- talks - lists.cs.princeton.edu

Linda Cai will present her MSE talk "Novel Behaviors in Combinatorial Auctions" on Friday, May 8, 2020 at 4pm via Zoom.
by Nicki Mahler 04 May '20

04 May '20

Linda Cai will present her MSE talk "Novel Behaviors in Combinatorial Auctions" on Friday, May 8, 2020 at 4pm via Zoom. Link to Zoom: [ https://us04web.zoom.us/j/5668379247 | https://us04web.zoom.us/j/5668379247 ] The members of her committee are as follows: Matt Weinberg (adviser) and Mark Braverman (reader) Understanding the power and limitations of efficient combinatorial auctions in various settings is a problem at the heart of modern day algorithmic game theory. There are two main notions of efficiency, computational efficiency (running the auction requires low computational power) and communication efficiency (a small number of bits need to be exchanged during the auction). For truthful combinatorial auctions with submodular valuation functions, these two notions of efficiency differ doubly exponentially in the approximation guarantees they can provide, and this distinction stems from the fact that demand queries are communication efficient but hard to even approximate in a computationally efficient way. We suggest a mild and natural relaxation of truthfulness under which this distinction disappears. Under our relaxation, we define a notion of demand queries, called bicriterion approximate demand queries, that can be computed in polynomial time. We use these demand queries to design computationally efficient combinatorial auctions satisfying our relaxed notion of truthfulness with approximation guarantees matching those known for communication efficient auctions. We also briefly discuss the problem of revenue maximization in repeated auctions when bidders are using mean-based no regret learning algorithms.

1 0

Molly Pan will present her MSE talk, "Memory-Efficient Membership Encoding in Switches" on Tuesday, May 5, 2020 at 3pm via Zoom
by Nicki Mahler 04 May '20

04 May '20

Molly Pan will present her MSE talk, "Memory-Efficient Membership Encoding in Switches" on Tuesday, May 5, 2020 at 3pm via Zoom. Link to Zoom: [ https://princeton.zoom.us/j/7979634003 | https://princeton.zoom.us/j/7979634003 ] The members of her committee are as follows: Jennifer Rexford (adviser), and Wyatt Lloyd (reader). All are invited to attend. The talk abstract follows below. Network applications often define policies to manage network traffic based on its attributes. For example, service chaining forwards traffic to reach the middleboxes it wants to visit, and access control restricts traffic by checking the permission flags it carries. These policies match against packets' attributes in switches before being applied. However, the prior works of attribute encoding all incur a high memory cost to identify the attributes in the data plane. This paper presents MEME, an encoding scheme that clusters the attributes which tend to appear together in the traffic to reduce the memory usage. Naive clustering would still fail since it is ineffective when a cluster contains an excessive number of attributes. To tackle this, MEME breaks the clusters into smaller ones by encoding a minimal number of attributes separately and by taking advantage of the special structures within the attributes. MEME also leverages match-action tables and reconfigurable parsers on modern hardware switches to achieve a final 87.7% lower memory usage, and applies an approximate graph algorithm to achieve 1-2 orders of magnitude faster compilation time than the prior state of the art. These performance gains pave the way for deployment of a real traffic management system desired by the world's largest Internet Exchange Points.

1 0

Suriya Kodeswaran will present his MSE talk "Tracking P4 Program Execution in the Data Plane" on Friday, May 8, 2020 at 1pm via Zoom.
by Nicki Mahler 01 May '20

