Probabilistic Automated Numerical Analysis in Machine learning and Artificial intelligence

Project Information

PANAMA

Grant agreement ID: 757275

DOI 10.3030/757275

Project closed

EC signature date 17 October 2017

Start date 1 March 2018

End date 28 February 2023

Funded under

• EXCELLENT SCIENCE - European Research Council (ERC)

Total cost € 1 450 000,00

EU contribution € 1 450 000,00

Coordinated by

EBERHARD KARLS UNIVERSITAET TUEBINGEN
Germany

CORDIS provides links to public deliverables and publications of HORIZON projects.

Links to deliverables and publications from FP7 projects, as well as links to some specific result types such as dataset and software, are dynamically retrieved from OpenAIRE .

Publications

Conference proceedings (46)

Calibrated Adaptive Probabilistic ODE Solvers

Author(s): Bosch, Nathanael; Hennig, Philipp; Tronarp, Filip
Published in: International Conference on Artificial Intelligence and Statistics, Issue 130, 2021, Page(s) 3466-3474
Publisher: PMLR
DOI: 10.48550/arxiv.2012.08202

Counterfactual mean embeddings

Author(s): Krikamol Muandet, Motonobu Kanagawa, Sorawit Saengkyongam, Sanparith Marukatat
Published in: The Journal of Machine Learning Research, Issue 22 (1), 2021, Page(s) 7322-7392
Publisher: The Journal of Machine Learning Research

BackPACK: Packing more into backprop

Author(s): Dangel, Felix; Kunstner, Frederik; Hennig, Philipp
Published in: ICLR, Issue 8, 2020
Publisher: ICLR

High-Dimensional Gaussian Process Inference with Derivatives

Author(s): de Roos, Filip; Gessner, Alexandra; Hennig, Philipp
Published in: International Conference on Machine Learning, Issue 139, 2021, Page(s) 2535-2545
Publisher: PMLR

Probabilistic ODE solutions in millions of dimensions

Author(s): Nicholas Krämer, Nathanael Bosch, Jonathan Schmidt, Philipp Hennig
Published in: International Conference on Machine Learning, Issue 162, 2022, Page(s) 11634-11649
Publisher: PMLR

Model-based Kernel Sum Rule: Kernel Bayesian Inference with Probabilistic Models

Author(s): Nishiyama, Yu; Kanagawa, Motonobu; Gretton, Arthur; Fukumizu, Kenji
Published in: Machine Learning, Issue 109, 2020, Page(s) 939–972
Publisher: Springer Link

Probabilistic Numerical Method of Lines for Time-Dependent Partial Differential Equations

Author(s): Nicholas Krämer, Jonathan Schmidt, Philipp Hennig
Published in: Proceedings of The 25th International Conference on Artificial Intelligence and Statistics, Issue 151, 2022, Page(s) 625-639
Publisher: PMLR

An Infinite-Feature Extension for Bayesian ReLU Nets That Fixes Their Asymptotic Overconfidence

Author(s): Kristiadi, Agustinus; Hein, Matthias; Hennig, Philipp
Published in: Advances in Neural Information Processing Systems, Issue 34, 2021, Page(s) 18789-18800
Publisher: Curran Associates Inc
DOI: 10.48550/arxiv.2010.02709

Integrals over Gaussians under Linear Domain Constraints

Author(s): Gessner, Alexandra; Kanjilal, Oindrila; Hennig, Philipp
Published in: Proceedings of Machine Learning Research, 2020, Page(s) 2764-2774
Publisher: PMLR

Convergence Guarantees for Adaptive Bayesian Quadrature Methods

Author(s): Kanagawa, Motonobu; Hennig, Philipp
Published in: Advances in Neural Information Processing Systems (NeurIPS 2019), Issue 32, 2019
Publisher: Advances in Neural Information Processing Systems 32 (NeurIPS 2019)
DOI: 10.48550/arxiv.1905.10271

Descending through a Crowded Valley - Benchmarking Deep Learning Optimizers

Author(s): Schmidt, Robin M.; Schneider, Frank; Hennig, Philipp
Published in: International Conference on Machine Learning, Issue 139, 2021, Page(s) 9367-9376
Publisher: PMLR
DOI: 10.48550/arxiv.2007.01547

Model Selection for Simulator-based Statistical Models: A Kernel Approach

Author(s): Kajihara, Takafumi; Kanagawa, Motonobu; Nakaguchi, Yuuki; Khandelwal, Kanishka; Fukumiziu, Kenji
Published in: Machine Learning, Issue 109, 2020, Page(s) 939–972
Publisher: Springer
DOI: 10.48550/arxiv.1902.02517

