Christopher (Chris) Re is an assistant professor in the Department of Computer Science at Stanford University and a Robert N. Noyce Family Faculty Scholar. His work's goal is to enable users and developers to build applications that more deeply understand and exploit data. Chris received his PhD from the University of Washington in Seattle under the supervision of Dan Suciu. For his PhD work in probabilistic data management, Chris received the SIGMOD 2010 Jim Gray Dissertation Award. He then spent four wonderful years on the faculty of the University of Wisconsin, Madison, before moving to Stanford in 2013. He helped discover the first join algorithm with worst-case optimal running time, which won the best paper at PODS 2012. He also helped develop a framework for feature engineering that won the best paper at SIGMOD 2014. In addition, work from his group has been incorporated into scientific efforts including the IceCube neutrino detector and PaleoDeepDive, and into Cloudera's Impala and products from Oracle, Pivotal, and Microsoft's Adam. He received an NSF CAREER Award in 2011, an Alfred P. Sloan Fellowship in 2013, a Moore Data Driven Investigator Award in 2014, the VLDB early Career Award in 2015, and the MacArthur Foundation Fellowship in 2015.

• NIH. Thank you to the NIH Big Data to Knowledge (BD2K) program for supporting our work on big data for mobility data led by Scott Delp.
• DARPA. Thank you to DARPA Memex for supporting our work. We are really excited to be part of this program to use extraction for good!
• NSF. Thank you to the NSF for supporting New Frontiers in Join Algorithms: Optimality, Noise, and Richer Languages which is joint with Atri Rudra and Hung Ngo!
• AirForce. Thank you to the Air Force for supporting Mathematical Foundations of Secure Computing Clouds; this is the hard work of Jordan Ellenberg (Math), Ben Recht (CS), Tom Ristenpart (CS), Rob Nowak (EE), and Steve Wright (CS).
• Oracle. Thank you for continuing support to support the Hazy Research group! This gift will be used to continue our work on feature engineering for structured analytics.
• ONR Thank you to the Office of Naval Research for supporting Ben Recht, Steve Wright, and my proposal about An Architecture for Integrating Information and Simplifying Large-scale Statistical Data Analysis (Award No. N000141310129)
• AmFam Thank you to American Family Insurance for their generous support of the Hazy group's research. We're very excited about the collaboration.
• DDR&E Thank you to DDR&E, DARPA, and Raytheon for funding Ben Recht and my proposal about operator splitting for information fusion applications.
• DARPA Thank you to DARPA (DEFT) for funding Jude Shavlik, Sriraam Natarajan (Wake Forest), and my proposal Creating Robust Relation Extractors and Anomaly Detectors via Probabilistic Logic-Based Reasoning and Learning.
• NSF. Thank you to the NSF for funding an EAGER to work on 'extracting Dark Data' with Shanan Peters from UW Geoscience and Miron Livny from CS.
• Google Research Award. Thank you to GOOGLE for supporting our proposal, GeoDeepDive: Machine Reading of Measurements.
• Oracle and Oracle Labs. Thank you to the Oracle Labs and the Oracle Analytics team for their generous support of the Hazy group's work! We are really excited to learn what customers need from in-database analytics. This will help support Arun's work. He and I are both very excited!
• Greenplum/EMC. Thank you to Greenplum/EMC for their generous support of the Hazy group's work! We are really excited to learn from this collaboration -- and to push some of Aaron and Arun's stuff in to MADlib!, an awesome open-source library for scalable in-database analytics.
• Office of Naval Research. Thank you to the ONR for support of my work under award no. N000141210041! This funding will allow our group to embark on a theoretical investigation of the foundations of building a large-scale, easy-to-use data-analysis system.
• NSF CAREER. I recently received the NSF CAREER award (IIS-1054009). Thank you to the NSF for their generous support of Hazy.
• IceCube. The Hazy group is extremely excited to announce funding for an exploratory data analysis project. The goal of the project is to apply Hazy's ideas to the problem of detecting neutrinos from the Big Bang in collaboration with the IceCube Neutrino Detector and Wisconsin Institutes for Discovery.
• LogicBlox The Hazy group is excited to collaborate with LogicBlox! Thank you, LogicBlox, for your generous research gift to support our ongoing work on Tuffy and Felix.
  
