The GraphLab project spanned almost a decade and had profound academic and industrial impact on large-scale machine learning and graph processing systems. There were numerous papers written describing the innovations in GraphLab including the original vertex-centric [8] and edge-centric [3] programming abstractions, high-performance asynchronous execution engines [9], out-of-core graph computation [6], tabular graph-systems [4], and even new statistical inference algorithms [2] enabled by the GraphLab project. This work became the basis of multiple PhD theses [1, 5, 7]. The GraphLab open-source project had broad academic and industrial adoption and ultimately lead to the launch of Turi. In this talk, we tell the story of GraphLab, how it began and the key ideas behind it. We will focus on the approach to achieving scalable asynchronous systems in machine learning. During our talk, we will explore the impact that GraphLab has had on the development of graph processing systems, graph databases, and AI/ML; Additionally, we will share our insights and opinions into where we see the future of these fields heading. In the process, we highlight some of the lessons we learned and provide guidance for future students.

In this paper, we present a purpose-built data management system, MLdp, for all machine learning (ML) datasets. ML applications pose some unique requirements different from common conventional data processing applications, including but not limited to: data lineage and provenance tracking, rich data semantics and formats, integration with diverse ML frameworks and access patterns, trial-and-error driven data exploration and evolution, rapid experimentation, reproducibility of the model training, strict compliance and privacy regulations, etc. Current ML systems/services, often named MLaaS, to-date focus on the ML algorithms, and offer no integrated data management system. Instead, they require users to bring their own data and to manage their own data on either blob storage or on file systems. The burdens of data management tasks, such as versioning and access control, fall onto the users, and not all compliance features, such as terms of use, privacy measures, and auditing, are available. MLdp offers a minimalist and flexible data model for all varieties of data, strong version management to guarantee re-producibility of ML experiments, and integration with major ML frameworks. MLdp also maintains the data provenance to help users track lineage and dependencies among data versions and models in their ML pipelines. In addition to table-stake features, such as security, availability and scalability, MLdp's internal design choices are strongly influenced by the goal to support rapid ML experiment iterations, which cycle through data discovery, data exploration, feature engineering, model training, model evaluation, and back to data discovery. The contributions of this paper are: 1) to recognize the needs and to call out the requirements of an ML data platform, 2) to share our experiences in building MLdp by adopting existing database technologies to the new problem as well as by devising new solutions, and 3) to call for actions from our communities on future challenges.

In this paper, we propose a novel method to efficiently compute the top-K most similar items given a query item, where similarity is defined by the set of items that have the highest vector inner products with the query. The task is related to the classical k-Nearest-Neighbor problem, and is widely applicable in a number of domains such as information retrieval, online advertising and collaborative filtering. Our method assumes an in-memory representation of the dataset and is designed to scale to query lengths of 100,000s of terms. Our algorithm uses a generalized Holder's inequality to upper bound the inner product with the norms of the constituent vectors. We also propose a novel compression scheme that computes bounds for groups of candidate items, thereby speeding up computation and minimizing memory requirements per query. We conduct extensive experiments on the publicly available Wikipedia dataset, and demonstrate that, with a memory overhead of 21%, our method can provide 1-3 orders of magnitude improvement in query run-time compared to naive methods and state of the art competing methods. Our median top-10 word query time is 25 us on 7.5 million words and 2.3 million documents.

While high-level data parallel frameworks, like MapReduce, simplify the design and implementation of large-scale data processing systems, they do not naturally or efficiently support many important data mining and machine learning algorithms and can lead to inefficient learning systems. To help fill this critical void, we introduced the GraphLab abstraction which naturally expresses asynchronous, dynamic, graph-parallel computation while ensuring data consistency and achieving a high degree of parallel performance in the shared-memory setting. In this paper, we extend the GraphLab framework to the substantially more challenging distributed setting while preserving strong data consistency guarantees. We develop graph based extensions to pipelined locking and data versioning to reduce network congestion and mitigate the effect of network latency. We also introduce fault tolerance to the GraphLab abstraction using the classic Chandy-Lamport snapshot algorithm and demonstrate how it can be easily implemented by exploiting the GraphLab abstraction itself. Finally, we evaluate our distributed implementation of the GraphLab abstraction on a large Amazon EC2 deployment and show 1-2 orders of magnitude performance gains over Hadoop-based implementations.

Historical user activity is key for building user profiles to predict the user behavior and affinities in many web applications such as targeting of online advertising, content personalization and social recommendations. User profiles are temporal, and changes in a user's activity patterns are particularly useful for improved prediction and recommendation. For instance, an increased interest in car-related web pages may well suggest that the user might be shopping for a new vehicle.In this paper we present a comprehensive statistical framework for user profiling based on topic models which is able to capture such effects in a fully \emph{unsupervised} fashion. Our method models topical interests of a user dynamically where both the user association with the topics and the topics themselves are allowed to vary over time, thus ensuring that the profiles remain current.

Content personalization is a key tool in creating attractive websites. Synergies can be obtained by integrating personalization between several Internet properties. In this paper we propose a hierarchical Bayesian model to address these issues. Our model allows the integration of multiple properties without changing the overall structure, which makes it easily extensible across large Internet portals. It relies at its lowest level on Latent Dirichlet Allocation, while making use of latent side features for cross-property integration. We demonstrate the efficiency of our approach by analyzing data from several properties of a major Internet portal.