Fenrir: Physics-Enhanced Regression for Initial Value Problems

Author(s): Filip Tronarp, Nathanael Bosch, Philipp Hennig
Published in: International Conference on Machine Learning, Issue 162, 2022, Page(s) 21776--21794
Publisher: PMLR

DeepOBS: A Deep Learning Optimizer Benchmark Suite

Author(s): Schneider, Frank; Balles, Lukas; Hennig, Philipp
Published in: International Conference on Learning Representations, 2019
Publisher: International Conference on Learning Representations
DOI: 10.48550/arxiv.1903.05499

The Geometry of Sign Gradient Descent

Author(s): Balles, Lukas; Pedregosa, Fabian; Le Roux, Nicolas
Published in: ICLR 2020, 2020
Publisher: ICLR

Fast Predictive Uncertainty for Classification with Bayesian Deep Networks

Author(s): Hobbhahn, Marius; Kristiadi, Agustinus; Hennig, Philipp
Published in: Uncertainty in Artificial Intelligence, Issue 180, 2022, Page(s) 822-832
Publisher: PMLR
DOI: 10.48550/arxiv.2003.01227

Preconditioning for scalable Gaussian process hyperparameter optimization

Author(s): Jonathan Wenger, Geoff Pleiss, Philipp Hennig, John Cunningham, Jacob Gardner
Published in: International Conference on Machine Learning, Issue 162, 2022, Page(s) 23751-23780
Publisher: PMLR

Learnable Uncertainty under Laplace Approximations

Author(s): Kristiadi, Agustinus; Hein, Matthias; Hennig, Philipp
Published in: Proceedings of the Thirty-Seventh Conference on Uncertainty in Artificial Intelligence, Issue 161, 2021, Page(s) 344-353
Publisher: PMLR
DOI: 10.48550/arxiv.2010.02720

Laplace Redux -- Effortless Bayesian Deep Learning

Author(s): Daxberger, Erik; Kristiadi, Agustinus; Immer, Alexander; Eschenhagen, Runa; Bauer, Matthias; Hennig, Philipp
Published in: Advances in Neural Information Processing Systems (NeurIPS), Issue 34, 2021, Page(s) 20089-20103
Publisher: Advances in Neural Information Processing Systems (NeurIPS)
DOI: 10.48550/arxiv.2106.14806

Probabilistic solvers enable a straight-forward exploration of numerical uncertainty in neuroscience models

Author(s): Jonathan Oesterle, Nicholas Krämer, Philipp Hennig, Philipp Berens
Published in: Journal of Computational Neuroscience, Issue 50 (4), 2022, Page(s) 485-503
Publisher: Springer US
DOI: 10.1007/s10827-022-00827-7

Being Bayesian, Even Just a Bit, Fixes Overconfidence in ReLU Networks

Author(s): Kristiadi, Agustinus; Hein, Matthias; Hennig, Philipp
Published in: ICML, Issue 4, 2019
Publisher: ICML

Probabilistic Linear Solvers for Machine Learning

Author(s): Wenger, Jonathan; Hennig, Philipp
Published in: Advances in Neural Information Processing Systems, Issue 33, 2020, Page(s) 6731 - 6742
Publisher: Curran Associate Inc.
DOI: 10.48550/arxiv.2010.09691

Limitations of the empirical Fisher approximation for natural gradient descent

Author(s): Kunstner, Frederik; Hennig, Philipp; Balles, Lukas
Published in: Advances in Neural Information Processing Systems 32, Issue 32, 2019, Page(s) {4158--4169
Publisher: Curran Associates, Inc.

Pick-and-mix information operators for probabilistic ODE solvers

Author(s): Nathanael Bosch, Filip Tronarp, Philipp Hennig
Published in: International Conference on Artificial Intelligence and Statistics, Issue 151, 2022, Page(s) 10015-10027
Publisher: PMLR

A Fourier State Space Model for Bayesian ODE Filters

Author(s): Kersting, Hans; Mahsereci, Maren
Published in: Workshop on Invertible Neural Networks, Normalizing Flows, and Explicit Likelihood Models, ICML, 2020
Publisher: ICML

Cockpit: A Practical Debugging Tool for the Training of Deep Neural Networks

Author(s): Schneider, Frank; Dangel, Felix; Hennig, Philipp
Published in: Advances in Neural Information Processing Systems, Issue 34, 2021, Page(s) 20825-20837
Publisher: Curran Associates Inc.
DOI: 10.48550/arxiv.2102.06604

