If you are interested in pursuing a PhD (or Postdoc) with me, please submit your application to our online application system (PhD only) and drop me an email.
Right now I am looking especially for these projects and profiles
- Program generation for performance in machine learning, optimization, program analysis, other domains: strong programming skills (meta programming a plus), compilers, mathematics
- Meta programming, generative programming
- Fast program analysis with abstract domains
- Algebraic signal processing: strong math, signal processing
- Hardware design for deep neural nets and other machine learning algorithms: experience with FPGAs and Verilog or VHDL
Master theses (for ETH students) on various topics are always available - just email me.
Current Research Areas and Interests
I like to combine techniques from mathematics, computer science, and engineering to attack difficult problems.
Program generation for performance. For functionality of mathematical nature we aim to automatically generate highest performance code from a high level mathematical description. With Spiral and LGen we have built such a generators for the domain of linear transforms and small scale linear algebra. The project combines techniques from mathematics, programming languages, symbolic computation, and compilers. We are looking into building generators and providing performance in other domains including machine learning.
Generators and design of streaming IP cores. We study the design and computer generation of streaming IP cores on FPGAs. Examples:
Algebraic signal processing. We have developed a new theory of discrete signal processing that makes it possible to generalize it to signals (or data) with various index domains including graphs or power sets.
Fast abstract domains. We have developed a new methodoolgy to speed up certian types of program analysis by orders of magnitude. Examples:
Some other recent research directions
+41 44 632 8580
8092 Zurich, Switzerland
- Chris Wendler
- Eliza Wszola
- Gagandeep Singh (with Martin Vechev)
- Francois Serre
- Alen Stojanov
Current master theses:
- Jonas Stulz: Low-Precision Quantized Matrix-Matrix Multiplication for Neural Networks
Spring 2018: sabbatical - yeah!
Dan Alistarh (IST Austria), Paolo Bientinesi (RWTH Aaachen), Martin Jaggi (EPFL), Andreas Krause (ETH), Tiark Rompf (Purdue), Martin Vechev (ETH)
Dept. of Computer Science
ETH Zurich, CAB H69.3
8092 Zurich, Switzerland
phone: +41 44 632 7303
email: pueschel at inf.ethz.ch