14.01.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Mahdi Soltanolkotabi (University of Southern California)
• Learning via early stopping and untrained neural nets
• see video

21.01.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Taiji Suzuki (The University of Tokyo, and Center for Advanced Intelligence Project, RIKEN, Tokyo)
• Statistical efficiency and optimization of deep learning from the viewpoint of non-convexity
• see video

28.01.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Suriya Gunasekar (Microsoft Research, Redmond)
• Rethinking the role of optimization in learning
• see video

02.02.2021, 18:00 Uhr, Deep Learning Theory Group Seminar

• Johannes Müller (MPI MiS, Leipzig)
• Monte Carlo rates and neural network approximation (Bytes)

04.02.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Tailin Wu (Stanford University)
• Phase transitions on the universal tradeoff between prediction and compression in machine learning
• see video

09.02.2021, 18:00 Uhr, Deep Learning Theory Group Seminar

• Hanna Tseran (MPI MiS, Leipzig)
• Effects of Neural Network Initialization (Paper review)

11.02.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Samuel L. Smith (DeepMind)
• A Backward Error Analysis for Random Shuffling SGD
• see video

16.02.2021, 18:00 Uhr, Deep Learning Theory Group Seminar

• Johannes Müller (MPI MiS, Leipzig)
• An explicitely solvable toy model from RL

18.02.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Umut Şimşekli (INRIA - École Normale Supérieure (Paris))
• Towards Building a Heavy-Tailed Theory of Stochastic Gradient Descent for Deep Neural Networks
• see video

25.02.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Zhiyuan Li (Princeton University)
• Reconciling Modern Deep Learning with Traditional Optimization Analyses: The Intrinsic Learning Rate
• see video (MPI MiS internal only)

02.03.2021, 18:00 Uhr, Deep Learning Theory Group Seminar

• Pierre Bréchet (MPI MiS, Leipzig)
• Discrete Wasserstein Gradient Flow (Bytes)

04.03.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Marcus Hutter (DeepMind and Australian National University)
• Learning Curve Theory
• see video

09.03.2021, 18:00 Uhr, Deep Learning Theory Group Seminar

• Hanna Tseran (MPI MiS, Leipzig)
• Topology for Characterizing Capacity of Neural Networks (Bytes)

11.03.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Spencer Frei (Department of Statistics, UCLA)
• Generalization of SGD-trained neural networks of any width in the presence of adversarial label noise
• see video

16.03.2021, 18:00 Uhr, Deep Learning Theory Group Seminar

• Yuguang Wang (MPI MiS, Leipzig)
• Information bottleneck for GNNs

18.03.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Ryo Karakida (AIST (National Institute of Advanced Industrial Science and Technology), Tokyo)
• Understanding Approximate Fisher Information for Fast Convergence of Natural Gradient Descent in Wide Neural Networks
• see video

25.03.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Pratik Chaudhari (University of Pennsylvania)
• Learning with few labeled data
• see video

30.03.2021, 18:00 Uhr, Deep Learning Theory Group Seminar

• Pradeep Banerjee (MPI MiS, Leipzig)
• Information complexity and flat minima

01.04.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Roi Livni (Tel Aviv University)
• Regularization, what is it good for?
• see video

06.04.2021, 18:00 Uhr, Deep Learning Theory Group Seminar

• Pierre Bréchet (MPI MiS, Leipzig)
• Curvature in the representation space of GANs (Bytes)

08.04.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Cristian Bodnar (University of Cambridge)
• Fabrizio Frasca (Twitter)
• Weisfeiler and Lehman Go Topological: Message Passing Simplicial Networks
• see video

13.04.2021, 18:00 Uhr, Deep Learning Theory Group Seminar

• Hanna Tseran (MPI MiS, Leipzig)
• Neural Architecture Search (Topical overview)

15.04.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Stanislav Fort (Stanford University)
• Neural Network Loss Landscapes in High Dimensions: Theory Meets Practice

20.04.2021, 18:00 Uhr, Deep Learning Theory Group Seminar

• Yuguang Wang (MPI MiS, Leipzig)
• Generalization of GNNs

22.04.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Matus Jan Telgarsky (UIUC)
• Recent advances in the analysis of the implicit bias of gradient descent on deep networks
• see video

29.04.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Mehrdad Farajtabar (DeepMind)
• Catastrophic Forgetting in Continual Learning of Neural Networks
• see video

06.05.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Haizhao Yang (Purdue University)
• A Few Thoughts on Deep Learning-Based PDE Solvers
• see video

13.05.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Eirikur Agustsson (Google research)
• Universally Quantized Neural Compression

27.05.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Sebastian Reich (University of Potsdam and University of Reading)
• Statistical inverse problems and affine-invariant gradient flow structures in the space of probability measures
• see slides

01.06.2021, 18:00 Uhr, Deep Learning Theory Group Seminar

• Pierre Bréchet (MPI MiS, Leipzig)
• Dynamical analysis of GANs (Bytes)

