to <em>widely varying degrees</em> by any given policy decision. In <strong>Part III</strong>, we study a key potential impact of the background information/deliberation phase of a DM: increases in the extent to which participants consider <em>how others in their society may be affected</em> by different policy options. In <strong>Part IV</strong>, we discuss why our contributions can be useful regardless of how DMs ultimately fare in the political sphere, and we highlight how the enclosed research illustrates new ways to combine tools from political science and computer science. </p><p><strong>Thesis Committee</strong></p><p>Ariel Procaccia (Chair) (Carnegie Mellon University / Harvard University)<br>Nihar Shah<br>Anupam Gupta (New York University)<br>Nika Haghtalab (University of California, Berkeley)<br>Ashish Goel (Stanford University)</p><p><em>In Person and </em><a href="https://cmu.zoom.us/j/2671719231" target="_blank"><em>Zoom</em></a><em> Participation. See announcement.</em></p>
<time datetime="2024-04-30T17:30:00Z">April 30, 2024 1:30pm</time>
<time datetime="2024-04-30T19:30:00Z">April 30, 2024 3:30pm</time>
https://sites.google.com/andrew.cmu.edu/baileyflanigan/home
Ph.D. Candidate, Computer Science Department, Carnegie Mellon University
Thesis Oral
<a href="https://csd.cs.cmu.edu/research/research-areas/theory" hreflang="en">Theory</a>
Reddy Conference Room, Gates Hillman 4405 and ZOom
Tue, 30 Apr 2024 17:30:00 +0000Anonymous222334927 at https://csd.cs.cmu.eduComputer Science Thesis Proposal
https://csd.cs.cmu.edu/calendar/thesis-proposal-PRASAD-2024-04-26
<span>Computer Science Thesis Proposal</span>
Gordon Bell Conference Room, Gates Hillman 5117
<span><span>Anonymous (not verified)</span></span>
<span><time datetime="2024-04-26T10:00:00-04:00" title="Friday, April 26, 2024 - 10:00">Fri, 04/26/2024 - 10:00</time>
</span>
In Person
Mechanism Design and Integer Programming in the Data Age
SIDDHARTH PRASAD
<p>Modern-day human-scale marketplaces such as recommender systems, advertisement markets, matching platforms, supply chain industries, electronic commerce platforms, and others must reckon with a balancing act of (i) understanding and respecting the incentives of the system's participants, (ii) obtaining optimal outcomes subject to those incentives, and (iii) ensuring that data is used in a sound manner to improve overall efficiency. The area of mechanism design from economics provides a rich language and toolkit to understand incentives and integer programming is an expressive optimization language that is the workhorse behind most practical solutions to real-world discrete optimization problems. This thesis studies how data-driven decisions can be integrated into fundamental algorithms from both areas to improve performance (economic, memory, run-time, etc.). </p><p>Within mechanism design, I focus on the design of revenue optimal mechanisms from a data-driven lens. I design algorithms, model new learning paradigms, and invent new mechanism classes for a variety of settings including two-part tariffs, combinatorial auctions, shrinking markets, and general multi-dimensional mechanism design. A highlight here is the first general tunable framework for integrating side information into mechanisms to boost revenue while preserving welfare and incentives. Within integer programming, I develop principled new methods for cutting plane configuration, which is one of the most important components in state-of-the-art branch-and-cut solvers. I develop a comprehensive generalization theory for cut selection that (i) unveils new geometric and combinatorial structure in the branch-and-cut algorithm and the class of Gomory cuts, (ii) improves and subsumes prior work via an abstract model of the underlying tree search, and (iii) is validated through experiments that demonstrate the impact of data-dependent parameter tuning. I also derive a