As more and more online search queries come from voice, automatic speech recognition becomes a key component to deliver relevant search results. Errors introduced by automatic speech recognition (ASR) lead to irrelevant search results returned to the user, thus causing user dissatisfaction. In this paper, we introduce an approach, "Mondegreen", to correct voice queries in text space without depending on audio signals, which may not always be available due to system constraints or privacy or bandwidth (for example, some ASR systems run on-device) considerations. We focus on voice queries transcribed via several proprietary commercial ASR systems. These queries come from users making internet, or online service search queries. We first present an analysis showing how different the language distribution coming from user voice queries is from that in traditional text corpora used to train off-the-shelf ASR systems. We then demonstrate that Mondegreen can achieve significant improvements in increased user interaction by correcting user voice queries in one of the largest search systems in Google. Finally, we see Mondegreen as complementing existing highly-optimized production ASR systems, which may not be frequently retrained and thus lag behind due to vocabulary drifts.

Many recent advances in neural information retrieval models, which predict top-K items given a query, learn directly from a large training set of (query, item) pairs. However, they are often insufficient when there are many previously unseen (query, item) combinations, often referred to as the cold start problem. Furthermore, the search system can be biased towards items that are frequently shown to a query previously, also known as the "rich get richer" (a.k.a. feedback loop) problem. In light of these problems, we observed that most online content platforms have both a search and a recommender system that, while having heterogeneous input spaces, can be connected through their common output item space and a shared semantic representation. In this paper, we propose a new Zero-Shot Heterogeneous Transfer Learning framework that transfers learned knowledge from the recommender system component to improve the search component of a content platform. First, it learns representations of items and their natural-language features by predicting (item, item) correlation graphs derived from the recommender system as an auxiliary task. Then, the learned representations are transferred to solve the target search retrieval task, performing query-to-item prediction without having seen any (query, item) pairs in training. We conduct online and offline experiments on one of the world's largest search and recommender systems from Google, and present the results and lessons learned. We demonstrate that the proposed approach can achieve high performance on offline search retrieval tasks, and more importantly, achieved significant improvements on relevance and user interactions over the highly-optimized production system in online experiments.

Reinforcement learning methods for recommender systems optimize recommendations for long-term user engagement. However, since users are often presented with slates of multiple items---which may have interacting effects on user choice---methods are required to deal with the combinatorics of the RL action space. We develop SlateQ, a decomposition of value-based temporal-difference and Q-learning that renders RL tractable with slates. Under mild assumptions on user choice behavior, we show that the long-term value (LTV) of a slate can be decomposed into a tractable function of its component item-wise LTVs. We demonstrate our methods in simulation, and validate the scalability and effectiveness of decomposed TD-learning on YouTube.

Most practical recommender systems focus on estimating immediate user engagement without considering the long-term effects of recommendations on user behavior. Reinforcement learning (RL) methods offer the potential to optimize recommendations for long-term user engagement. However, since users are often presented with slates of multiple items - which may have interacting effects on user choice - methods are required to deal with the combinatorics of the RL action space. In this work, we address the challenge of making slate-based recommendations to optimize long-term value using RL. Our contributions are three-fold. (i) We develop SLATEQ, a decomposition of value-based temporal-difference and Q-learning that renders RL tractable with slates. Under mild assumptions on user choice behavior, we show that the long-term value (LTV) of a slate can be decomposed into a tractable function of its component item-wise LTVs. (ii) We outline a methodology that leverages existing myopic learning-based recommenders to quickly develop a recommender that handles LTV. (iii) We demonstrate our methods in simulation, and validate the scalability of decomposed TD-learning using SLATEQ in live experiments on YouTube.

Generalized linear models with nonlinear feature transformations are widely used for large-scale regression and classification problems with sparse inputs. Memorization of feature interactions through a wide set of cross-product feature transformations are effective and interpretable, while generalization requires more feature engineering effort. With less feature engineering, deep neural networks can generalize better to unseen feature combinations through low-dimensional dense embeddings learned for the sparse features. However, deep neural networks with embeddings can over-generalize and recommend less relevant items when the user-item interactions are sparse and high-rank. In this paper, we present Wide & Deep learning---jointly trained wide linear models and deep neural networks---to combine the benefits of memorization and generalization for recommender systems. We productionized and evaluated the system on Google Play, a commercial mobile app store with over one billion active users and over one million apps. Online experiment results show that Wide & Deep significantly increased app acquisitions compared with wide-only and deep-only models. We have also open-sourced our implementation in TensorFlow.

