arXiv subjects

Eytan Bakshy

Publications and source records attributed to Eytan Bakshy.

48 records · Page 3Linked to original sources

arXiv2020

Real-world Video Adaptation with Reinforcement Learning

Client-side video players employ adaptive bitrate (ABR) algorithms to optimize user quality of experience (QoE). We evaluate recently proposed RL-based ABR methods in Facebook's web-based video streaming platform. Real-world ABR contains several challenges that requires customized designs beyond off-the-shelf RL algorithms -- we implement a scalable neural network architecture that supports videos with arbitrary bitrate encodings; we design a training method to cope with the variance resulting from the stochasticity in network conditions; and we leverage constrained Bayesian optimization for reward shaping in order to optimize the conflicting QoE objectives. In a week-long worldwide deployment with more than 30 million video streaming sessions, our RL approach outperforms the existing human-engineered ABR algorithms.

cs.NI↗

arXiv2019

Thompson Sampling for Contextual Bandit Problems with Auxiliary Safety Constraints

Recent advances in contextual bandit optimization and reinforcement learning have garnered interest in applying these methods to real-world sequential decision making problems. Real-world applications frequently have constraints with respect to a currently deployed policy. Many of the existing constraint-aware algorithms consider problems with a single objective (the reward) and a constraint on the reward with respect to a baseline policy. However, many important applications involve multiple competing objectives and auxiliary constraints. In this paper, we propose a novel Thompson sampling algorithm for multi-outcome contextual bandit problems with auxiliary constraints. We empirically evaluate our algorithm on a synthetic problem. Lastly, we apply our method to a real world video transcoding problem and provide a practical way for navigating the trade-off between safety and performance using Bayesian optimization.

cs.LG↗

arXiv2019

PlanAlyzer: Assessing Threats to the Validity of Online Experiments

Online experiments are ubiquitous. As the scale of experiments has grown, so has the complexity of their design and implementation. In response, firms have developed software frameworks for designing and deploying online experiments. Ensuring that experiments in these frameworks are correctly designed and that their results are trustworthy---referred to as *internal validity*---can be difficult. Currently, verifying internal validity requires manual inspection by someone with substantial expertise in experimental design. We present the first approach for statically checking the internal validity of online experiments. Our checks are based on well-known problems that arise in experimental design and causal inference. Our analyses target PlanOut, a widely deployed, open-source experimentation framework that uses a domain-specific language to specify and run complex experiments. We have built a tool, PlanAlyzer, that checks PlanOut programs for a variety of threats to internal validity, including failures of randomization, treatment assignment, and causal sufficiency. PlanAlyzer uses its analyses to automatically generate *contrasts*, a key type of information required to perform valid statistical analyses over experimental results. We demonstrate PlanAlyzer's utility on a corpus of PlanOut scripts deployed in production at Facebook, and we evaluate its ability to identify threats to validity on a mutated subset of this corpus. PlanAlyzer has both precision and recall of 92% on the mutated corpus, and 82% of the contrasts it automatically generates match hand-specified data.

cs.PL↗

arXiv2019

Bayesian Optimization for Policy Search via Online-Offline Experimentation

Online field experiments are the gold-standard way of evaluating changes to real-world interactive machine learning systems. Yet our ability to explore complex, multi-dimensional policy spaces - such as those found in recommendation and ranking problems - is often constrained by the limited number of experiments that can be run simultaneously. To alleviate these constraints, we augment online experiments with an offline simulator and apply multi-task Bayesian optimization to tune live machine learning systems. We describe practical issues that arise in these types of applications, including biases that arise from using a simulator and assumptions for the multi-task kernel. We measure empirical learning curves which show substantial gains from including data from biased offline experiments, and show how these learning curves are consistent with theoretical results for multi-task Gaussian process generalization. We find that improved kernel inference is a significant driver of multi-task generalization. Finally, we show several examples of Bayesian optimization efficiently tuning a live machine learning system by combining offline and online experiments.

