Dagstuhl-Seminar 18112
Coding Theory for Inference, Learning and Optimization
( 11. Mar – 16. Mar, 2018 )
(zum Vergrößern in der Bildmitte klicken)
Permalink
Bitte benutzen Sie folgende Kurz-Url zum Verlinken dieser Seite:
https://www.dagstuhl.de/18112
Organisatoren
- Po-Ling Loh (University of Wisconsin - Madison, US)
- Arya Mazumdar (University of Massachusetts - Amherst, US)
- Dimitris Papailiopoulos (University of Wisconsin - Madison, US)
- Rüdiger Urbanke (EPFL - Lausanne, CH)
Kontakt
- Michael Gerke (für wissenschaftliche Fragen)
- Annette Beyer (für administrative Fragen)
Coding theory has recently found new applications in areas such as distributed machine learning, dimension reduction, and variety of statistical problems involving estimation and inference. In machine learning applications that use large-scale data, it is desirable to communicate the results of distributed computations in an efficient and robust manner. In dimension reduction applications, the pseudorandom properties of algebraic codes may be used to construct projection matrices that are deterministic and facilitate algorithmic efficiency. Finally, relationships that have been forged between coding theory and problems in theoretical computer science, such as k-SAT or the planted clique problem, lead to a new interpretation of the sharp thresholds encountered in these settings in terms of thresholds in channel coding theory.
The aim of this Dagstuhl Seminar is to draw together researchers from industry and academia that are working in coding theory, particularly in these different (and somewhat disparate) application areas of machine learning and inference. The discussions and collaborations facilitated by this seminar will help spark new ideas about how coding theory may be used to improve and inform modern techniques for data analytics.
Creative Commons BY 3.0 DE
Po-Ling Loh, Arya Mazumdar, Dimitris Papailiopoulos, and Rüdiger Urbanke
Codes are widely used in engineering applications to offer reliability and fault tolerance. The high-level idea of coding is to exploit redundancy in order to create robustness against system noise. The theoretical properties of codes have been studied for decades both from a purely mathematical point of view, as well as in various engineering contexts. The latter have resulted in constructions that have been incorporated into our daily lives: No storage device, cell phone transmission, or Wi-Fi connection would be possible without well-constructed codes.
Recent research has connected concepts in coding theory to non-traditional applications in learning, computation and inference, where codes have been used to design more efficient inference algorithms and build robust, large-scale, distributed computational pipelines. Moreover, ideas derived from Shannon theory and the algebraic properties of random codes have resulted in novel research that sheds light on fundamental phase transition phenomena in several long-standing combinatorial and graph-theoretic problems.
The main goal of our seminar was to accelerate research in the growing field of coding theory for computation and learning, and maximize the transformative role of codes in non-traditional application areas. The seminar brought together 22 researchers from across the world specializing in information theory, machine learning, theoretical computer science, optimization, and statistics. The schedule for each day included a tutorial talk by a senior researcher, followed by shorter talks by participants on recent or ongoing work. The afternoons were devoted to informal breakout sessions for groups to discuss open questions. Two of the larger breakout sessions focused on distributed optimization and group testing.
Seminar participants reported that they enjoyed hearing about new ideas, as well as delving into deeper technical discussions about open problems in coding theory. Some topics deserving special mention include the use of techniques in statistical mechanics; locally decodable and recoverable codes; submodular function optimization; hypergraph clustering; private information retrieval; and contagion on graphs. All participants valued the ample time for discussions between and after talks, as it provided a fruitful atmosphere for collaborating on new topics.
Po-Ling Loh, Arya Mazumdar, Dimitris Papailiopoulos, and Rüdiger Urbanke
- Dimitris Achlioptas (University of California - Santa Cruz, US) [dblp]
- Alexander Barg (University of Maryland - College Park, US) [dblp]
- Martin Bossert (Universität Ulm, DE) [dblp]
- Elette Boyle (The Interdisciplinary Center - Herzliya, IL) [dblp]
- Amin Coja-Oghlan (Goethe-Universität - Frankfurt am Main, DE) [dblp]
- Anna Gál (University of Texas - Austin, US) [dblp]
- Venkatesan Guruswami (Carnegie Mellon University - Pittsburgh, US) [dblp]
- Hamed S. Hassani (University of Pennsylvania - Philadelphia, US) [dblp]
- Sihuang Hu (RWTH Aachen, DE) [dblp]
- Sidharth Jaggi (The Chinese University of Hong Kong, HK) [dblp]
- Marc Lelarge (ENS - Paris, FR) [dblp]
- Po-Ling Loh (University of Wisconsin - Madison, US) [dblp]
- Nicolas Macris (EPFL - Lausanne, CH) [dblp]
- Arya Mazumdar (University of Massachusetts - Amherst, US) [dblp]
- Olgica Milenkovic (University of Illinois - Urbana Champaign, US) [dblp]
- Dimitris Papailiopoulos (University of Wisconsin - Madison, US) [dblp]
- Ankit Singh Rawat (MIT - Cambridge, US) [dblp]
- Changho Suh (KAIST - Daejeon, KR) [dblp]
- Itzhak Tamo (Tel Aviv University, IL) [dblp]
- Rüdiger Urbanke (EPFL - Lausanne, CH) [dblp]
- Pascal Vontobel (The Chinese University of Hong Kong, HK) [dblp]
- Mary Wootters (Stanford University, US) [dblp]
Klassifikation
- artificial intelligence / robotics
- data structures / algorithms / complexity
- optimization / scheduling
Schlagworte
- Coding theory
- distributed optimization
- machine learning
- sparse recovery
- threshold phenomena