01 May '20

Suriya Kodeswaran will present his MSE talk "Tracking P4 Program Execution in the Data Plane" on Friday, May 8, 2020 at 1pm via Zoom. Zoom link: [ https://princeton.zoom.us/j/3304136967 | https://princeton.zoom.us/j/3304136967 ] The members of his committee are as follows: Jennifer Rexford (adviser) and Andrew APpel (reader). All are welcome to attend. Abstract follows below. While programmable switches provide operators with much needed control over the network, they also increase the potential sources of packet-processing errors. Bugs can happen anywhere: in the P4 program, the controller installing rules into tables, or the compiler that maps the P4 program into the resource-constrained switch pipelines. Most of these bugs manifest themselves after certain sequences of packets with certain combinations of rules in the tables. Tracking each packet’s execution path through the P4 program, i.e., the sequence of tables hit and the actions applied, directly in the data plane is useful in localizing such bugs as they occur in real time. The fact that programmable data planes require P4 programs to be loop-free and can perform simple integer arithmetic operations make them amenable to Ball-Larus encoding, a well-known technique in profiling execution paths in software programs that can efficiently encode all N paths in a single ⌈loд(N)⌉-bit variable. However, for real-world P4 programs, the path variable can get quite large, making it inefficient for integer arithmetic at line rate. Moreover, the encoding could require a subset of tables, that would otherwise have no data dependency, to update the same variable. By carefully breaking up the P4 program into disjoint partitions and tracking each partition’s execution path separately, we show how to minimally augment P4 programs to track the execution path of each packet. With this system in place, we then provide intuition towards using the tracked path to detect and localize bugs in P4 programs.

1 0

Joseph Eichenhofer will present his MSE talk "Voiceover: Censorship-Circumventing Protocol Tunnels with Generative Modelling" on Thursday May 7, 2020 at 11am via Zoom.
by Nicki Mahler 30 Apr '20

30 Apr '20

Joseph Eichenhofer will present his MSE talk "Voiceover: Censorship-Circumventing Protocol Tunnels with Generative Modelling" on Thursday May 7, 2020 at 11am via Zoom. Zoom Link: [ https://princeton.zoom.us/j/92944512708 | https://princeton.zoom.us/j/92944512708 ] The members of his committee are as follows: Prateek Mittal (adviser), and Arvind Narayanan (reader). All are welcome to attend. Abstract: Censorship regimes are continuously adopting and deploying state-of-the-art techniques to detect and prosecute open communication on the internet. Multimedia protocol tunneling seeks to disguise covert data communication by processing it directly through a legitimate audio/video communication system. Systems like VoIP and video streaming services use variable bitrate encoding schemes, which leak characteristics of the content they carry through packet sizes and timing. In what we call a content mismatch attack, censors can distinguish between a channel carrying the legitimate media content and one carrying covert data content. We address content mismatch attacks by introducing a novel traffic shaping technique that models the normal media content and applies its properties to the covert content. We constructed a generative machine learning model to restrict covert data transmission such that its timing properties match properties learned from real two-person conversations. Our evaluation finds that modeling the timing properties in the application layer content reduces distinguishing features in the encrypted network traffic. This mitigates content mismatch attacks on coarse-grained timing properties.

1 0

Haochen Li will present his MSE Talk "Emergence of race and gender in face recognition systems" on Thursday May 7th, 2020 at 2pm via Zoom
by Nicki Mahler 29 Apr '20

29 Apr '20

Haochen Li will present his MSE Talk " Emergence of race and gender in face recognition systems" on Thursday May 7th, 2020 at 2pm. The talk will take place via Zoom: [ https://princeton.zoom.us/j/96991086598 | https://princeton.zoom.us/j/96991086598 ] The members of his committee are Arvind Narayanan (adviser) and Olga Russakovsky (reader). Everyone is invited to attend his talk. His abstract follows below. Abstract: Can computer vision recover attributes such as gender and race(or some approximation thereof) from facial images without being trained for such classification? We show that face recognition systems accurately reproduce human categorizations of race and gender in photos of faces, despite receiving no training labels for such classification. We show race and gender categorization both in FaceNet, a state-of-the-art vector embedding, and Microsoft’s Bing Face API, a deployed black-box system. Given that deployed AI systems today use facial expressions to make decisions about people, our work shows that withholding sensitive attributes from the model does not ensure that the models have no access to them.

