In the realm of recommender systems, the ubiquitous adoption of deep neural networks has emerged as a dominant paradigm for modeling diverse business objectives. As user bases continue to expand, the necessity of personalization and frequent model updates have assumed paramount significance to ensure the delivery of relevant and refreshed experiences to a diverse array of members. In this work, we introduce an innovative meta-learning solution tailored to the personalization of models for individual members and other entities, coupled with the frequent updates based on the latest user interaction signals. Specifically, we leverage the Model-Agnostic Meta Learning (MAML) algorithm to adapt per-task sub-networks using recent user interaction data. Given the near infeasibility of productionizing original MAML-based models in online recommendation systems, we propose an efficient strategy to operationalize meta-learned sub-networks in production, which involves transforming them into fixed-sized vectors, termed meta embeddings, thereby enabling the seamless deployment of models with hundreds of billions of parameters for online serving. Through extensive experimentation on production data drawn from various applications at LinkedIn, we demonstrate that the proposed solution consistently outperforms the best performing baseline models of those applications, including strong baselines such as using wide-and-deep ID based personalization approach. Our approach has enabled the deployment of a range of highly personalized AI models across diverse LinkedIn applications, leading to substantial improvements in business metrics as well as refreshed experience for our members.

Social media and technology have drastically transformed the social and information networks around us. They have impacted how we communicate with others, search for information, and even how we express our personal opinions. Further, in this era of big data, not only are the online services collecting vast variety of user data, but we, as users, are also readily divulging significant amounts of information. Together, massive datasets obtained from diverse sources such as organizations and user generated content give us the opportunity to explore and understand complex behavior of both individuals and communities. This proposal aims at designing generalizable and scalable data-driven frameworks to gain a deeper understanding of the users, explain their actions and preferences, and infer personal traits. The proposed models will enable us to move beyond asking the conventional questions of who and what, and reveal answers about how and why. Given the varying digital persona of users motivated by their personal preferences and social attributes, we characterize users in two distinct domains: online health and peace studies. The models are designed to solve various real-world challenges to maximize their broader impact.

Stochastic Gradient Descent (SGD) is a simple yet very efficient online learning algorithm for optimizing convex (and often non-convex) functions and one of the most popular stochastic optimization methods in machine learning today. One drawback of SGD is that it is sensitive to the learning rate hyper-parameter. The Adaptive Sub-gradient Descent, AdaGrad, dynamically incorporates knowledge of the geometry of the data observed in earlier iterations to calculate a different learning rate for every feature. In this work, we implement a distributed version of AdaGrad for large-scale machine learning tasks using Apache Spark. Apache Spark is a fast cluster computing engine that provides similar scalability and fault tolerance properties to MapReduce, but in contrast to Hadoop's two-stage disk-based MapReduce paradigm, Spark's multi-stage in-memory primitives allow user programs to load data into a cluster's memory and query it repeatedly, which makes it ideal for building scalable machine learning applications. We empirically evaluate our implementation on large-scale real-world problems in the machine learning canonical tasks of classification and regression. Comparing our implementation of AdaGrad with the SGD scheduler currently available in Spark's Machine Learning Library (MLlib), we experimentally show that AdaGrad saves time by avoiding manually setting a learning-rate hyperparameter, converges fast and can even achieve better generalization errors.

As robots enter human spaces, unique perception challenges are emerging. Sensing human activity, adapting to highly dynamic environments, and acting coherently and contingently is challenging when robots transition from structured environments to human-centric ones. We approach this problem by employing context-based perception, a biologically-inspired, low-cost approach to sensing that leverages noisy, global features. Across several months, our mobile robot collected real-world, multimodal data from multi-use locations; where the same space might be used for many different activities. We then ran a series of unimodal and multimodal classification experiments. We successfully classified several aspects of situational context from noisy data, and, to our knowledge are the first group to do so. This work represents an important step toward enabling robots that can readily leverage context to solve perceptual tasks.