Google research awards go to cloud-service security, probabilistic programming of SDNs, and privacy-preserving deep learning

Several exciting and wide-ranging projects at Cornell have won Faculty Research Awards from Google.

Vitaly Shmatikov (with Adam Smith of Penn State):  a system for privacy-preserving deep learning.  Deep learning based on artificial neural networks has led to dramatic improvements in speech and image recognition, language translation, and other AI tasks, yet centralized collection of training data from millions of users presents serious privacy risks.  This project aims to design and implement a practical system that will enable data holders to learn accurate neural-network models without sharing their training datasets but still benefitting from other participants who are concurrently learning similar models.

Nate Foster and Dexter Kozen: a new probabilistic framework for programming software-defined networks (SDNs). Most existing SDN languages are deterministic, which makes it difficult to give satisfactory treatments of phenomena such as congestion given uncertain traffic patterns or reliability in the presence of failures. This project will develop a new programming language based on probabilistic semantics and tools for compiling programs and automatically verifying formal properties.

Ari Juels and Tom Ristenpart: security property enforcement (SPE) for cloud services. SPE applies in settings in which a cloud provider such as Google wishes to apply proprietary algorithms to customers sensitive data. SPE provides strong evidence to  customers (or auditors working on their behalf) that the provider is not abusing its access to sensitive data and to the provider that malicious customers cannot exploit sensitive information about proprietary provider algorithms.

Rafael Pass: large-scale privacy-preserving computation. Cryptographic methods such as secure multi-party computation enable securely and privately performing any computation on individuals private inputs. Such methods, however, do not scale to the modern regime of large-scale distributed, parallel, data processing.  This project will develop methods to securely and privately process large amounts of data using parallel distributed algorithms.

Date Posted: 9/01/2015

The dark side of "smart contracts"

Ari Juels and Elaine Shi explain to MIT Technology Review that "smart contracts" could be problematic. Together with Ahmed Kosba of University of Maryland, the team, in the provocatively titled paper "The Ring of Gyges: Using smart contracts for crime", show that "criminal smart contracts" can  facilitate leakage of confidential information, theft of cryptographic keys, and various real-world "calling-card" crimes, such as murder, arson, and terrorism.

For example, a criminal organization might put out a smart contract on someone's life: an unpredictable feature of the crime-to-be, such as the time and location of the killing, is provided in a cryptographically sealed form in advance of the crime; the smart contract automatically releases the funds when this proof of responsiblity for the crime is opened by the criminals who perpetrated it.

“In some ways this is the perfect vehicle for criminal acts, because it’s meant to create trust in situations where otherwise it’s difficult to achieve,” says Juels.  “We are optimistic about [these contracts'] beneficial applications, but crime is something that is going to have to be dealt with in an effective way if those benefits are to bear fruit,” says Shi.

Date Posted: 8/28/2015

NSF award on cryptocurrencies in the news.

PIs Elaine Shi and Gün Sirer, with Berkeley PI Down Song, have received an NSF Large Collaborative grant on the science and applications of crypto-currency, leveraging cryptography, game theory, programming languages and systems security techniques to establish a rigorous scientific foundation for crypto-currencies and to build trustworthy infrastructure to enable their next generation applications. This work is key because crypto-currencies are a billion-dollar market, and hundreds of companies are entering this space, promising exciting new markets and eco-systems; but most crypto-currencies rely on heuristic designs without a solid appreciation of the necessary security properties, or any formal basis upon which strong assurance of such properties might be achieved.

Shi and Sirer have been quoted  about the grant in Bitcoin Magazine, Coindesk, and CoinTelegraph; the latter described the award as "probably the [NSF's] broadest effort in the Bitcoin space to date".

Co-PIs on the grant are Michael Hicks, David van Horn, and Jonathan Katz of UMD.

Date Posted: 8/24/2015

Cecil, clicks and Kleinberg in the New York Times

Using the explosion of interest in the story of Cecil the lion as a starting point, the New York Times recently explored the question of whether "online news has deteriorated, and that it is now focused on the viral at the expense of the substantive." While the founder of The Verge quipped that "Reading disposable web journalism is 'like eating a whole bag of Doritos'", Jon Kleinberg comments that these kinds of stories have probably always caught on, and what has changed now is just that the internet "makes visible and searchable what was once ephemeral".