Kernel Recursive ABC: Point Estimation with Intractable Likelihood

Author(s): Takafumi Kajihara, Motonobu Kanagawa, Keisuke Yamazaki, and Kenji Fukumizu
Published in: Proceedings of the International Conference on Machine Learning, Issue 35, 2018, Page(s) 2400-2409
Publisher: PMLR (Proceedings of Machine Learning Research

Convergence Guarantees for Adaptive Bayesian Quadrature Methods

Author(s): Motonobu Kanagawa, Philipp Hennig
Published in: Advances in Neural Information Processing Systems (NeurIPS), Issue 32, 2019, Page(s) 6234--6245
Publisher: Curran Associates, Inc.

Dissecting Adam: The Sign, Magnitude and Variance of Stochastic Gradients

Author(s): Lukas Balles, Philipp Hennig
Published in: Proceedings of the 35th International Conference on Machine Learning (ICML), Issue 35, 2018, Page(s) 404--413
Publisher: PMLR

DeepOBS: A Deep Learning Optimizer Benchmark Suite

Author(s): Schneider, Frank; Balles, Lukas; Hennig, Philipp
Published in: International Conference on Learning Representations (ICLR), Issue 7, 2019
Publisher: ICLR

Limitations of the empirical Fisher approximation for natural gradient descent

Author(s): Frederik Kunstner, Philipp Hennig, Lukas Balles
Published in: Advances in Neural Information Processing Systems (NeurIPS), Issue 32, 2019, Page(s) 4158--4169
Publisher: Curran Associates, Inc.

Active Probabilistic Inference on Matrices for Pre-Conditioning in Stochastic Optimization

Author(s): Filip de Roos, Philipp Hennig
Published in: Proceedings of the 22nd International Conference on Artificial Intelligence and Statistics (AISTATS), Issue 22, 2019, Page(s) 1448--1457
Publisher: PMLR

Fast and Robust Shortest Paths on Manifolds Learned from Data

Author(s): Georgios Arvanitidis, Soren Hauberg, Philipp Hennig, Michael Schober
Published in: Proceedings of the 22nd International Conference on Artificial Intelligence and Statistics (AISTATS), Issue 22, 2019, Page(s) 1506--1515
Publisher: JMLR

Active Multi-Information Source Bayesian Quadrature

Author(s): Alexandra Gessner, Javier Gonzalez, Maren Mahsereci
Published in: Conference on Uncertainty in Artificial Intelligence (UAI), Issue 35, 2019
Publisher: UAI

Convergence rates of Gaussian ODE filters

Author(s): Hans Kersting; Timothy Sullivan; Philipp Hennig
Published in: Statistics and computing, Issue 30 (6), 2020, Page(s) 1791-1816
Publisher: Springer US
DOI: 10.1007/s11222-020-09972-4

Being a Bit Frequentist Improves Bayesian Neural Networks

Author(s): Kristiadi, Agustinus; Hein, Matthias; Hennig, Philipp
Published in: International Conference on Artificial Intelligence and Statistics, Issue 151, 2022, Page(s) 529-545
Publisher: PMLR
DOI: 10.48550/arxiv.2106.10065

Linear-Time Probabilistic Solutions of Boundary Value Problems

Author(s): Krämer, Nicholas; Hennig, Philipp
Published in: Advances in Neural Information Processing Systems 34 (NeurIPS 2021), Issue 34, 2021, Page(s) 11160-11171
Publisher: Advances in Neural Information Processing Systems 34 (NeurIPS 2021)

Probabilistic DAG Search

Author(s): Grosse, Julia; Zhang, Cheng; Hennig, Philipp
Published in: Uncertainty in Artificial Intelligence, Issue 161, 2021, Page(s) 1424-1433
Publisher: PMLR
DOI: 10.48550/arxiv.2106.08717

A Probabilistic State Space Model for Joint Inference from Differential Equations and Data

Author(s): Schmidt, Jonathan; Krämer, Nicholas; Hennig, Philipp
Published in: Advances in Neural Information Processing Systems 34 (NeurIPS 2021), Issue 34, 2021, Page(s) 12374-12385
Publisher: Advances in Neural Information Processing Systems 34 (NeurIPS 2021)
DOI: 10.48550/arxiv.2103.10153

