Mahmoud Abo Khamis

Joins via Geometric Resolutions: Worst-case and Beyond

We present a simple geometric framework for the relational join. Using this framework, we design an algorithm that achieves the fractional hypertree-width bound, which generalizes classical and recent worst-case algorithmic results on computing joins. In addition, we use our framework and the same algorithm to show a series of what are colloquially known as beyond worst-case results. The framework allows us to prove results for data stored in Btrees, multidimensional data structures, and even multiple indices per table. A key idea in our framework is formalizing the inference one does with an index as a type of geometric resolution; transforming the algorithmic problem of computing joins to a geometric problem. Our notion of geometric resolution can be viewed as a geometric analog of logical resolution. In addition to the geometry and logic connections, our algorithm can also be thought of as backtracking search with memoization.

Computing Join Queries with Functional Dependencies

Recently, Gottlob, Lee, Valiant, and Valiant (GLVV) presented an output size bound for join queries with functional dependencies (FD), based on a linear program on polymatroids. GLVV bound strictly generalizes the bound of Atserias, Grohe and Marx (AGM) for queries with no FD, in which case there are known algorithms running within the AGM-bound and thus are worst-case optimal. A main result of this paper is an algorithm for computing join queries with FDs, running within GLVV bound up to a poly-log factor. In particular, our algorithm is worst-case optimal for any query where the GLVV bound is tight. As an unexpected by-product, our algorithm manages to solve a harder problem, where (some) input relations may have prescribed maximum degree bounds, of which both functional dependencies and cardinality bounds are special cases. We extend Gottlob et al. framework by replacing all variable subsets with the lattice of closed sets (under the given FDs). This gives us new insights into the structure of the worst-case bound and worst-case instances. While it is still open whether GLVV bound is tight in general, we show that it is tight on distributive lattices and some other simple lattices. Distributive lattices capture a strict superset of queries with no FD and with simple FDs. We also present two simpler algorithms which are also worst-case optimal on distributive lattices within a single-log factor, but they do not match GLVV bound on a general lattice. Our algorithms are designed based on a novel principle: we turn a proof of a polymatroid-based output size bound into an algorithm.

In-Database Learning with Sparse Tensors

In-database analytics is of great practical importance as it avoids the costly repeated loop data scientists have to deal with on a daily basis: select features, export the data, convert data format, train models using an external tool, reimport the parameters. It is also a fertile ground of theoretically fundamental and challenging problems at the intersection of relational and statistical data models. This paper introduces a unified framework for training and evaluating a class of statistical learning models inside a relational database. This class includes ridge linear regression, polynomial regression, factorization machines, and principal component analysis. We show that, by synergizing key tools from relational database theory such as schema information, query structure, recent advances in query evaluation algorithms, and from linear algebra such as various tensor and matrix operations, one can formulate in-database learning problems and design efficient algorithms to solve them. The algorithms and models proposed in the paper have already been implemented and deployed in retail-planning and forecasting applications, with significant performance benefits over out-of-database solutions that require the costly data-export loop.

AC/DC: In-Database Learning Thunderstruck

We report on the design and implementation of the AC/DC gradient descent solver for a class of optimization problems over normalized databases. AC/DC decomposes an optimization problem into a set of aggregates over the join of the database relations. It then uses the answers to these aggregates to iteratively improve the solution to the problem until it converges. The challenges faced by AC/DC are the large database size, the mixture of continuous and categorical features, and the large number of aggregates to compute. AC/DC addresses these challenges by employing a sparse data representation, factorized computation, problem reparameterization under functional dependencies, and a data structure that supports shared computation of aggregates. To train polynomial regression models and factorization machines of up to 154K features over the natural join of all relations from a real-world dataset of up to 86M tuples, AC/DC needs up to 30 minutes on one core of a commodity machine. This is up to three orders of magnitude faster than its competitors R, MadLib, libFM, and TensorFlow whenever they finish and thus do not exceed memory limitation, 24-hour timeout, or internal design limitations.

Juggling Functions Inside a Database

We define and study the Functional Aggregate Query (FAQ) problem, which captures common computational tasks across a very wide range of domains including relational databases, logic, matrix and tensor computation, probabilistic graphical models, constraint satisfaction, and signal processing. Simply put, an FAQ is a declarative way of defining a new function from a database of input functions. We present InsideOut, a dynamic programming algorithm, to evaluate an FAQ. The algorithm rewrites the input query into a set of easier-to-compute FAQ sub-queries. Each subquery is then evaluated using a worst-case optimal relational join algorithm. The topic of designing algorithms to optimally evaluate the classic multiway join problem has seen exciting developments in the past few years. Our framework tightly connects these new ideas in database theory with a vast number of application areas in a coherent manner, showing potentially that -- with the right abstraction, blurring the distinction between data and computation -- a good database engine can be a general purpose constraint solver, relational data store, graphical model inference engine, and matrix/tensor computation processor all at once. The InsideOut algorithm is very simple, as shall be described in this paper. Yet, in spite of solving an extremely general problem, its runtime either is as good as or improves upon the best known algorithm for the applications that FAQ specializes to. These corollaries include computational tasks in graphical model inference, matrix/tensor operations, relational joins, and logic. Better yet, InsideOut can be used within any database engine, because it is basically a principled way of rewriting queries. Indeed, it is already part of the LogicBlox database engine, helping efficiently answer traditional database queries, graphical model inference queries, and train a large class of machine learning models inside the database itself.