Open Research Clubs 2020

April 2020

  • "LIM: A Privacy-Respecting Android Malware Classifier Using Federated Learning" (Rafa Galvez / KU Leuven)
  • "ReLU Code Space: A basis for rating network quality besides accuracy" (Natalia Shepeleva / SCCH)
  • "Towards a ReLU network based distance for comparing GANs with small samples" (Michal Lewandowski (Werner Zellinger, Bernhard Moser et al) / SCCH)

Rafa Galvez (KU Leuven) on LIM: A Privacy-Respecting Android Malware Classifier Using Federated Learning

Natalia Shepeleva (SCCH) on ReLU Code Space: A basis for rating network quality besides accuracy

Michal Lewandowski (Werner Zellinger, Bernhard Moser et al, SCCH) towards a ReLU network based distance for comparing GANs with small samples

May 2020

  • Distance Function and Effective Dimension in a Kernel Analysis of Deep Networks (Sergei V. Pereverzyev / RICAM)
  • Discussion of "Kernel Analysis of Deep Networks" (Journal of Machine Learning Research 12 (2011) 2563-2581), moderated by Sergei V. Pereverzyev / RICAM
  • Provable Robustness of ReLU networks via Maximization of Linear Regions (Francesco Croce / University of Tübingen)

Sergei V. Pereverzyev (RICAM) on Distance Function and Effective Dimension in a Kernel Analysis of Deep Networks

Sergei V. Pereverzyev (RICAM) discusses "Kernel Analysis of Deep Networks" (Journal of Machine Learning Research 12 (2011) 2563-2581)

Francesco Croce (University of Tübingen) on Provable Robustness of ReLU networks via Maximization of Linear Regions

June 2020

  • Credit Assignment in Reinforcement Learning (Marius-Constantin Dinu / LIT AI LAB / Institute for Machine Learning / JKU)

Marius-Constantin Dinu  (LIT AI LAB / Institute for Machine Learning / JKU) on Credit Assignment in Reinforcement Learning

July 2020

  • Fast minimum-norm adversarial attacks through adaptive norm constraints (Maura Pintor, Battista Biggio, Wieland Brendel / Max-Planck-Institut für Biologische Kybernetik / University of Cagliari)
  • Self-Supervised Prototypical Transfer Learning for Few-Shot Classification (Arnout Devos / EPFL, Swiss Federal Institute of
    Technology Lausanne)

Arnout Devos (EPFL, Swiss Federal Institute of Technology Lausanne) on Self-Supervised Prototypical Transfer Learning for Few-Shot Classification

Maura Pintor (University of Cagliari) on Fast minimum-norm adversarial attacks through adaptive norm constraints

September 2020

  • On the Inductive Biases in Data Augmentation and Adversarial Robustness (Hamid Eghbal-zadeh / LIT-AI Lab / Inst. f. Perceptional Computation / JKU)
  • Optimization of Differential and Informational Privacy, An Information Theoretic Approach (Mohit Kumar / SCCH)

Hamid Eghbal-zadeh (LIT-AI Lab / Inst. f. Perceptional Computation / JKU) on the Inductive Biases in Data Augmentation and Adversarial Robustness

Mohit Kumar (SCCH) on "Optimization of Differential and Informational Privacy, An Information Theoretic Approach"

October 2020

  • Deep neural network tropicalization (Anton Ponomarchuk / RICAM)
  • Functional regression and some of its applications (Nguyen Duc Hoan / RICAM)

Anton Ponomarchuk (RICAM) on Deep neural network tropicalization

Nguyen Duc Hoan (RICAM) on Functional regression and some of its applications

November 2020

  • Empirical study of linear regions in ReLU networks (Michael Lewandowski / SCCH, Hamid Eghbal-Zadeh, Vaios Laschos)
  • Overlapping Multi-Patch Structures in Isogeometric Analysis (S. Kargaran / SCCH)

S. Kargaran (SCCH) on overlapping multi-patch structures in isogeometric analysis

M. Lewandowsky (SCCH) on empirical study of linear regions in ReLU networks

December 2020

  • LiM: A privacy-respecting Android malware classifier using Federated Learning (Rafa Gálvez, Veelasha Moonsamy, Claudia Díaz / COSIC / KU Leuven)
  • Domain-Adversarial Offline Reinforcement Learning: Learning generative policies from demonstrations without environment
    interactions (Marius-Constantin Dinu / LIT-AI / IML / JKU)

Marius-Constantin Dinu (LIT-AI / IML / JKU) on Domain-Adversarial Offline Reinforcement Learning: Learning generative policies from demonstrations without environment
interactions

Rafa Gálvez (COSIC / KU Leuven) on LiM: A privacy-respecting Android malware classifier using Federated Learning