Felix Petersen

My primary research interest is differentiable relaxations in machine learning. In particular, I'm interested in differentiable logic gate networks for extremely efficient inference, differentiable sorting (networks) for weakly-supervised learning, differentiable rendering, as well as general stochastic gradient estimation and opti miza tion.

Currently, I am a postdoctoral researcher at Stanford University in Stefano Ermon's group.

Previously, I have been working, i.a., at the University of Konstanz, at TAU, DESY, PSI, and CERN.

News

Sep 2024 We have 3 accepted papers at NeurIPS 2024, one of which is an oral! 🎊

Sep 2024 Presenting my research at MIT-IBM, Harvard, Boston University, and CMU.

Apr 2024 Our second Differentiable Almost Everything workshop has been accepted for ICML 2024 🎻!

Mar 2024 Our paper "Grounding Everything: Emerging Localization Properties in Vision-Language Transformers" was accepted to CVPR 2024! Congrats to Walid!

Jan 2024 Our paper "Uncertainty Quantification via Stable Distribution Propagation" was accepted to ICLR 2024!

Research

The focus of my research lies in differentiable relaxations of non-differentiable operations in machine learning. Differentiable relaxations enable a plethora of optimization tasks: from optimizing logic gate networks [1, 2] and optimizing through the 3D rendering pipeline [3, 4, 5] to differentiating sorting and ranking [6, 7] for supervised [8] and self-supervised [9] learning. Beyond differentiable algorithms, this branches out into various domains including stochastic gradient estimation [10, 11], analytical distribution propagation [12], second-order optimization [10, 13], uncertainty quantification [12], fairness [14, 15], and efficient neural architectures [1, 2, 16].

