Don Dennis

Research Fellow
Machine Learning and Optimization Group
Microsoft Research India
 t-dodenn [at] microsoft [dot] com
 donkdennis [at] gmail [dot] com
Its Me

Latest News
[Jan '19]   Manuscript on shallow recurrent networks submitted to ICML '19.
[Dec '18]   Manuscript describing EMI-RNN algorithm accepted to NeurIPS '18.
[Dec '18]   GesturePod presented at NeurIPS '18 at the MLPCD workshop.
[Dec '18]   GesturePod submitted to CHI'19 and covered by ZD Net.
[Nov '18]   EMI-RNN implementation now part of Microsoft's EdgeML.

I am a Research Fellow at Microsoft Research India, where I work with Prateek Jain and Harsha Simhadri in the Machine Learning and Optimisation (MLO) group and Intelligent Devices Expedition group. I do research on bringing state-of-the-art machine learning to severely resource-constrained devices. I am slated to finish my research fellowship at Microsoft Research in Spring 2019 after which I hope to work towards a PhD.

My current research focuses on designing theoretically sound and practical ML algorithms and systems. For instance,

  • With my EMI-RNN work (accepted at NeurIPS '18), we speed-up RNN inference by up to 72x while simultaneously gaining on performance. As a concrete application, EMI-RNN brought real-time key-word spotting to devices like Raspberry Pi0 and MXChip. A demonstration was accepted at the MLPCD workshop at NeurIPS '18.
  • The ProtoNN (prototypes + kNN) based machine learning pipeline I developed for GesturePod (in submission, CHI '19) does robust, accurate and real-time gesture recognition on 32kB of RAM and a 48MHz processor. GesturePod has been widely covered by the media and was Microsoft's official demonstration at NeurIPS '18.

Research Interests
    · Machine Learning (ML) - Theoretical and Applied Aspects
    · Robust and Reliable ML for Systems and Devices
    · Resource Efficient ML
    · ML Embedded in Devices (Robotics/IoT/Autonomous Systems)

My current focus is primarily academic, but I strongly feel about the impact cutting-edge technology can have for social good.

During our undergraduate years, Rito, Karan, Prashant and I started a venture - Weave, focused on creating micro-entrepreneurs in a marginalized weaver community. Although the start-up itself failed to gain much traction, it was an important lesson in building products with empathy.

It also set the ball rolling for our future endeavours: the journey somewhere along the line transformed into a new venture, rooting for healthcare access and affordability, using deep learning for diagnostics at ChironX. I continue to actively collaborate with Rito and Prashant at ChironX on designing their deep learning solutions and architectures.