03.06.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Stanisław Jastrzębski (Molecule.one and Jagiellonian University)
• Reverse-engineering implicit regularization due to large learning rates in deep learning
• see video

08.06.2021, 18:00 Uhr, Deep Learning Theory Group Seminar

• Jiayi Li (University of California, Los Angeles)
• Graph Coarsening with neural networks (Paper review)

10.06.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Isabel Valera (Saarland University and MPI for Intelligent Systems)
• Algorithmic recourse: theory and practice
• see video

17.06.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Ziv Goldfeld (Cornell University)
• Scaling Wasserstein distances to high dimensions via smoothing
• see video

22.06.2021, 18:00 Uhr, Deep Learning Theory Group Seminar

• Hanna Tseran (MPI MiS, Leipzig)
• Neural architecture search (Bytes)

29.06.2021, 18:00 Uhr, Deep Learning Theory Group Seminar

• Johannes Müller (MPI MiS, Leipzig)
• Metrics and continuity in reinforcement learning (Paper review)

01.07.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Dominik Janzing (Amazon Research, Tuebingen, Germany)
• Why we should prefer simple causal models

13.07.2021, 18:00 Uhr, Deep Learning Theory Group Seminar

• Pierre Bréchet (MPI MiS, Leipzig)
• Convergence in GANs: The discrete case

15.07.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Hossein Mobahi (Google Research)
• Self-Distillation Amplifies Regularization in Hilbert Space
• see video

22.07.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Soledad Villar (Johns Hopkins University)
• Scalars are universal: Gauge-equivariant machine learning, structured like classical physics
• see video

27.07.2021, 18:00 Uhr, Deep Learning Theory Group Seminar

• Jiayi Li (University of California, Los Angeles)
• Rademacher Complexity for Adversarial Robust Generalization (Paper review)

24.08.2021, 18:00 Uhr, Deep Learning Theory Group Seminar

• Hanna Tseran (MPI MiS, Leipzig)
• Decision Boundary of Neural Networks (Topic overview)

26.08.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Guodong Zhang (University of Toronto)
• Differentiable Game Dynamics: Hardness and Complexity of Equilibrium Learning
• see video

02.09.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Matthew Trager (Amazon US)
• A Geometric View of Functional Spaces of Neural Networks

07.09.2021, 18:00 Uhr, Deep Learning Theory Group Seminar

• Johannes Müller (MPI MiS, Leipzig)
• NNs for PDEs: Differences to supervised learning (Bytes)

09.09.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Shaowei Lin (working in a stealth startup)
• All you need is relative information
• see video

14.09.2021, 18:00 Uhr, Deep Learning Theory Group Seminar

• Benjamin Bowman (University of California, LA)
• Can we use the NTK outside the NTK regime?

16.09.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Pavel Izmailov (New York University)
• What Are Bayesian Neural Network Posteriors Really Like?
• see video

21.09.2021, 18:00 Uhr, Deep Learning Theory Group Seminar

• Pierre Bréchet (MPI MiS, Leipzig)
• Role(s) of the regularization in the convergence of GANs

23.09.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Huan Xiong (MBZUAI, Abu Dhabi)
• On the Number of Linear Regions of Convolutional Neural Networks with Piecewise Linear Activations
• see video

28.09.2021, 18:00 Uhr, Deep Learning Theory Group Seminar

• Hanna Tseran (MPI MiS, Leipzig)
• How to Start Training: The Effect of Initialization and Architecture (Paper review)

07.10.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Roger Grosse (University of Toronto)
• Self-tuning networks: Amortizing the hypergradient computation for hyperparameter optimization
• see video

12.10.2021, 18:00 Uhr, Deep Learning Theory Group Seminar

• Johannes Müller (MPI MiS, Leipzig)
• Degree of the value function (Progress report)

14.10.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Mathias Drton (Technical University of Munich)
• A Bayesian information criterion for singular models
• see video

21.10.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• David Stutz (MPI for Informatics, Saarbrücken)
• Relating Adversarial Robustness and Weight Robustness Through Flatness
• see video

04.11.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Hyeyoung Park (Kyungpook National University)
• Effect of Geometrical Singularity on Learning Dynamics of Neural Networks
• see video (MPI MiS internal only)

11.11.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Christoph Hertrich (TU Berlin)
• Towards Lower Bounds on the Depth of ReLU Neural Networks
• see video

18.11.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Daniel Soudry (Technion)
• Algorithmic Bias Control in Deep learning
• see video

25.11.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Michael Joswig (TU Berlin)
• Geometric Disentanglement by Random Convex Polytopes
• see video

02.12.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Ekaterina Lobacheva (HSE University)
• Maxim Kodryan (HSE University)
• On the Periodic Behavior of Neural Network Training with Batch Normalization and Weight Decay
• see video

16.12.2021, 17:00 Uhr, Math Machine Learning seminar MPI MIS + UCLA

• Zhi-Qin John Xu (Shanghai Jiao Tong University)
• Occam’s razors in neural networks: frequency principle in training and embedding principle of loss landscape
• see video