Explore the latest research from Facebook

In order to construct accurate proposers for Metropolis-Hastings Markov Chain Monte Carlo, we integrate ideas from probabilistic graphical models and neural networks in a framework we call Lightweight Inference Compilation (LIC). LIC implements amortized inference within an open-universe declarative probabilistic programming language (PPL).

We focus on Internet advertising auctions, fair division problems, content recommendation systems, and robust abstractions of large-scale markets.

Although there have been numerous studies about underrepresentation and misrepresentation of people in advertising, most have focused on traditional channels such as television, print, and radio, rather than on digital channels. In this paper, we seek to contribute to the body of knowledge by utilizing a mix of quantitative and qualitative methods to explore people’s attitudes toward diversity in online advertising, the current state of representation, and the impact of diversity on digital campaign performance.

In this paper, we present a novel approach for using aggregated and anonymized Facebook location data to measure displacement patterns in the weeks and months after disasters.

Motivated by the needs of online large-scale recommender systems, we specialize the decoupled extended Kalman filter (DEKF) to factorization models, including factorization machines, matrix and tensor factorization, and illustrate the effectiveness of the approach through numerical experiments on synthetic and on real-world data.

We explore how social network exposure to COVID-19 cases shapes individuals’ social distancing behavior during the early months of the ongoing pandemic. We work with de-identified data from Facebook to show that U.S. users whose friends live in areas with worse coronavirus outbreaks reduce their mobility more than otherwise similar users whose friends live in areas with smaller outbreaks.

Using the recently deployed Facebook Blood Donation tool, we conduct the first large-scale algorithmic matching of blood donors with donation opportunities. In both simulations and real experiments we match potential donors with opportunities, guided by a machine learning model trained on prior observations of donor behavior.

Our empirical evaluation demonstrates that qEHVI is computationally tractable in many practical scenarios and outperforms state-of-the-art multi-objective BO algorithms at a fraction of their wall time.

We develop effective models that leverage the structure of the search space to enable contextual policy optimization directly from the aggregate rewards using Bayesian optimization.

We outline a way to deploy a privacy-preserving protocol for multiparty Randomized Controlled Trials on the scale of 500 million rows of data and more than a billion gates.