Mathias Lécuyer

Dispatch: secure, resilient mobile reporting

Kanak Biscuitwala kanak@cs.stanford.edu Willem Bult wbult@stanford.edu Mathias Lecuyer† ml3302@columbia.edu T.J. Purtell tpurtell@cs.stanford.edu Madeline K.B. Ross‡ mkr2132@columbia.edu Augustin Chaintreau† augustin@cs.columbia.edu Chris Haseman§ haseman@tumblr.com Monica S. Lam lam@cs.stanford.edu Susan E. McGregor‡ sem2196@columbia.edu Computer Science Department, Stanford University †Computer Science Department, Columbia University ‡Graduate School of Journalism, Columbia University §Tumblr, Inc.

Enhancing Selectivity in Big Data

Today’s companies collect immense amounts of personal data and enable wide access to it within the company. This exposes the data to external hackers and privacy-transgressing employees. This study shows that, for a wide and important class of workloads, only a fraction of the data is needed to approach state-of-the-art accuracy. We propose selective data systems that are designed to pinpoint the data that is valuable for a company’s current and evolving workloads. These systems limit data exposure by setting aside the data that is not truly valuable.

Harvesting Randomness to Optimize Distributed Systems

We view randomization through the lens of statistical machine learning: as a powerful resource for offline optimization. Cloud systems make randomized decisions all the time (e.g., in load balancing), yet this randomness is rarely used for optimization after-the-fact. By casting system decisions in the framework of reinforcement learning, we show how to collect data from existing systems, without modifying them, to evaluate new policies, without deploying them. Our methodology, called harvesting randomness, has the potential to accurately estimate a policys performance without the risk or cost of deploying it on live traffic. We quantify this optimization power and apply it to a real machine health scenario in Azure Compute. We also apply it to two prototyped scenarios, for load balancing (Nginx) and caching (Redis), with much less success, and use them to identify the systems and machine learning challenges to achieving our goal. Our long-term agenda is to harvest the randomness in distributed systems to develop non-invasive and efficient techniques for optimizing them. Like CPU cycles and bandwidth, we view randomness as a valuable resource being wasted by the cloud, and we seek to remedy this.

Web Transparency for Complex Targeting: Algorithms, Limits, and Tradeoffs

Big Data promises important societal progress but exacerbates the need for due process and accountability. Companies and institutions can now discriminate between users at an individual level using collected data or past behavior. Worse, today they can do so in near perfect opacity. The nascent field of web transparency aims to develop the tools and methods necessary to reveal how information is used, however today it lacks robust tools that let users and investigators identify targeting using multiple inputs. In this paper, we formalize for the first time the problem of detecting and identifying targeting on combinations of inputs and provide the first algorithm that is asymptotically exact. This algorithm is designed to serve as a theoretical foundational block to build future scalable and robust web transparency tools. It offers three key properties. First, our algorithm is service agnostic and applies to a variety of settings under a broad set of assumptions. Second, our algorithms analysis delineates a theoretical detection limit that characterizes which forms of targeting can be distinguished from noise and which cannot. Third, our algorithm establishes fundamental tradeoffs that lead the way to new metrics for the science of web transparency.