Events

Epidemic processes can model anything that spreads. As such, they are a useful tool for studying not only human diseases, but also network attacks, chains of activation in the brain, the propagation of real or fake news, the spread of viral tweets, and other processes. In this talk, we investigate epidemics spreading on a graph in the presence of various forms of uncertainty. We present in particular a result about controlling the spread of an epidemic when there is uncertainty about who exactly is infected. We show first that neither algorithms nor results are robust to uncertainty. In other words, uncertainty fundamentally changes how we must approach epidemics on graphs. We also present two related results about learning the graph underlying an epidemic process when there is uncertainty about when people were infected or what infected them.

Despite recent advances in natural language processing and generation, communication between humans and machines is in still its infancy. Existing intelligent home and mobile assistant technologies excel at scripted tasks such as weather or news reports and music control, yet typically fail at more advanced personalization. This calls for a centralized repository for personal knowledge about each user, which will then be a distant source of background knowledge for personalization in downstream applications. Such personal knowledge repository will be beneficial as a reusable asset; it should be both explainable and scrutable, giving full control to the owning user on editing and sharing stored information with selected service providers.

In this talk, I will discuss our efforts on automated personal knowledge extraction. We can easily obtain personal knowledge of famous people from biographies or news articles, however, such resources are not available for ordinary users. Hence, we turn to the task of inferring personal attributes from users' utterances in conversations, e.g., guessing a person's occupation from "I was sitting the whole day in front of my computer today, trying to finish a grant proposal for my research." I will highlight our Hidden Attribute Models (HAM) to solve the problem, a neural architecture leveraging attention mechanisms and embeddings, as well as an ongoing work on its extension to address challenging attributes such as hobbies and travel preferences with wide sets of multi-faceted attribute values. Finally, I will present an outlook on what we can further infer from users' activities, particularly in relation with their mood and emotion.

In the past, information and knowledge dissemination was relegated to the brick-and-mortar classrooms, newspapers, radio, and television. As these processes were simple and centralized, the models behind them were well understood and so were the empirical methods for optimizing them. In today's world, the internet and social media has become a powerful tool for information and knowledge dissemination: Wikipedia gets more than 1 million edits per day, Stack Overflow has more than 17 million questions, 25% of US population visits Yahoo! News for articles and discussions, Twitter has more than 60 million active monthly users, and Duolingo has 25 million users learning languages online.

These developments have introduced a paradigm shift in the process of dissemination. Not only has the nature of the task moved from being centralized to decentralized, but the developments have also blurred the boundary between the creator and the consumer of the content, i.e., information and knowledge. These changes have made it necessary to develop new models, which are better suited to understanding and analysing the dissemination, and to develop new methods to optimize them.

At a broad level, we can view the participation of users in the process of dissemination as falling in one of two settings: collaborative or competitive. In the collaborative setting, the participants work together in crafting knowledge online, e.g., by asking questions and contributing answers, or by discussing news or opinion pieces. In contrast, as competitors, they vie for the attention of their followers on social media. The first part of the thesis will propose models for the complexity of discussions and the evolution of expertise. The latter part of the thesis will explore the competitive setting where I will propose computational methods for measuring, and increasing, the attention received from followers on social media.

Module systems like that of Haskell permit only a weak form of modularity in which module implementations depend directly on other implementations and must be processed in dependency order. Module systems like that of ML, on the other hand, permit a stronger form of modularity in which explicit interfaces express assumptions about dependencies and each module can be typechecked and reasoned about independently.

In this thesis, I present Backpack, a new language for building separately-typecheckable packages on top of a weak module system like Haskell’s. The design of Backpack is the first to bring the rich world of type systems to the practical world of packages via mixin modules. It’s inspired by the MixML module calculus of Rossberg and Dreyer but by choosing practicality over expressivity Backpack both simplifies that semantics and supports a flexible notion of applicative instantiation. Moreover, this design is motivated less by foundational concerns and more by the practical concern of integration into Haskell. The result is a new approach to writing modular software at the scale of packages.

The semantics of Backpack is defined via elaboration into sets of Haskell modules and binary interface files, thus showing how Backpack maintains interoperability with Haskell while retrofitting it with interfaces. In my formalization of Backpack I present a novel type system for Haskell modules and I prove a key soundness theorem to validate Backpack’s semantics.

Cloud providers are nervous about recent research showing how Rowhammer attacks affect many types of DRAM including DDR4 and ECC-equipped DRAM.  Unfortunately, cloud providers lack a systematic way to test the DRAM present in their servers for the threat of a Rowhammer attack. Building such a methodology needs to overcome two difficult challenges: (1) devising a CPU instruction sequence that maximizes the rate of DRAM row activations on a given system, and (2) determining the adjacency of rows internal to DRAM. This talk will present an end-to-end methodology that overcomes these challenges to determine if cloud servers are susceptible to Rowhammer attacks. With our methodology, a cloud provider can construct worst-case testing conditions for DRAM.

We used our methodology to create worst-case DRAM testing conditions on the hardware used by a major cloud provider for a recent generation of its servers. Our findings show that none of the instruction sequences used in prior work to mount Rowhammer attacks create worst-case DRAM testing conditions. Instead, we construct an instruction sequence that issues non-explicit load and store instructions. Our new sequence leverages microarchitectural side-effects to ``hammer'' DRAM at a near-optimal rate on modern Skylake platforms. We also designed a DDR4 fault injector capable of reverse engineering row adjacency inside a DRAM device. When applied to our cloud provider's DIMMs, we find that rows inside DDR4 DRAM devices do not always follow a linear map.

Joint work with Lucian Cojocar (VU Amsterdam), Jeremie Kim, Minesh Patel, Onur Mutlu (ETH Zurich), Lily Tsai (MIT), and Alec Wolman (MSR)