new method of sequence-independent lifting for cutting planes that I validate through rigorous theory—by deriving broad conditions under which the new cuts define facets of the integer polytope—and extensive experiments. </p><p>I conclude with avenues for future work; most notably my preliminary ideas on combinatorial auctions with side information which calls for a blend of innovations in auction design and integer programming techniques. </p><p><strong>Thesis Committee:</strong></p><p>Maria-Florina Balcan (Co-chair)<br>Tuomas Sandholm (Co-chair)<br>Gérard Cornuéjols<br>Craig Boutilier (Google Research)<br>Peter Cramton (University of Maryland)</p><p><a href="https://sid-prasad.github.io/files/thesis_proposal.pdf" target="_blank">Additional Information</a></p>
<time datetime="2024-04-26T14:00:00Z">April 26, 2024 10:00am</time>
<time datetime="2024-04-26T14:00:00Z">April 26, 2024 10:00am</time>
https://sid-prasad.github.io/
Ph.D. Student, Computer Science Department, Carnegie Mellon University
<a href="mailto:matthewstewart@cmu.edu">matthewstewart@cmu.edu</a>
Thesis Proposal
<a href="https://csd.cs.cmu.edu/people/doctoral-student/siddharth-prasad" hreflang="en">Siddharth Prasad</a>
<a href="https://csd.cs.cmu.edu/research/research-areas/theory" hreflang="en">Theory</a>
Gordon Bell Conference Room, Gates Hillman 5117
Fri, 26 Apr 2024 14:00:00 +0000Anonymous222334877 at https://csd.cs.cmu.eduAlgorithms, Combinatorics and Optimization Seminar
https://csd.cs.cmu.edu/calendar/seminar-series-ACO-2024-04-25
<span>Algorithms, Combinatorics and Optimization Seminar</span>
Wean Hall 8220
<span><span>Anonymous (not verified)</span></span>
<span><time datetime="2024-04-25T15:00:00-04:00" title="Thursday, April 25, 2024 - 15:00">Thu, 04/25/2024 - 15:00</time>
</span>
In Person
Perfect matchings in the random bipartite geometric graph
XAVIER PÉREZ GIMÉNEZ
<p>We consider the standard random bipartite geometric graph process in which n red vertices and n blue vertices are placed at random on the unit d-dimensional cube and edges are added sequentially, between vertices of different colors, in increasing order of edge-length. A natural question is to ask whether the first edge in the process that results in the minimum degree being at least one coincides, with high probability, with the first edge that creates a perfect matching. While this was already known to be false when d=2, as the thresholds are not even of the same order, we are able to positively answer it for dimension d at least 3. </p><p><em>This is joint work with Abigail Raz. </em></p><p><em>Tea and cookies at 4pm in the Math Lounge, Wean 6220 (bring your own cup if possible)</em></p>
<time datetime="2024-04-25T19:00:00Z">April 25, 2024 3:00pm</time>
<time datetime="2024-04-25T20:00:00Z">April 25, 2024 4:00pm</time>
https://math.unl.edu/xperezgimenez2
Associate Professor, Department of Mathematics, University of Nebraska-Lincoln
https://aco.math.cmu.edu/abs-23-24/apr25.html
<a href="mailto:alanlew@andrew.cmu.edu">alanlew@andrew.cmu.edu</a>
Seminar Series
<a href="https://csd.cs.cmu.edu/research/research-areas/algorithms-and-complexity" hreflang="en">Algorithms and Complexity</a>
<a href="https://csd.cs.cmu.edu/research/research-areas/theory" hreflang="en">Theory</a>
Wean Hall 8220
Thu, 25 Apr 2024 19:00:00 +0000Anonymous222334912 at https://csd.cs.cmu.eduJoint Theory Seminar / Computer Science Speaking Skills Talk
https://csd.cs.cmu.edu/calendar/speaking-skills-WHITEHOUSE-2024-04-24
<span>Joint Theory Seminar / Computer Science Speaking Skills Talk</span>
Gates Hillman 8102
<span><span>Anonymous (not verified)</span></span>
<span><time datetime="2024-04-24T14:00:00-04:00" title="Wednesday, April 24, 2024 - 14:00">Wed, 04/24/2024 - 14:00</time>
</span>
In Person
Brownian Noise Reduction: Maximizing Privacy Subject to Accuracy Constraints
JUSTIN WHITEHOUSE