Date Posted: 8/18/2015

The CoinTelegraph covers Ittay Eyal's "The Miner's Dilemma"

The CoinTelegraph has covered Ittay Eyal's work on "the Bitcoin Miner's Dilemma", which is concerned with block withholding attacks, where an individual miner can cheat the Bitcoin mining pool he or she belongs to, contribute nothing, but still receive rewards. 

Previously, it had been believed that miners had no incentive to engage in this attack, due to the need to expand resources on it. Eyal argues that the system is currently faced with a version of the classical prisoner’s dilemma: it would be better for all Bitcoin mining pools to not attack each other, as no hashing power would get wasted on attacks in that case; but it is better for all individual mining pools to carry out the attack if they are the only one doing so. However, it is unsustainable to be attacked and not launch a counterattack – logically leading to all pools attacking each other. Eyal concludes that smaller and private mining pools would lead to be a better environment for Bitcoin as a whole.

Date Posted: 8/18/2015

Ari Juels, Mor Naaman, and Vitaly Shmatikov receive Google expeditions grant on the "Open Web of Things"

Ari Juels, Mor Naaman, and Vitaly Shmatikov have received an Expedition Lead Grant from Google's "Open Web of Things" program.  Describing the effort, Juels writes, "we believe a key threat is underexplored: the correlation between dissimilar data sources, such as physical sensors and social media. We call this threat parallax privacy infringement. The term “parallax” denotes the displacement of an object when seen from two vantage points, which reveals distance and depth. Similarly, the vantage points afforded by disparate data streams can reveal a great deal about users. Our role in the expedition program will be to explore the implications to user privacy."

The mission of Google's program is to "enable effective use and broad adoption of the Internet of Things by making it as easy to discover and interact with connected devices as it is to find and use information on the open web"; grants were awarded to CMU, Cornell Tech, Illinois and Stanford.

Date Posted: 8/18/2015

Science News reports on Vlad Niculae and Cristian Danescu-Niculesu-Mizil's work on linguistic predictors of betrayal

Science News interviewed Cristian Danescu-Niculescu-Mizil in a zippy article about his work with Cornell PhD student Vlad Niculae and colleagues at University of Maryland and UC-Boulder on linguistic signals of upcoming betrayals. Using the online game of Diplomacy as the source of data, the team found that "Players who were excessively polite in general were more likely to betray, and people who were suddenly more polite were more likely to become victims of betrayal"; also, "An increase [in] planning-related language by the soon-to-be victim also indicated impending betrayal, a signal that emerges a few rounds before the treachery ensues. And correspondence of soon-to-be betrayers had an uptick in positive sentiment in the lead-up to their breach." But "more important than the clues themselves is the shift in the balance of behavior in the relationship. Positive or negative sentiment of one player isn’t what matters, it’s the asymmetry of the behavior of the two people in the relationship." The resulting system built by the Cornell researchers and their co-authors Srijan Kumar and Jordan Boyd-Graber was more accurate than humans at detecting future betrayal.

The article starts with the observation that "Whether it’s Katy Perry poaching dancers from once-BFF Taylor Swift or Clytemnestra orchestrating the murder of her husband Agamemnon, betrayal is a dark, persistent part of the human condition. Unlike garden-variety deception, betrayal happens in established relationships, destroying trust that has developed over time. It’s usually unexpected, and it yields a unique, often irreparable, wound. In fact, betrayers have a special place in hell, literarily: In Dante’s Inferno, they occupy the ninth and final circle; mere fraudsters dwell in the eighth."

Date Posted: 8/11/2015

KDD 2015 Best Student Paper Award

PhD student Wenlei Xie, David Bindel, Al Demers, and Johannes Gerhke have won one of two Best Student Paper awards at KDD 2015 for their paper "Edge-Weighted Personalized PageRank: Breaking A Decade-Old Performance Barrier". Their algorithm outperforms existing methods for computing PageRank on general graphs when the edge weights are personalized by nearly five orders of magnitude.

The ACM SIGKDD Conference on Knowledge Discovery and Data Mining is the largest and highest quality conference on Data Mining, Data Science, and Knowledge Discovery.

Date Posted: 8/11/2015


Archived News: