Dr. Maximus Mutschler

Background

Since March 2018
Research assistant at the Department of Cognitive Systems, University of Tübingen
2015 - 2018
2015 - 2018: Master of Science in Computer Science at the University of Tübingen
2011 - 2014
Bachelor of Science in Medical Computer Science at the University of Heidelberg and the University of Heilbronn

Research Interests

  • Machine Learning
  • Deep Neural Networks
  • Optimization Methods

Teaching

Project Work

Administrator of the Training Center for Machine Learning Cluster (TCML-Cluster) with 40 nodes and 160 GPUs

Supervised Theses

2019 Bachelor thesis Analysis of the Generalization Capability of a New Deep Neural Network Optimizer
2020 Bachelor thesis

Sensitivity Analysis of DNN Optimizers on different weight initializations

2020 Bachelor thesis

Forcing Deep Neural Networks to explore the Loss Landscape

2020 Master thesis

Adaptive Approximation niedrig dimensionaler Mannigfaltigkeiten zwischen lokalen Minima in den Fehlerlandschaften von Neuronalen Netzen

 

Theses

Optimizing a motor cortex model by evolution of connectivity patterns
Master's thesis, University of Tübingen, October 2017

Softwareplagiatserkennung auf Java-Bytecodebasis
Bachelor's thesis, University of Heidelberg and University of Heilbronn, June 2014

Publications

[1] Maximus Mutschler and Andreas Zell. “Parabolic Approximation Line Search for DNNs” .  Accepted at NeurIPS 2020.
[2] Maximus Mutschler and Andreas Zell. “A straightforward line search approach on the expected empirical loss for stochastic deep learning problems”  (2020).   (discontinued work)
[3] Maximus Mutschler and Andreas Zell. "Emperically explaining SGD from a line search perspective" (2021).  Accepted at ICANN 2021.
[4] Maximus Mutschler, Kevin Laube and Andreas Zell. "Using a one dimensional parabolic model of the full-batch loss to estimate learning rates during training" (2021). Accepted at NeurIPS 2021 Optimization Workshop.

ResearchGate profile  

Reviews

  •  Reviewed three works at ICANN 2021