• DARPA DARPA's Machine Reading Program has the goal of understanding information expressed as free-form text. We are building a scalable engine to process a probabilistic logic called Markov Logic to support this effort.
• Thank You! The Hazy group would like to thank our sponsors in the past and coming year: The Microsoft Jim Gray Lab, DARPA/AFOSR via SRI, the NSF, Google, Johnson Controls Inc., the University of Wisconsin-Madison, the Office of Naval Research, and Physical Layer Systems. In addition, we would like to thank our collaborators at the Wisconsin Institutes for Discovery, HP Labs-China, LogicBlox, Greenplum, Oracle and IBM.

Current PhD Students

• Chris Aberger Coadvisor: Kunle Olukotun
• Xiao Cheng
• Chris De Sa Coadvisor: Kunle Olukotun
• Emily Mallory (Biomedical Informatics) Coadvisor: Russ Altman
• Stefan Hadjis
• Rohan Puttagunta
• Alex Ratner
• Jaeho Shin
• Feiran Wang
• Sen Wu

Current MS and Coterm Students

• Konstantin Lopyrev
• Juhana Kangaspunta
• Thomas Palomares
• Susan Tu

Current Postdocs

• Stephen Bach Coadvisor: Jure Leskovec
• Madalina Fiterau Coadvisor: Scott Delp
• Jason Fries Coadvisor: Scott Delp
• Ioannis Mitliagkas Coadvisor: Lester Mackey
• Theodoros Rekatsinas
• Ce Zhang

Stanford Students and Postdocs who I regularly harass:

• Manas Joglekar Advisor: Hector Garcia-Molina
• Johannes Birgmeier Advisor: Gill Bejerano
• Jiyan Yang Advisor: Michael Saunders (ICME) and Michael Mahoney (Berkeley)
• Kun-Hsing Yu Advisor: Michael Snyder (BioE)

Alumni (Degree year, First Employment)

• Ce Zhang (PhD 2015, Postdoc at Stanford)
• Michael Fitzpatrick (MS 2015, Google)
• Firas Abuzaid (MS 2015, MIT for PhD)
• Zifei Shan (MS 2015, ClearCut)
• Adam Goldberg (BS 2015, Rubrik)
• Adam Perelman (BS 2015, Good Eggs)
• Srikrishna Sridhar (PhD 2014, GraphLab) Main Advisor: Stephen J. Wright
• Victor Bittorf (MS 2014, Cloudera)
• Vidhya Govindaraju (MS 2014, Oracle)
• Mark Wellons (MS 2013, Amazon)
• Arun Kumar (MS 2013, Wisconsin for PhD)
• Feng Niu (PhD 2012, Google)
• Xixi Luo (MS in Industrial Engineering 2012, Oracle)
• Vinod Ramachandran (MS 2011, Oracle)
• M. Levent Koc (MS 2011, Google)
• Balaji Gopalan (MS 2010, Google)

• (1) Building Knowledge Bases. The world is full of interesting data and knowledge, but it's difficult to use. Part of the challenge is that knowledge is buried in all kinds of crazy places, e.g., in the text, tables, and figures of scholarly articles, webpages, and government reports. We are building a trained system called DeepDive to help extract this knowledge. The system does everything from optical character recognition to natural language processing to object reconciliation to regression. A key goal is to demonstrate real value to end users with these systems, and we are working with scientists in several disciplines to understand this challenge.
  
  DeepDive is our attempt to understand a new type of database system. Our new approach can be summarized as follows: the data, the output of various tools, the input from users — including the program the developer writes — are observations from which the system statistically infers the answer. This view is a radical departure from traditional data processing systems, which assume that the data is one-hundred percent correct. A key problem in DeepDive is that the system needs to consider many possible interpretations for each data item. In turn, we need to explore a huge number of combinations during probabilistic inference, which is one of the core technical challenges.
  • Demos, Examples, and Papers.
    With Shanan Peters (UW Geoscience) and Miron Livny (Condor), we are combining Macrostrat with DeepDive to (hopefully!) deliver value for Geoscientists. One key challenge is extracting all the measurement information that is reported in the literature, that is buried in the dark data of text, graphs, and figures. A demo video and a new video about quality that is higher than the volunteers who have been at this for the last decade. This is all powered by DeepDive. Thank you to the National Science Foundation and Google for supporting this work.
    • Portions of DeepDive are available. Elementary. Gibbs Sampling on Factor Graphs on TBs in files, Accumulo, or HBase! Now with BUGS support! Tuffy is updated, which uses an RDBMS to process Markov Logic. The most recent version of DeepDive's engine is described in a recent SIGMOD paper. A new release is coming soon!
Our goal is to acquire more sources of data for DeepDive to understand more deeply to change the way that science and industry operate.