	Convolutional Differentiable Logic Gate Networks Felix Petersen, Hilde Kuehne, Christian Borgelt, Julian Welzel, Stefano Ermon in Proceedings of the 38th International Conference on Neural Information Processing Systems (NeurIPS 2024) (Oral) YouTube Code (will be updated soon) NeurIPS Oral
	TrAct: Making First-layer Pre-Activations Trainable Felix Petersen, Christian Borgelt, Stefano Ermon in Proceedings of the 38th International Conference on Neural Information Processing Systems (NeurIPS 2024) YouTube
	Newton Losses: Using Curvature Information for Learning with Differentiable Algorithms Felix Petersen, Christian Borgelt, Tobias Sutter, Hilde Kuehne, Oliver Deussen, Stefano Ermon in Proceedings of the 38th International Conference on Neural Information Processing Systems (NeurIPS 2024) YouTube
	Generalizing Stochastic Smoothing for Differentiation and Gradient Estimation Felix Petersen, Christian Borgelt, Aashwin Mishra, Stefano Ermon arXiv preprint
	Grounding Everything: Emerging Localization Properties in Vision-Language Transformers Walid Bousselham, Felix Petersen, Vittorio Ferrari, Hilde Kuehne in Proc. of the Conference on Computer Vision and Pattern Recognition (CVPR 2024) Demo Code
	Uncertainty Quantification via Stable Distribution Propagation Felix Petersen, Aashwin Mishra, Hilde Kuehne, Christian Borgelt, Oliver Deussen, Mikhail Yurochkin in Proc. of the International Conference on Learning Representations (ICLR 2024)
	Learning by Sorting: Self-supervised Learning with Group Ordering Constraints Nina Shvetsova, Felix Petersen, Anna Kukleva, Bernt Schiele, Hilde Kuehne in Proc. of the International Conference on Computer Vision (ICCV 2023) Code
	Neural Machine Translation for Mathematical Formulae Felix Petersen, Moritz Schubotz, Andre Greiner-Petter, Bela Gipp in Proc. of the 61st Annual Meeting of the Association for Computational Linguistics (ACL 2023) YouTube
	ISAAC Newton: Input-based Approximate Curvature for Newton's Method Felix Petersen, Tobias Sutter, Christian Borgelt, Dongsung Huh, Hilde Kuehne, Yuekai Sun, Oliver Deussen in Proc. of the International Conference on Learning Representations (ICLR 2023) YouTube Code
	Deep Differentiable Logic Gate Networks Felix Petersen, Christian Borgelt, Hilde Kuehne, Oliver Deussen in Proceedings of the 36th International Conference on Neural Information Processing Systems (NeurIPS 2022) Code
	Domain Adaptation meets Individual Fairness. And they get along. Debarghya Mukherjee, Felix Petersen, Mikhail Yurochkin, Yuekai Sun in Proceedings of the 36th International Conference on Neural Information Processing Systems (NeurIPS 2022)
	Learning with Differentiable Algorithms Felix Petersen PhD thesis (summa cum laude), University of Konstanz
	Differentiable Top-k Classification Learning Felix Petersen, Hilde Kuehne, Christian Borgelt, Oliver Deussen in Proceedings of the 39th International Conference on Machine Learning (ICML 2022) YouTube Code
	GenDR: A Generalized Differentiable Renderer Felix Petersen, Christian Borgelt, Bastian Goldluecke, Oliver Deussen in Proceedings of the Conference on Computer Vision and Pattern Recognition (CVPR 2022) YouTube Code
	Monotonic Differentiable Sorting Networks Felix Petersen, Christian Borgelt, Hilde Kuehne, Oliver Deussen in Proceedings of the International Conference on Learning Representations (ICLR 2022) YouTube Code / diffsort library
	Style Agnostic 3D Reconstruction via Adversarial Style Transfer Felix Petersen, Hilde Kuehne, Bastian Goldluecke, Oliver Deussen in Proceedings of the IEEE Winter Conf. on Applications of Computer Vision (WACV 2022) YouTube
	Learning with Algorithmic Supervision via Continuous Relaxations Felix Petersen, Christian Borgelt, Hilde Kuehne, Oliver Deussen in Proceedings of the 35th International Conference on Neural Information Processing Systems (NeurIPS 2021) YouTube Code / AlgoVision library
	Post-processing for Individual Fairness Felix Petersen, Debarghya Mukherjee, Yuekai Sun, Mikhail Yurochkin in Proceedings of the 35th International Conference on Neural Information Processing Systems (NeurIPS 2021) YouTube Code
	Differentiable Sorting Networks for Scalable Sorting and Ranking Supervision Felix Petersen, Christian Borgelt, Hilde Kuehne, Oliver Deussen in Proceedings of the 38th International Conference on Machine Learning (ICML 2021) YouTube
	AlgoNet: C^∞ Smooth Algorithmic Neural Networks Felix Petersen, Christian Borgelt, Oliver Deussen
	Pix2Vex: Image-to-Geometry Reconstruction using a Smooth Differentiable Renderer Felix Petersen, Amit H. Bermano, Oliver Deussen, Daniel Cohen-Or
	Towards Formula Translation using Recursive Neural Networks Felix Petersen, Moritz Schubotz, Bela Gipp in Proceedings of the 11th Conference on Intelligent Computer Mathematics (CICM), 2018
	LaTeXEqChecker - A framework for checking mathematical semantics in LaTeX documents Felix Petersen Presented in the Special Session of the 11th Conference on Intelligent Computer Mathematics (CICM), 2018 Slides

Media Coverage

	MIT Technology Review: The next generation of neural networks could live in hardware
	AIM Analytics India Magazine: This New Logic Gate Network Reduces Inference Speed to Only 4 Nanoseconds
	Local newspaper article about me as a 19 years old PhD student: Dieser junge Mann ist 19 Jahre alt – und schreibt gerade seine Doktorarbeit. Über ein hochbegabtes Talent der Konstanzer Uni [English PDF]
	campus.kn article about my Schülerstudium (studying while in school): English: “At university, knowledge counts more than age” – “Im Gespräch” interviews Felix Petersen, who completed a Schülerstudium German: „An der Uni ist die Expertise wichtiger als das Alter“ – Im Gespräch mit Schülerstudium-Absolvent Felix Petersen
	German TV: Short interview about my participation at Jugend forscht: „Jugend forscht“: Hamburger Schüler stellen Forschungsprojekte vor
	Südkurier article about my work at DESY: Streng geheimes Forschungsprojekt: 17-jähriger Informatik-Student entwickelt neuartigen Röntgenlaser

Based on ✩.