Computer Laboratory

Profile  picture

Michael Schaarschmidt

PhD Student, Networks and Operating Systems Group
Computer Laboratory, University of Cambridge


Note: I submitted my dissertation and am now working fulltime as a research scientist at DeepMind. My new personal website is here.

I am a final year PhD candidate under the supervision of Eiko Yoneki (2nd Jon Crowcroft). Prior to starting my PhD, I completed an MPhil in Advanced Computer Science in Cambridge where I won the Google prize for the best MPhil project.

My PhD is on data processing aspects and computer systems applications of deep reinforcement learning. In particular, I am interested in providing controllers for runtime configurations in distributed systems with as little manual modelling as possible. RL is difficult to apply in these domains as problems are both expensive to evaluate (in comparison to typical simulations), and have large action spaces, thus making pure online learning impractical. My work particularly centers on leveraging information from log data (e.g. in database systems) to guide control models without intermediate simulation. Previously, I have worked on cloud infrastructure for real time caching with baqend (see our joint paper at VLDB 2017).

One of my more useful past projects is TensorForce, a TensorFlow library for applied RL: link

More recently, I have been working on RLgraph, a new software framework for designing and executing RL algorithms at scale: link.

My PhD research is supported by a Google PhD fellowship.

Google scholar profile: link 

I spent summer 2019 working away at DeepMind for further research into systems applications.


RLgraph: Flexible Computation Graphs for Deep Reinforcement Learning. [ArXiv preprint]
Michael Schaarschmidt*, Sven Mika*, Kai Fricke, Eiko Yoneki *equal contribution
Proceedings of the 2nd Conference on Systems and Machine Learning (SysML), Palo Alto, CA, April 2019.

LIFT: Reinforcement Learning in Computer Systems by Learning From Demonstrations. [ArXiv preprint]
Michael Schaarschmidt, Alexander Kuhnle, Ben Ellis, Kai Fricke, Felix Gessert, Eiko Yoneki

Quaestor: Query Web Caching for Database-as-a-Service Providers.[pdf]
Felix Gessert*, Michael Schaarschmidt*, Wolfram Wingerath, Erik Witt, Eiko Yoneki, Norbert Ritter *equal contribution
Proceedings of the 43rd International Conference on Very Large Databases (PVLDB 2017), Munich, Germany, August 2017

BOAT: Building Auto-Tuners with Structured Bayesian Optimization. [pdf] [featured by the morning paper]
Valentin Dalibard, Michael Schaarschmidt, Eiko Yoneki
Proceedings of the 26th World Wide Web Conference, Systems and Infrastructure track (WWW 2017), Perth, Australia, April 2017

Learning Runtime Parameters in Computer Systems with Delayed Experience Injection. [pdf][arXiv]
Michael Schaarschmidt, Felix Gessert, Valentin Dalibard, Eiko Yoneki
Deep Reinforcement Learning Workshop, NIPS 2016, Barcelona, Spain, December 2016

Tuning the Scheduling of Distributed Stochastic Gradient Descent with Bayesian Optimization. [pdf] [arXiv]
Valentin Dalibard, Michael Schaarschmidt, Eiko Yoneki
Workshop on Bayesian Optimization, NIPS 2016, Barcelona, Spain, December 2016

Towards Automated Polyglot Persistence. [pdf]
Michael Schaarschmidt, Felix Gessert, Norbert Ritter,
BTW 2015, Hamburg, Germany, March 2015

The Cache Sketch: Revisiting Expiration-based Caching in the Age of Cloud Data Management. [pdf]
Felix Gessert, Michael Schaarschmidt, Steffen Friedich, Norbert Ritter.
BTW 2015, Hamburg, Germany, March 2015

Towards a Scalable and Unified REST API for Cloud Data Stores. [pdf]
Felix Gessert, Steffen Friedrich, Wolfram Wingerath, Michael Schaarschmidt, Norbert Ritter.
Data Management in the Cloud (DMC 2014), Stuttgart, Germany, September 2014


Email: michael[dot]schaarschmidt[at]
Computer Laboratory, University of Cambridge
William Gates Building
15 JJ Thompson Avenue
Cambridge CB3 0FD