Biomedical Data Analysis Group

About me

From 2017 to 2022, I pursued my doctorate in computer science in the Biomedical Data Analysis group of Professor Volker Roth. Before that I completed my bachelor’s and master’s degree in physics at the Swiss Federal Institute of Technology (ETH) in Zurich and spent a semester in the Complex Systems Master’s programme at École normale supérieure (ENS) de Lyon.

I have a broad interest in machine learning and application of data analytics tools. In my research I worked on information theoretic approaches to deep latent variable models and structuring latent representations, image segmentation in real-world settings, as well as Deep Learning theory in the context of Gaussian Processes and the Neural Tangent Kernel to reason about practical preformance aspects of deep neural networks. In our weObserve research project, we used these methods to identify soil degradation areas from aerial images and investigate bird migration with the goal of establishing new insights into these phenomena by combing heterogeneous data sources and Citizen Science. Further applications involved facilitating the discovery of novel molecules with desirable properties.

In my projects, I mainly worked in Python and the frameworks and platforms of TensorFlow, PyTorch, JAX, CUDA and others. I also teach a Python crash course once a semester, and I tutored in the lectures on Bioinformatics Algorithms and Information Theory.

My projects

weObserve: Integrating Citizen Observers and High Throughput Sensing Devices for Big Data Collection, Integration, and Analysis (SNF NRP 75 Big Data)

Contact Information

E-mail: samarinm17@gmail.com

Community Service

• Program committee member for workshops: Bridging the Gap: From Research to Clinics (NeurIPS 2021), Machine Learning for Drug Discovery (ICLR 2022).
• Reviewer for machine learning conferences: NeurIPS (2019*, 2020, 2021; *top 50% ranking reviewer), ICML (2020; top 33% ranking reviewer), ICLR (2021; outstanding reviewer).
• Member of the German Association for Pattern Recognition (DAGM).

Publications & Preprints

• Machine Learning for Informed Representation Learning.
  M. Samarin.
  PhD thesis, 2022.
  [edoc]
• Mesh-free Eulerian Physics-Informed Neural Networks.
  F. Arend Torres, M. Negri, M. Nagy-Huber, M. Samarin & V. Roth.
  arXiv preprint arXiv:2206.01545, 2022.
  [arXiv]
• Feature Learning and Random Features in Standard Finite-Width Convolutional Neural Networks: An Empirical Study.
  M. Samarin, V. Roth & D. Belius.
  Accepted to the Conference on Uncertainty in Artificial Intelligence (UAI) 2022.
  [paper, OpenReview]
• Learning Invariances with Generalised Input-Convex Neural Networks.
  V. Nesterov, F. Arend Torres, M. Nagy-Huber, M. Samarin & V. Roth.
  arXiv preprint arXiv:2204.07009, 2022.
  [arXiv]
• Learning Conditional Invariance through Cycle Consistency.
  M. Samarin*, V. Nesterov*, M. Wieser, A. Wieczorek, S. Parbhoo, & V. Roth.
  Conference on Pattern Recognition (GCPR), 2021.
  [video, paper, arXiv, code]
• Investigating Causal Factors of Shallow Landslides in Grassland Regions of Switzerland.
  L. Zweifel, M. Samarin, K. Meusburger, & C. Alewell.
  Natural Hazards and Earth System Sciences (NHESS), 2021.
  [paper]
• Learning Extremal Representations with Deep Archetypal Analysis.
  S. M. Keller, M. Samarin, F. Arend Torres, M. Wieser & V. Roth.
  International Journal on Computer Vision (IJCV), 2020.
  [paper, arXiv, code]
• Identifying Soil Erosion Processes in Alpine Grasslands on Aerial Imagery with a U-Net Convolutional Neural Network.
  M. Samarin*, L. Zweifel*, V. Roth & C. Alewell.
  Remote Sensing, 2020.
  [paper, code]
• On the Empirical Neural Tangent Kernel of Standard Finite-Width Convolutional Neural Network Architectures.
  M. Samarin, V. Roth & D. Belius.
  arXiv preprint arXiv:2006.13645, 2020.
  [arXiv]
• Deep Archetypal Analysis.
  S. M. Keller, M. Samarin, M. Wieser & V. Roth.
  German Conference on Pattern Recognition (GCPR), 2019.
  GCPR 2019 Honorable Mention Paper Award.
  [paper, arXiv, code]

Talks

• Segmentation of Soil Degradation Sites in Swiss Alpine Grasslands with Deep Learning.
  M. Samarin*, L. Zweifel*, C. Alewell & V. Roth.
  NeurIPS Workshop on “AI for Earth Sciences”, December 12, online, 2020.
  [video]
• Visual Understanding in Semantic Segmentation of Soil Erosion Sites in Swiss Alpine Grasslands.
  M. Samarin, M. Nagy-Huber, L. Zweifel, K. Meusburger, C. Alewell & V. Roth.
  EGU General, May 4-8, online, 2020.
• Identification of Soil Erosion in Alpine Grasslands on High-Resolution Aerial Images: Switching from Object-based Image Analysis to Deep Learning?
  L. Zweifel, M. Samarin, K. Meusburger, V. Roth & C. Alewell.
  EGU General Assembly, May 4-8, online, 2020.
• Object Segmentation for Identification of Soil Erosion Sites in Swiss Alpine Grasslands with Neural Networks.
  M. Samarin & V. Roth.
  EGU General Assembly, April 7-12, Vienna, 2019.
• Detection of Shallow Landslides on Aerial Images using Convolutional Neural Networks.
  M. Samarin, L. Zweifel, K. Meusburger, C. Alewell & V. Roth.
  Applied Machine Learning Days, January 26-29, Lausanne, 2019.

Conference Posters

• Exploring Data Through Archetypal Representatives.
  S. M. Keller, F. Arend Torres, M. Samarin, M. Wieser & V. Roth.
  NeurIPS Workshop on “Learning Meaningful Representations of Life”, December 13, Vancouver, 2019.
  [poster]
• Identifying Soil Degradation in Swiss Alpine Grasslands using different Machine Learning Approaches.
  L. Zweifel, M. Samarin, K. Meusburger, V. Roth & C. Alewell.
  EGU General Assembly, April 7-12, Vienna, 2019.
• Identifying Shallow Landslides on Swiss Alpine Grasslands using Machine Learning.
  L. Zweifel, M. Samarin, K. Meusburger, V. Roth & C. Alewell.
  1st Swiss Workshop on Machine Learning for Environmental and Geosciences, January 17, Dübendorf, 2019.
• Spatio-temporal Analysis of Soil Degradation in Swiss Alpine Grasslands.
  L. Zweifel, M. Samarin, K. Meusburger & C. Alewell.
  Women in Big Data Workshop, Zurich, 2018.