stat.ML↗

arXiv2019

Shrinkage Estimators in Online Experiments

We develop and analyze empirical Bayes Stein-type estimators for use in the estimation of causal effects in large-scale online experiments. While online experiments are generally thought to be distinguished by their large sample size, we focus on the multiplicity of treatment groups. The typical analysis practice is to use simple differences-in-means (perhaps with covariate adjustment) as if all treatment arms were independent. In this work we develop consistent, small bias, shrinkage estimators for this setting. In addition to achieving lower mean squared error these estimators retain important frequentist properties such as coverage under most reasonable scenarios. Modern sequential methods of experimentation and optimization such as multi-armed bandit optimization (where treatment allocations adapt over time to prior responses) benefit from the use of our shrinkage estimators. Exploration under empirical Bayes focuses more efficiently on near-optimal arms, improving the resulting decisions made under uncertainty. We demonstrate these properties by examining seventeen large-scale experiments conducted on Facebook from April to June 2017.

stat.ME↗

arXiv2018

Constrained Bayesian Optimization with Noisy Experiments

Randomized experiments are the gold standard for evaluating the effects of changes to real-world systems. Data in these tests may be difficult to collect and outcomes may have high variance, resulting in potentially large measurement error. Bayesian optimization is a promising technique for efficiently optimizing multiple continuous parameters, but existing approaches degrade in performance when the noise level is high, limiting its applicability to many randomized experiments. We derive an expression for expected improvement under greedy batch optimization with noisy observations and noisy constraints, and develop a quasi-Monte Carlo approximation that allows it to be efficiently optimized. Simulations with synthetic functions show that optimization performance on noisy, constrained problems outperforms existing methods. We further demonstrate the effectiveness of the method with two real-world experiments conducted at Facebook: optimizing a ranking system, and optimizing server compiler flags.

stat.ML↗

arXiv2017

Uncertainty in Online Experiments with Dependent Data: An Evaluation of Bootstrap Methods

Many online experiments exhibit dependence between users and items. For example, in online advertising, observations that have a user or an ad in common are likely to be associated. Because of this, even in experiments involving millions of subjects, the difference in mean outcomes between control and treatment conditions can have substantial variance. Previous theoretical and simulation results demonstrate that not accounting for this kind of dependence structure can result in confidence intervals that are too narrow, leading to inaccurate hypothesis tests. We develop a framework for understanding how dependence affects uncertainty in user-item experiments and evaluate how bootstrap methods that account for differing levels of dependence perform in practice. We use three real datasets describing user behaviors on Facebook - user responses to ads, search results, and News Feed stories - to generate data for synthetic experiments in which there is no effect of the treatment on average by design. We then estimate empirical Type I error rates for each bootstrap method. Accounting for dependence within a single type of unit (i.e., within-user dependence) is often sufficient to get reasonable error rates. But when experiments have effects, as one might expect in the field, accounting for multiple units with a multiway bootstrap can be necessary to get close to the advertised Type I error rates. This work provides guidance to practitioners evaluating large-scale experiments, and highlights the importance of analysis of inferential methods for dependence structures common to online systems.

stat.ME↗

arXiv2017

Bias and high-dimensional adjustment in observational studies of peer effects

Peer effects, in which the behavior of an individual is affected by the behavior of their peers, are posited by multiple theories in the social sciences. Other processes can also produce behaviors that are correlated in networks and groups, thereby generating debate about the credibility of observational (i.e. nonexperimental) studies of peer effects. Randomized field experiments that identify peer effects, however, are often expensive or infeasible. Thus, many studies of peer effects use observational data, and prior evaluations of causal inference methods for adjusting observational data to estimate peer effects have lacked an experimental "gold standard" for comparison. Here we show, in the context of information and media diffusion on Facebook, that high-dimensional adjustment of a nonexperimental control group (677 million observations) using propensity score models produces estimates of peer effects statistically indistinguishable from those from using a large randomized experiment (220 million observations). Naive observational estimators overstate peer effects by 320% and commonly used variables (e.g., demographics) offer little bias reduction, but adjusting for a measure of prior behaviors closely related to the focal behavior reduces bias by 91%. High-dimensional models adjusting for over 3,700 past behaviors provide additional bias reduction, such that the full model reduces bias by over 97%. This experimental evaluation demonstrates that detailed records of individuals' past behavior can improve studies of social influence, information diffusion, and imitation; these results are encouraging for the credibility of some studies but also cautionary for studies of rare or new behaviors. More generally, these results show how large, high-dimensional data sets and statistical learning techniques can be used to improve causal inference in the behavioral sciences.