ProbNum: Probabilistic Numerics in Python

Author(s): Jonathan Wenger, Nicholas Krämer, Marvin Pförtner, Jonathan Schmidt, Nathanael Bosch, Nina Effenberger, Johannes Zenn, Alexandra Gessner, Toni Karvonen, François-Xavier Briol, Maren Mahsereci, Philipp Hennig
Published in: 2021
Publisher: arXiv
DOI: 10.48550/arxiv.2112.02100

Bayesian Quadrature on Riemannian Data Manifolds

Author(s): Fröhlich, Christian; Gessner, Alexandra; Hennig, Philipp; Schölkopf, Bernhard; Arvanitidis, Georgios
Published in: International Conference on Machine Learning, Issue 139, 2021, Page(s) 3459-3468
Publisher: PMLR
DOI: 10.48550/arxiv.2102.06645

Integrals over Gaussians under Linear Domain Constraints

Author(s): Gessner, Alexandra; Kanjilal, Oindrila; Hennig, Philipp
Published in: International Conference on Artificial Intelligence and Statistics. Proceedings of Machine Learning Research (PMLR), Issue 1, 2020
Publisher: MLR Press
DOI: 10.48550/arxiv.1910.09328

Modular Block-diagonal Curvature Approximations for Feedforward Architectures

Author(s): Dangel, Felix; Harmeling, Stefan; Hennig, Philipp
Published in: Proceedings of the Twenty Third International Conference on Artificial Intelligence and Statistics, Issue 2, 2020, Page(s) 799-808
Publisher: PMLR

Differentiable Likelihoods for Fast Inversion of 'Likelihood-Free' Dynamical Systems

Author(s): Kersting, Hans; Krämer, Nicholas; Schiegg, Martin; Daniel, Christian; Tiemann, Michael; Hennig, Philipp
Published in: International Conference on Machine Learning (ICML), Issue 11, 2020
Publisher: ICML

Resnet after all: Neural odes and their numerical solution

Author(s): Katharina Ott, Prateek Katiyar, Philipp Hennig, Michael Tiemann
Published in: International Conference on Learning Representations, 2021
Publisher: International Conference on Learning Representations

Conjugate Gradients for Kernel Machines

Author(s): Bartels, Simon; Hennig, Philipp
Published in: Journal of Machine Learning Research, 2020, Page(s) 1-42
Publisher: Journal of Machine Learning Research

Peer reviewed articles (5)

Bayesian ODE solvers: the maximum a posteriori estimate

Author(s): Filip Tronarp; Simo Särkkä; Philipp Hennig
Published in: Statistics and Computing, 31(3), Issue 31, 2021, ISSN 0960-3174
Publisher: Kluwer Academic Publishers
DOI: 10.1007/s11222-021-09993-7

Convergence Analysis of Deterministic Kernel-Based Quadrature Rules in Misspecified Settings

Author(s): Motonobu Kanagawa, Bharath K. Sriperumbudur, Kenji Fukumizu
Published in: Foundations of Computational Mathematics, 2019, Page(s) 1-40, ISSN 1615-3375
Publisher: Springer Verlag
DOI: 10.1007/s10208-018-09407-7

On the positivity and magnitudes of Bayesian quadrature weights

Author(s): Toni Karvonen, Motonobu Kanagawa, Simo Särkkä
Published in: Statistics and Computing, Issue 29/6, 2019, Page(s) 1317-1333, ISSN 0960-3174
Publisher: Kluwer Academic Publishers
DOI: 10.1007/s11222-019-09901-0

Probabilistic solutions to ordinary differential equations as nonlinear Bayesian filtering: a new perspective

Author(s): Filip Tronarp, Hans Kersting, Simo Särkkä, Philipp Hennig
Published in: Statistics and Computing, Issue 29/6, 2019, Page(s) 1297-1315, ISSN 0960-3174
Publisher: Kluwer Academic Publishers
DOI: 10.1007/s11222-019-09900-1

Probabilistic linear solvers: a unifying view

Author(s): Simon Bartels, Jon Cockayne, Ilse C. F. Ipsen, Philipp Hennig
Published in: Statistics and Computing, Issue 29/6, 2019, Page(s) 1249-1263, ISSN 0960-3174
Publisher: Kluwer Academic Publishers
DOI: 10.1007/s11222-019-09897-7

Monographic books (1)

Probabilistic Numerics: Computation as Machine Learning

Author(s): Philipp Hennig, Michael A. Osborne, Hans P. Kersting
Published in: 2022, ISBN 9781316681411
Publisher: Cambridge University Press
DOI: 10.1017/9781316681411