Send email Copy Email Address
Research Group

Stich

Optimization is at the heart of many machine learning algorithms. The focus of our research group is on fundamental research on optimization techniques for training machine learning models. The research topics include decentralized methods (collaborative learning, federated learning), but also adaptive stochastic methods, generalization performance, deep theory of deep learning, and privacy and security in machine learning.

Head of Group

Sebastian Stich

Email

Phone

Address

Im Oberen Werk 1
66386 St. Ingbert (Germany)

Members

Placeholder

Xiaowen Jiang
Placeholder

Ali Zindari
Placeholder

Md Sakhawat Hosain
Placeholder

Yuan Gao
Placeholder

Parham Yazdkhasti

Sohom Muhkerjee
Placeholder

Anton Rodomanov
Placeholder

Shireen

Yasin Esfandiari

Most Recent Publications

Year 2023

2023-12-11

Adaptive SGD with Polyak stepsize and Line-search: Robust Convergence and Variance Reduction

Conference / Medium

NeurIPS
Conference on Neural Information Processing SystemsConference on Neural Information Processing Systems

Tags
Algorithmic Foundations and Cryptography
Authors
Download Visit Detail Page
2023-07-27

Why Random Pruning Is All We Need to Start Sparse

Conference / Medium

ICML
International Conference on Machine Learning40th International Conference on Machine Learning

Tags
Algorithmic Foundations and Cryptography
Authors
Download Visit Detail Page
2023-07-23

Revisiting Gradient Clipping: Stochastic bias and tight convergence guarantees

Conference / Medium

ICML
Proceedings of the 40th International Conference on Machine LearningInternational Conference on Machine Learning (ICML)

Tags
Algorithmic Foundations and Cryptography
Authors
Download Visit Detail Page
2023-07-23

Special Properties of Gradient Descent with Large Learning Rates

Conference / Medium

ICML
Proceedings of the 40th International Conference on Machine LearningInternational Conference on Machine Learning (ICML)

Tags
Algorithmic Foundations and Cryptography
Authors
Download Visit Detail Page
2023-06-20

On the effectiveness of partial variance reduction in federated learning with heterogeneous data

Conference / Medium

CVPR
Proceedings of IEEE Conference on Computer Vision and Pattern RecognitionIEEE Conference on Computer Vision and Pattern Recognition (CVPR)

Tags
Algorithmic Foundations and Cryptography
Authors
Download Visit Detail Page