Skip to content (access key 's')
Logo of Technion
Logo of CS Department
Logo of CS4People

The Taub Faculty of Computer Science Events and Talks

CSpecial Talk: Neuro-Symbolic Program Synthesis
event speaker icon
Alex (Oleksandr) Polozov (Microsoft Research AI – Redmond)
event date icon
Monday, 19.11.2018, 12:00
event location icon
Room 337 Taub Bld.
Program synthesis, the task of automatically finding a program that satisfies a given user intent specification, has been successfully applied to aid commercial data wrangling, software engineering, and question answering. While data-driven (deep learning) and symbolic (formal logic) techniques are both commonly used for program synthesis, both of them have their strengths and weaknesses. Symbolic techniques guarantee correctness of the generated program with respect to the specification, but are difficult to design, develop, and tune. Data-driven techniques are easier to train and tune given sufficient customer data, but they cannot guarantee program correctness and have less insight into the program semantics.

This talk describes three recent projects from Microsoft Research AI that address these issues by combining the strengths of data-driven and symbolic worlds. First, I will describe how deep learning models boost the performance of Microsoft PROSE SDK, a mass-market framework for programming by examples, by guiding its deductive search process. Second, I will present execution-guided decoding, a technique for making neural SQL program generation more semantics-aware, which achieves state-of-the-art accuracy on the Salesforce WikiSQL dataset. Finally, I will present a system for automatic completion of C# expressions given their surrounding context, powered by a novel combination of static analysis, attribute grammars, and gated graph neural networks

Alex (Oleksandr) Polozov, a researcher in the Deep Procedural Intelligence group at Microsoft Research AI, Redmond. Works on neural program synthesis from input-output examples and natural language, intersections of machine learning and software engineering, and neuro-symbolic architectures. Interested in combining neural and symbolic techniques to tackle the next generation of AI problems, including program synthesis, planning, and reasoning.

His main passion of the last several years has been PROSE, a program synthesis framework for mass-market development of by-example technologies. He completed his Ph.D. in the Paul G. Allen School of Computer Science & Engineering at the University of Washington. His advisors were Sumit Gulwani and Zoran Popović. Before joining UW, he received his B.S. in System Analysis with honors from the National Technical University of Ukraine “Kyiv Polytechnic Institute”