The problem. We consider the problem of implementing a consistent replicated object in a partially synchronous message passing distributed system susceptible to process and communication failures. The object is a generic shared resource, such as a data structure, a file, or a lock. The processes implementing the replicated object access it by applying operations to it at unpredictable times and potentially concurrently.1 The object should be linearizable: it should behave as if each operation applied to it takes effect at a distinct instant in time during the interval between its invocation and its response. The main reason for replicating an object is fault tolerance; if a copy of the object becomes inaccessible, the object can still be accessed through the other copies. Another reason for replicating an object is performance; a process that requires the object can access its local copy. The benefit of accessing a local copy, however, must be weighed against the cost of keeping the copies consistent.

This talk will focus on our experience in managing the complexity of Sibyl, a large scale machine learning system that is widely used within Google. We believe that a large fraction of the challenges faced by Sibyl are inherent to large scale production machine learning and that other production systems are likely to encounter them as well [1]. Thus, these challenges present interesting opportunities for future research. The Sibyl system is complex for a number of reasons. We have learnt that a complete end-to-end machine learning solution has to have subsystems to address a variety of different needs: data ingestion, data analysis, data verification, experimentation, model analysis, model serving, configuration, data transformations, support for different kinds of loss functions and modeling, machine learning algorithm implementations, etc. Machine learning algorithms themselves constitute a relatively small fraction of the overall system. Each subsystem consists of a number of distinct components to support the variety of product needs. For example, Sibyl supports more than 5 different model serving systems, each with its own idiosyncrasies and challenges. In addition, Sibyl configuration contains more lines of code than the core Sibyl learner itself. Finally existing solutions for some of the challenges don't feel adequate and we believe these challenges present opportunities for future research. Though the overall system is complex, our users need to be able to deploy solutions quickly. This is because a machine learning deployment is typically an iterative process of model improvements. At each iteration, our users experiment with new features, find those that improve the model's prediction capability, and then "launch" a new model with those improved features. A user may go through 10 or more such productive launches. Not only is speed of iteration crucial to our users, but they are often willing to sacrifice the improved prediction quality of a high quality but cumbersome system for the speed of iteration of a lower quality but nimble system. In this talk I will give an example of how simplification drives systems design and sometimes the design of novel algorithms.

Page 1. Tushar Chandra, Eugene Ie, Kenneth Goldman, Tomas Lloret Llinares, Jim McFadden, Fernando Pereira, Joshua Redstone, Tal Shaked, Yoram Singer Sibyl: a system for large scale machine learning Page 2. Machine Learning Background Use the past to predict the future Core technology for internet-based prediction tasks Examples of problems that can be solved with machine learning: • Classify email as spam or not • Estimate relevance of an impression in context: • Search, advertising, videos, etc. • Rank candidate impressions The internet adds a scaling challenge: • 100s of millions of users interacting every day • Good solutions require a mix of theory and systems Page 3. Overview of Results Built a large scale machine learning system: • Used recently developed machine learning algorithm • Algorithms have provable convergence & quality guarantees …

We describe efficient algorithms for projecting a vector onto the l1-ball. We present two methods for projection. The first performs exact projection in O(n) expected time, where n is the dimension of the space. The second works on vectors k of whose elements are perturbed outside the l1-ball, projecting in O(k log(n)) time. This setting is especially useful for online learning in sparse feature spaces such as text categorization applications. We demonstrate the merits and effectiveness of our algorithms in numerous batch and online learning tasks. We show that variants of stochastic gradient projection methods augmented with our efficient projection procedures outperform interior point methods, which are considered state-of-the-art optimization techniques. We also show that in online settings gradient updates with l1 projections outperform the exponentiated gradient algorithm while obtaining models with high degrees of sparsity.

Bigtable is a distributed storage system for managing structured data that is designed to scale to a very large size: petabytes of data across thousands of commodity servers. Many projects at Google store data in Bigtable, including web indexing, Google Earth, and Google Finance. These applications place very different demands on Bigtable, both in terms of data size (from URLs to web pages to satellite imagery) and latency requirements (from backend bulk processing to real-time data serving). Despite these varied demands, Bigtable has successfully provided a flexible, high-performance solution for all of these Google products. In this article, we describe the simple data model provided by Bigtable, which gives clients dynamic control over data layout and format, and we describe the design and implementation of Bigtable.

We describe our experience in building a fault-tolerant data-base using the Paxos consensus algorithm. Despite the existing literature in the field, building such a database proved to be non-trivial. We describe selected algorithmic and engineering problems encountered, and the solutions we found for them. Our measurements indicate that we have built a competitive system.