Jump to Content
Ketan Mandke

Ketan Mandke

Ketan is a software engineer at Google, where he works on making machine learning more accessible to programmers. He previously worked on mmWave wireless networks and mobile networking within Alphabet as well. Prior to joining Google, Ketan implemented software defined radio systems and developed signal processing algorithms for detection and estimation in geolocation applications. Ketan holds a Ph.D. in Electrical Engineering from the University of Texas at Austin.
Authored Publications
Google Publications
Other Publications
Sort By
  • Title
  • Title, descending
  • Year
  • Year, descending
    A Flexible Approach to Autotuning Multi-Pass Machine Learning Compilers
    Berkin Ilbeyi
    Bjarke Roune
    Blake Hechtman
    Emma Wang
    Karthik Srinivasa Murthy
    Mike Burrows
    Nikhil Sarda
    Rezsa Farahani
    Samuel J. Kaufman
    Shen Wang
    Sudip Roy
    Yuanzhong Xu
    PACT (2021)
    Preview abstract Search-based techniques have been demonstrated effective in solving complex optimization problems that arise in domain-specific compilers for machine learning (ML). Unfortunately, deploying such techniques in production compilers is impeded by two limitations. First, prior works require factorization of a computation graph into smaller subgraphs over which search is applied. This decomposition is not only non-trivial but also significantly limits the scope of optimization. Second, prior works require search to be applied in a single stage in the compilation flow, which does not fit with the multi-stage layered architecture of most production ML compilers. This paper presents XTAT, an autotuner for production ML compilers that can tune both graph-level and subgraph-level optimizations across multiple compilation stages. XTAT applies XTAT-M, a flexible search methodology that defines a search formulation for joint optimizations by accurately modeling the interactions between different compiler passes. XTAT tunes tensor layouts, operator fusion decisions, tile sizes, and code generation parameters in XLA, a production ML compiler, using various search strategies. In an evaluation across 150 ML training and inference models on Tensor Processing Units (TPUs) at Google, XTAT offers up to 2.4x and an average 5% execution time speedup over the heavily-optimized XLA compiler. View details
    Operating a UAV Mesh & Internet Backhaul Network using Temporospatial SDN
    Brian Barritt
    Tatiana Kichkaylo
    Victor Lin
    2017 IEEE Aerospace Conference
    Preview abstract In this paper we describe an application of Temporospatial SDN (TS-SDN) to UAV networks. Airborne platforms (airplanes, balloons, airships) can be used to carry wireless communication nodes to provide direct-to-user as well as backhaul connections. Such networks also include ground nodes typically equipped with highly directional steerable transceivers. The physics of flight as well as state of the atmosphere lead to time-dynamic link metrics and availability. As nodes move around, the network topology and routing need to adjust to maintain connectivity. Further, mechanical aspects of the system, such as time required to mechanically steer antennas, makes the reactive repair approach more costly than in terrestrial applications. Instead, TS-SDN incorporates reasoning about physical evolution of the system to proactively adjust the network topology in anticipation of future changes. Using airborne networks under development at Google as an example, we discuss the benefits of the TS-SDN approach compared to reactive repair in terms of network availability. We also identify additional constraints one needs to account for when computing the network topology, such as non-interference with other stationary and moving sources. Existing SDN standards do not support scheduled updates necessary in a TS-SDN. We describe our extensions to control messages and software implementation used in field tests. View details
    No Results Found