A common applied statistics task involves building regression models to characterize non-linear relationships between variables. When we write a function that takes continuous values as inputs, we are essentially implying an infinite vector that only returns values (indexed by the inputs) when the function is called upon to do so. To make this notion of a "distribution over functions" more concrete, let's quickly demonstrate how we obtain realizations from a Gaussian process, which result in an evaluation of a function over a set of points. We are going generate realizations sequentially, point by point, using the lovely conditioning property of mutlivariate Gaussian distributions.

As a single host has limited storage and compute resources, our storage systems shard data items over multiple hosts and our batch jobs execute over clusters of thousands of workers, to scale and speed-up the computation. Our VLDB'17 paper, Social Hash Partitioner: A Scalable Distributed Hypergraph Partitioner, describes a new method for partitioning bipartite graphs while minimizing fan-out. We describe the resulting framework as a Social Hash Partitioner (SHP) because it can be used as the hypergraph partitioning component of the Social Hash framework introduced in our earlier NSDI'16 paper. The fan-out reduction model is applicable to many infrastructure optimization problems at Facebook, like data sharding, query routing and index compression.

In my opinion, the marriage of the leading professional social network and the world's largest software company demonstrates that we are decidedly at the start of a new era in software, where proprietary data is king, and will start to come bundled together with software. We've seen this rise in the consumer realm, where technology companies are fundamentally aggregating and analyzing user behavior, and providing value back to users (and, of course, advertisers.) There are countless other examples that also demonstrate that consumer technology puts behavioral and user data front and center, in a way that I expect we will start to see from the enterprise as the divide between these two segments starts to collapse. Taken together, this demonstrates that proven machine learning algorithms have both the horsepower and access to granular datasets that are unprecedented.

Intelligent systems that learn interactively from their end-users are quickly becoming widespread. Until recently, this progress has been fueled mostly by advances in machine learning; however, more and more researchers are realizing the importance of studying users of these systems. We present a number of case studies that characterize the impact of interactivity, demonstrate ways in which some existing systems fail to account for the user, and explore new ways for learning systems to interact with their users. We argue that the design process for interactive machine learning systems should involve users at all stages: explorations that reveal human interaction patterns and inspire novel interaction methods, as well as refinement stages to tune details of the interface and choose among alternatives.

The emerging literature on combined systems is directed at domains where the prediction of human behavior is not required. Professionals concerned with human outcomes make decisions that are intuitive or analytic or some combination of both. Justifications and methodology are presented for combining analytic and intuitive agents in an expert system that supports professional decision making. The system presented demonstrates the challenges and opportunities inherent in developing and using AI-collaborative technology to solve social problems.