<p>There is a disconnect between how researchers and practitioners handle privacy-utility tradeoffs. Researchers primarily operate from a privacy-first perspective, setting strict privacy requirements and minimizing risk subject to these constraints. Practitioners often desire an accuracy-first perspective, possibly satisfied with the greatest privacy they can get subject to obtaining sufficiently small error. Ligett, et al. have introduced a “noise reduction” algorithm to address the latter perspective. </p><p>The authors show that by adding correlated Laplace noise and progressively reducing it on demand, it is possible to produce a sequence of increasingly accurate estimates of a private parameter while only paying a privacy cost for the least noisy iterate released. In this work, we generalize noise reduction to the setting of Gaussian noise, introducing the Brownian mechanism. The Brownian mechanism works by first adding Gaussian noise of high variance corresponding to the final point of a simulated Brownian motion. Then, at the practitioner's discretion, noise is gradually decreased by tracing back along the Brownian path to an earlier time.</p><p>Our mechanism is more naturally applicable to the common setting of <em><font>ℓ</font></em><sub>2</sub>-sensitivity, empirically outperforms existing work on common statistical tasks, and provides customizable control of privacy loss over the entire interaction with the practitioner. Overall, our results demonstrate that one can meet utility constraints while still maintaining strong levels of privacy.</p><p><em>Presented as part of the </em><a href="https://www.cs.cmu.edu/~theorylunch/" target="_blank"><em>Theory Lunch Seminar</em></a><em>. </em></p><p><em>Presented in Partial Fulfillment of the CSD Speaking Skills Requirement</em></p>
<time datetime="2024-04-24T18:00:00Z">April 24, 2024 2:00pm</time>
<time datetime="2024-04-24T19:00:00Z">April 24, 2024 3:00pm</time>
https://jwhitehouse11.github.io/
Ph.D Student, Computer Science Department, Carnegie Mellon University
<a href="mailto:matthewstewart@cmu.edu">matthewstewart@cmu.edu</a>
Speaking Skills
<a href="https://csd.cs.cmu.edu/people/doctoral-student/justin-whitehouse" hreflang="en">Justin Whitehouse</a>
<a href="https://csd.cs.cmu.edu/research/research-areas/theory" hreflang="en">Theory</a>
Gates Hillman 8102
Wed, 24 Apr 2024 18:00:00 +0000Anonymous222334910 at https://csd.cs.cmu.eduTheory Lunch Seminar
https://csd.cs.cmu.edu/calendar/seminar-series-THEORY-LUNCH-2024-04-24
<span>Theory Lunch Seminar</span>
Gates Hillman 8102
<span><span>Anonymous (not verified)</span></span>
<span><time datetime="2024-04-24T12:00:00-04:00" title="Wednesday, April 24, 2024 - 12:00">Wed, 04/24/2024 - 12:00</time>
</span>
In Person
Metric Clustering and MST with Strong and Weak Distance Oracles
CHEN WANG
<script type="text/x-mathjax-config">
MathJax.Hub.Config({
tex2jax: {
inlineMath: [ ['$','$'], ["\\(","\\)"] ],
processEscapes: true
}
});
</script><script type="text/javascript" src="https://cdn.mathjax.org/mathjax/latest/MathJax.js?config=TeX-AMS-MML_HTMLorMML"></script>
<p>I will discuss recent results of k-clustering and MST in a new weak-strong oracle model. In this model, for a fixed metric space $(\chi, d)$, we can compute distances in two ways: via a "strong" oracle that returns exact distances $d(x,y)$, and a "weak" oracle that returns distances $ \tilde{d}(x,y)$ which may be arbitrarily corrupted with some probability. This model captures the increasingly common trade-off between employing both an expensive similarity model (e.g. a large-scale embedding model) and a less accurate but cheaper model. Hence, the goal is to make as few queries to the strong oracle as possible. We consider both "point queries", where the strong oracle is queried on a set of points $ S \subset \chi$ and returns $d(x,y)$ for all $x,y \in S$, and "edge queries" where it is queried for individual distances $d(x,y)$.