1 0

Anastasiya Kravchuk-Kirilyuk will present her MSE talk "The B+-tree Index as a Verified Software Unit" on May 7th, 2020 at 1:30 pm via Zoom.
by Nicki Mahler 29 Apr '20

29 Apr '20

Anastasiya Kravchuk-Kirilyuk will present her MSE talk "The B+-tree Index as a Verified Software Unit" on May 7th, 2020 at 1:30 pm via Zoom. Zoom Link: [ https://princeton.zoom.us/j/9432858291 | https://princeton.zoom.us/j/9432858291 ] The members of her committee are as follows: Andrew Appel (adviser) and Lennart Beringer (reader) Everyone is invited to attend her talk. Please see below for talk abstract. Ordered indexes are used in relational databases to allow for a faster retrieval of records. An ordered index can be implemented using a variety of data structures, and the query planner can later decide which implementation to use for optimal performance. This means the implementation details of an ordered index need not be exposed to the user. In this work, we define an API that any instance of an ordered index should implement. Then, we show that a particular instance of an ordered index, the B+-tree with cursors, satisfies this general interface. Finally, we present the B+-tree index instance as a Verified Software Unit, a VST-verified module with explicitly defined sets of imported and exported functions. That is, a B+-tree implementation verified for functional correctness to a B+-tree specification in previous work, is formally proved to be functionally correct w.r.t. the abstract specification of an ordered index. We verify our code using the Coq proof assistant, the Verified Software Toolchain (VST), and CompCert.

1 0

Fwd: NLP guest lectures this week
by Danqi Chen 27 Apr '20

27 Apr '20

---------- Forwarded message --------- From: Danqi Chen <danqic(a)cs.princeton.edu> Date: Mon, Apr 27, 2020 at 12:34 PM Subject: NLP guest lectures this week To: <nlp-group(a)lists.cs.princeton.edu>, <aiml-group(a)lists.cs.princeton.edu> Hello everyone, We will have two guest lectures in my NLP graduate seminar this week from Jesse Thomason (UW) and Diyi Yang (GaTech). Everyone is welcome to attend the lectures if you are interested! Please find the details below. Time: April 28th/30th 1:30-2:50pm Zoom link: https://princeton.zoom.us/j/767550999 Jesse Thomason - April 28th 1:30-2:50pm *Title: *Language Grounding with Robots *Abstract* We use language to refer to objects like "toast", "plate", and "table" and to communicate requests such as "Could you make breakfast?" In this talk, I will present work on computational methods to tie language to physical, grounded meaning. Robots are an ideal platform for such work because they can perceive and interact with the world. I will discuss dialog and learning strategies I have developed to enable robots to learn from their human partners, similar to how people learn from one another through interaction. I will present methods enabling robots to understand language referring expressions like "the heavy, metallic mug", the first work showing that it is possible to learn to connect words to their perceptual properties in the visual, tactile, and auditory senses of a physical robot. I will also present benchmarks and models for translating high-level human language like "put the toast on the table" that imply latent, intermediate goals into executable sequences of agent actions with the help of low-level, step-by-step language instructions. Diyi Yang - April 30th 1:30-2:50pm *Title: *Language Understanding in Social Context *Abstract* Over the last few decades, natural language processing (NLP) has had increasing success and produced industrial applications like search, and personal assistants. Despite being sufficient to enable these applications, current NLP systems largely ignore the social part of language, e.g., who says it, in what context, for what goals. My research combines NLP, linguistics and social science to study how people use language in different social settings for their social goals, with the implications of developing systems to facilitate human-human and human-machine communication. In this talk, I will explain my research from two specific studies. The first part studies what makes language persuasive by introducing a semi-supervised neural network to recognize persuasion strategies in loan requests on crowdfunding platforms, and further designed neural encoder-decoder systems to automatically transform inappropriately subjective framing into a neutral point of view. The second focuses on modeling how people seek and offer support via language in online cancer support communities and building interventions to support patient communication. Through these examples, I show how we can accurately and efficiently build better language technologies for social contexts. Best, Danqi -- *Danqi Chen* Assistant Professor of Computer Science Princeton University http://www.cs.princeton.edu/~danqic/

