Dilsad Er

A cute little animation: a critically damped harmonic oscillator becomes unstable with integral control if the gain is too high. Here, at K_i = 2, a Hopf bifurcation occurs: two poles of the transfer function enter the right-hand s-plane and the closed-loop system becomes unstable.

I'm flying to Michigan today to present our new paper "A Pontryagin Perspective on Reinforcement Learning" at L4DC, where it has been nominated for the Best Paper Award! We ask the question: is it possible to learn an open-loop controller via RL? 🧵

We’re happy to announce our fantastic speakers for the 3rd TWiML Workshop!✨
Prof. Dr. Anna Levina, Apl. Prof. Dr. Britta Dorn, and Dr. @nnludwig.bsky.social who will be sharing their amazing research with us.

The 3rd TWiML Workshop is happening! Join us for great talks, discussions, and community connections.

Science is for everyone! No matter your gender, everyone is welcome.

Mark your calendars!📢
The third TWIML workshop is coming soon! 🎉
📅 When? March 14, 2025, from 9:00 to 13:00
📍 Where? Maria von Linden Strasse 6, Lecture Hall
Join us for an exciting half-day workshop featuring incredible speakers and insightful discussions. Stay tuned, more news is coming!

Big congratulations to Michael on winning the Collen Prize from Trinity College Dublin for his thesis and this excellent work! It’s been a pleasure supervising him and collaborating on this research.

Controlling Participation in Federated Learning with Feedback We address the problem of client participation in federated learning, where traditional methods typically rely on a random selection of a small subset of clients for each training round. In contrast, ...

The paper is under review. You can read the preprint here: arxiv.org/abs/2411.19242

FedBack models server-client communication as a dynamical system to achieve global stability and reduce client drift, even under tough network conditions.

We introduce FedBack, a federated optimization algorithm based on ADMM, which treats client participation as a control problem.

Controlling Participation in Federated Learning with Feedback We address the problem of client participation in federated learning, where traditional methods typically rely on a random selection of a small subset of clients for each training round. In contrast, ...

💻🤖 Let’s optimize client participation in federated learning using feedback and control!

arxiv.org/abs/2411.19242

Michael Cummins will be presenting the paper, “Controlling Participation in Federated Learning with Feedback,” at the Flower AI Summit 2025! (flower.ai)

A network for women to support each other There are still far fewer women than men working in the field of machine learning. With "TWiML", Claire Vernade has fostered a community for women in Tübingen.

With “Tübingen Women in Machine Learning”, Claire Vernade and her team have created a local community. Their aim is to empower women and make them more visible. Read more on our blog: www.machinelearningforscience.de/en/a-network... @twiml.bsky.social #TWiML #WiML