• (2) Statistical Analytics in Data Processing Systems. Enabling users to understand their data more deeply can improve society, science, and business. The new generation of analysis tools are statistical and predictive, which require different low-level processing than traditional SQL-style analytics. Our group is interested in understanding how these analytics change traditional database systems. As a result, we work on both integrating statistical algorithms into existing data processing systems and building our own frameworks. We have built several tools now used in science and industry.
  • Demos, Examples, and Papers.

    • For integrating statistical analytics into a data processing system, we wrote these two papers paper or this one. This work has been incorporated into MADlib (EMC/Greenplum/Pivotal), a product from Oracle, and in Cloudera's Impala.
    • In scientific applications, Mark Wellons, Ben Recht, and I have worked with the IceCube Neutrino Detector. Mark's code now runs in the detector on the South Pole and is used on over 250 Million events per day. More details are in this video, this video, this paper at the The International Cosmic Ray Conference 2013, or this paper. A paper was accepted to NIM A and described here. IceCube (and Mark) got the cover of Science! Awesome!
    • Code for our own engines are here and Hogwild! for. These process things like SVMs, logistic regression, matrix factorization, and other convex goodness without locking. Specialized versions of trace-norm regularization called Jellyfish and non-negative matrix factorization called HottTopix. Each of these is referenced in a paper of the same name. There are also videos about some of the technical portions of these projects Matrix Factorization, Seismic Data Interpolation, and a nowcasting framework (now called Ringtail).
    • See some recent thoughts about this problem including Brainwash and a spate of recent demos such as a nowcasting framework, Ringtail in VLDB and WebDB; and an enterprise feature-selection system with Oracle. I am also interested in making analytic systems self-tuning.
      
  Our goal is to make analytics systems easier to build, maintain, and tune.
• (3) Fundamentals of Data Processing. Almost all data processing systems have their intellectual roots in first order logic. The most computationally expensive (and most interesting) operation in such systems is the relational join. Recently, I helped discover the first join algorithm with optimal worst-case running time. This result uses a novel connection between logic, combinatorics, and geometry. We are using this connection to develop new attacks on classical problems in listing patterns in graphs and in statistical inference. Two threads have emerged:
  • The first theme is that these new worst-case-optimal algorithms are fundamentally different from the algorithms used in (most of) today's data processing systems. Although our algorithm is optimal in the worst case, commercial relational database engines have been tuned to work well on real data sets by smart people for about four decades. And so a difficult question is how does one translate these insights into real data processing systems?
  • The second theme is that we may need new techniques to get theoretical results strong enough to guide practice. As a result, I've started thinking about "beyond worst-case analysis" and things like conditioning for combinatorial problems to hopefully build theory that can inform practice to a greater extent. The first papers have just been posted.
  • Demos, Examples, and Papers.
    • Worst-case Optimal Joins. We have posted a survey for SIGMOD record about recent advances in join algorithms. Our goal is to give a high-level view of the results for practitioners and applied researchers. We also managed to simplify the arguments. A full version of our join algorithm with worst-case optimal running time is here. The LogicBlox guys have their own commercial worst-case optimal algorithm. Our new system, EmptyHeaded is based on this theory.
    • Beyond Worst-case Joins. This work is our attempt to go beyond worst-case analysis for join algorithms. We (with Dung Nguyen) develop a new algorithm that we call Minesweeper based on these ideas. The main theoretical idea is to formalize the amount of work any algorithm spends certifying (using a set of propositional statements) that the output set is complete (and not, say, a proper subset). We call this set of propositions the certificate. We manage to establish a dichotomy theorem for this stronger notion of complexity: if a query is what Ron Fagin calls beta-acyclic, then Minesweeper runs in time linear in the certificate; if a query is beta-cyclic than on some instance any algorithm takes time that is super linear in the certificate. The results get sharper and more fun.
    • Almost to one algorithm to rule them all? We have a much better description of beyond worst-case optimality with a resolution framework and a host of new results for different indexing strategies. This paper supercedes many of the results in Minesweeper and in a much nicer way!. We also hope to connect more of geometry and resolution... but we'll see!
    • A first part of our attack on conditioning for combinatorial problems is in NIPS and on Arxiv.
    • It is not difficult to get me interested in a theory problem. Ask around the Infolab if you don't believe me.
  Our goal is to understand the fundamentals of data processing systems.

• Fall 15: CS145, Introduction to Databases
• Spring 15: CS346, Database System Implementation
• Spring 15: CS341, Project in Mining Massive Data Sets
• Fall 14: CS145, Introduction to Databases
• Spring 14: CS346 Advanced Database Implementation
• Winter 14: CS345 Advanced Database Systems