stat.ME↗

arXiv2014

Designing and Deploying Online Field Experiments

Online experiments are widely used to compare specific design alternatives, but they can also be used to produce generalizable knowledge and inform strategic decision making. Doing so often requires sophisticated experimental designs, iterative refinement, and careful logging and analysis. Few tools exist that support these needs. We thus introduce a language for online field experiments called PlanOut. PlanOut separates experimental design from application code, allowing the experimenter to concisely describe experimental designs, whether common "A/B tests" and factorial designs, or more complex designs involving conditional logic or multiple experimental units. These latter designs are often useful for understanding causal mechanisms involved in user behaviors. We demonstrate how experiments from the literature can be implemented in PlanOut, and describe two large field experiments conducted on Facebook with PlanOut. For common scenarios in which experiments are run iteratively and in parallel, we introduce a namespaced management system that encourages sound experimental practice.

cs.HC↗

arXiv2013

Selection Effects in Online Sharing: Consequences for Peer Adoption

Most models of social contagion take peer exposure to be a corollary of adoption, yet in many settings, the visibility of one's adoption behavior happens through a separate decision process. In online systems, product designers can define how peer exposure mechanisms work: adoption behaviors can be shared in a passive, automatic fashion, or occur through explicit, active sharing. The consequences of these mechanisms are of substantial practical and theoretical interest: passive sharing may increase total peer exposure but active sharing may expose higher quality products to peers who are more likely to adopt. We examine selection effects in online sharing through a large-scale field experiment on Facebook that randomizes whether or not adopters share Offers (coupons) in a passive manner. We derive and estimate a joint discrete choice model of adopters' sharing decisions and their peers' adoption decisions. Our results show that active sharing enables a selection effect that exposes peers who are more likely to adopt than the population exposed under passive sharing. We decompose the selection effect into two distinct mechanisms: active sharers expose peers to higher quality products, and the peers they share with are more likely to adopt independently of product quality. Simulation results show that the user-level mechanism comprises the bulk of the selection effect. The study's findings are among the first to address downstream peer effects induced by online sharing mechanisms, and can inform design in settings where a surplus of sharing could be viewed as costly.

cs.SI↗

arXiv2012

Social Influence in Social Advertising: Evidence from Field Experiments

Social advertising uses information about consumers' peers, including peer affiliations with a brand, product, organization, etc., to target ads and contextualize their display. This approach can increase ad efficacy for two main reasons: peers' affiliations reflect unobserved consumer characteristics, which are correlated along the social network; and the inclusion of social cues (i.e., peers' association with a brand) alongside ads affect responses via social influence processes. For these reasons, responses may be increased when multiple social signals are presented with ads, and when ads are affiliated with peers who are strong, rather than weak, ties. We conduct two very large field experiments that identify the effect of social cues on consumer responses to ads, measured in terms of ad clicks and the formation of connections with the advertised entity. In the first experiment, we randomize the number of social cues present in word-of-mouth advertising, and measure how responses increase as a function of the number of cues. The second experiment examines the effect of augmenting traditional ad units with a minimal social cue (i.e., displaying a peer's affiliation below an ad in light grey text). On average, this cue causes significant increases in ad performance. Using a measurement of tie strength based on the total amount of communication between subjects and their peers, we show that these influence effects are greatest for strong ties. Our work has implications for ad optimization, user interface design, and central questions in social science research.

cs.SI↗

arXiv2012

The Role of Social Networks in Information Diffusion

Online social networking technologies enable individuals to simultaneously share information with any number of peers. Quantifying the causal effect of these technologies on the dissemination of information requires not only identification of who influences whom, but also of whether individuals would still propagate information in the absence of social signals about that information. We examine the role of social networks in online information diffusion with a large-scale field experiment that randomizes exposure to signals about friends' information sharing among 253 million subjects in situ. Those who are exposed are significantly more likely to spread information, and do so sooner than those who are not exposed. We further examine the relative role of strong and weak ties in information propagation. We show that, although stronger ties are individually more influential, it is the more abundant weak ties who are responsible for the propagation of novel information. This suggests that weak ties may play a more dominant role in the dissemination of information online than currently believed.

cs.SI↗