Our main contributions are optimal algorithms and lower bounds for clustering and Minimum Spanning Tree (MST) in this model. For $k$-centers, $k$-median, and $k$-means, we give constant factor approximation algorithms with only $ \tilde{\ O}(k)$ strong oracle point queries, and prove that $ \Omega{(k)}$ queries are required for any bounded approximation. For edge queries, our upper and lower bounds are both $ \tilde{\Theta}(k^2)$. Surprisingly, for the MST problem, we give an $ \ O{(\sqrt{ \log n})}$ approximation algorithm using no strong oracle queries at all, and we prove a matching $ \Omega{(\sqrt{ \log n})}$ lower bound which holds even if $ \tilde{\Omega}(n)$ strong oracle point queries are allowed.</p><p><em>Based on the </em><a href="https://arxiv.org/abs/2310.15863" target="_blank"><em>this paper</em></a></p>
<time datetime="2024-04-24T16:00:00Z">April 24, 2024 12:00pm</time>
<time datetime="2024-04-24T17:00:00Z">April 24, 2024 1:00pm</time>
https://sites.google.com/view/chen-wang/home
Postdoctoral Researcher, Texas A&M University and Rice University
https://www.cs.cmu.edu/~theorylunch/
<a href="mailto:mpittu@andrew.cmu.edu">mpittu@andrew.cmu.edu</a>
Seminar Series
<a href="https://csd.cs.cmu.edu/research/research-areas/theory" hreflang="en">Theory</a>
Gates Hillman 8102
Wed, 24 Apr 2024 16:00:00 +0000Anonymous222334932 at https://csd.cs.cmu.eduComputer Science Thesis Proposal
https://csd.cs.cmu.edu/calendar/thesis-oral-RAIZES-2024-04-22
<span>Computer Science Thesis Proposal</span>
Reddy Conference Room, Gates Hillman 4405 and Zoom
<span><span>Anonymous (not verified)</span></span>
<span><time datetime="2024-04-22T13:00:00-04:00" title="Monday, April 22, 2024 - 13:00">Mon, 04/22/2024 - 13:00</time>
</span>
In Person and Virtual - ET
Certified Deniability in a Quantum World
JUSTIN RAIZES
<p>Certified deletion is an influential paradigm that allows a user to delegate sensitive information to another user, and later verify that this information has been destroyed. Since its proposal by Broadbent and Islam (TCC '20), certified deletion has been extended to a wide variety of primitives, such as fully-homomorphic ciphertexts, encryption secret keys, signatures, and even programs.</p><p>However, current definitions of certified deletion do not capture all of the advantages of deletion. For example, in revocable signatures, an adversary who deletes a signature on a message <em>m </em>cannot output a valid signature for <em>m</em>, but may still be able to prove that it <em>used to possess one</em>. In other words, it proves that m was signed at some point.</p><p>The missing security property is the notion of deniability. Similarly to how Pass (CRYPTO '03) views deniability as a fundamental property of zero-knowledge, I view deniability as a fundamental property of certified deletion. Unfortunately, classical solutions for non-interactive deniability either lack public verifiability or reveal even more sensitive information in the future, contrary to the goals of certified deletion.</p><p>I propose to study how certified deletion can be strengthened to provide novel <em>deniability</em> guarantees for a wide variety of primitives, which I call <em>certified deniability</em>. These primitives include signatures, non-interactive zero-knowledge, and key leasing. Notably, the problem of deniable key leasing does not seem to have a classical counterpart to date. If time permits, I plan to extend these results to the related setting of unclonability under the name <em>single-copy explainability</em>, which guarantees that an adversary cannot turn one copy of some sensitive information into two copies that can both be explained - the validity of the other can be denied.</p><p><strong>Thesis Committee:</strong></p><p>Vipul Goyal (Chair) (Carnegie Mellon University / NTT Research)<br>Aayush Jain<br>Elaine Shi<br>Giulio Malavolta (Bocconi University / Max Planck Institute for Security and Privacy)</p><p><em>In Person and </em><a href="https://cmu.zoom.us/j/97866320651?pwd=QzFDbjg4UURLcDE0dEtGRlRmVkdEUT09" target="_blank"><em>Zoom</em></a><em> Participation. See announcement.</em></p>
<time datetime="2024-04-22T17:00:00Z">April 22, 2024 1:00pm</time>
<time datetime="2024-04-22T18:30:00Z">April 22, 2024 2:30pm</time>
https://sites.google.com/view/justinraizes/
Ph.D. Student, Computer Science Department, Carnegie Mellon University
<a href="mailto:matthewstewart@cmu.edu">matthewstewart@cmu.edu</a>
Thesis Proposal
<a href="https://csd.cs.cmu.edu/people/doctoral-student/justin-raizes" hreflang="en">Justin Raizes</a>
<a href="https://csd.cs.cmu.edu/research/research-areas/theory" hreflang="en">Theory</a>
Reddy Conference Room, Gates Hillman 4405 and Zoom
Mon, 22 Apr 2024 17:00:00 +0000Anonymous222334876 at https://csd.cs.cmu.edu