1 0

Vikram Ramaswamy will present his General Exam on May 4, 2020 at 3pm via Zoom.
by Nicki Mahler 25 Apr '20

25 Apr '20

Vikram Ramaswamy will present his General Exam on May 4, 2020 at 3pm via Zoom. Zoom link: [ https://princeton.zoom.us/j/617474328 | https://princeton.zoom.us/j/617474328 ] The members of his committee are as follows: Olga Russakovsky (adviser), Arvind Narayanan, and Jia Deng Link to Abstract and Reading List: [ https://docs.google.com/document/d/1DTptk7gMpg3XBDHn1tOSQ6VUqezmB9Q9x32EkqI… | https://docs.google.com/document/d/1DTptk7gMpg3XBDHn1tOSQ6VUqezmB9Q9x32EkqI… ]

1 0

Arunesh Mathur will present his Pre FPO "Identifying and Measuring Manipulative Practices at Scale on the Web" on May 1, 2020 at 10am via Zoom.
by Nicki Mahler 24 Apr '20

24 Apr '20

Arunesh Mathur will present his Pre FPO "Identifying and Measuring Manipulative Practices at Scale on the Web" on May 1, 2020 at 10am via Zoom. Zoom link: [ https://princeton.zoom.us/j/95923475880 | https://princeton.zoom.us/j/95923475880 ] The members of his committee are as follows: Marshini Chetty (adviser, reader), Arvind Narayanan (reader); Brandon Stewart, Jonathan Mayer, and Janet Vertesi (non-readers). All are welcome to attend. Please see abstract below. Abstract: The emerging threat of digital manipulation has raised numerous societal concerns, such as a loss of trust in online content, invasion of privacy, financial loss, and loss of autonomy. While recent work has begun to document the various kinds of manipulative threats that users encounter on a daily basis, we still lack a systematic and rigorous understanding of their prevalence on digital platforms. Building up this knowledge is necessary to gain a deeper understanding of the solutions we can develop to inform and protect users in light of these new threats. In this dissertation, I present automated methods that combine web crawling and machine learning to uncover the prevalence of these manipulative practices at scale. Using these methods, I examine three distinct practices on the web. First, I describe how nearly 90% of online advertising content (endorsements) on platforms like YouTube is not disclosed—misleading users into believing they are viewing unbiased, non-advertising content. Second, I describe how 1,818 dark patterns on 1,254 shopping websites mislead and deceive users into making decisions they would otherwise not make. Third, in ongoing research, I describe how emails sent by political campaigns and organizations in the US employ clickbait and other deceptive tactics to influence and solicit donations from the public. I conclude with how the lessons learned from these measurements can be used to build technical defenses and lay out policy recommendations to mitigate their effects.

1 0

Pranay Manocha will present his General Exam "Learning a Perceptual Audio Metric" on April 29, 2020 at 3pm via Zoom
by Nicki Mahler 24 Apr '20

24 Apr '20

Pranay Manocha will present his General Exam "Learning a Perceptual Audio Metric" on April 29, 2020 at 3pm via Zoom. Link to Zoom meeting: [ https://princeton.zoom.us/j/693084957 | https://princeton.zoom.us/j/693084957 ] The members of his committee are as follows: Adam Finkelstein (adviser), Karthik Narasimhan, and Szymon Rusinkiewicz Links to his abstract and reading list can be found here: [ https://docs.google.com/document/d/1yJ3saKtnxT7KYqP_NF1l2Ey9lI1vzdvCipzf2Vy… | https://docs.google.com/document/d/1yJ3saKtnxT7KYqP_NF1l2Ey9lI1vzdvCipzf2Vy… ]

1 0