Studying the impact of data heterogeneity in Split Learning. Specifically, we conduct a systematic analysis with Deep Neural Networks (ResNet, VGG, MobileNet) using various datasets. As a result, we identified the existence of catastrophic forgetting (CF) in the training. Finally, we proposed a new ML solution for tackling CF caused by non-IID data.
Joana Tirana (Ιωάννα Τιράνα)
Postdoc at the Bioinformatics group of WUR
Utrecht, the Netherlands
joana.tirana@wur.nl
GitHub ,Google Scholar, Linkedin
Small Bio
Machine Learning researcher specialized in distributed training like split learning and federated learning, and optimization for resource-constrained systems. Experienced in both academic and industrial research, with strong programming skills.
Full CV: CV-pdf
Programming Skills
- Languages: Python, C, C++, Java, SQL, Matlab
- Machine Learning: PyTorch, Huggingface, LibTorch, Tensorflow
- Cloud Tools: Docker, Kubernetes, AWS CLI, Ansible
- Parallel Programming: CUDA, OpenMP
- Operating systems: Linux, MacOS, Android
- Optimization: Gurobi, cvxpy
- Other: GIT, Buildbot
NEWS
- 01/April/2026: PostDoc @ WUR Started my postdoc at BIF group, working for the PRIORITY project
- 23/February/2026: PhD Viva@UCD Successfully defended my Phd! See post Linkedin
- 25/November/2025: Today I started my one-month secondment visit at TU Delft Faculty of Technology, Policy and Management, hosted by Dr. Y. (Aaron) Ding, as part of the Marie Curie ENSURE 6G program research program.
- 10/November/2025: AAAI 2026: Our paper "Data Heterogeneity and Forgotten Labels in Split Federated Learning" got accepted at the main technical track! Check here git-repo
Experience
Postdoc at the Bioinforatics Group at Wageningen University & Research, The Netherlands
April 2026 - Current
supervisors: Marnix Medema, Justin van der Hooft, and Daniel Probst
Research Intern at Telefónica Innovación Diginal, Barcelona Spain
July - December 2024
This work is part of my PhD research. During this internship, we studied the effect of Catastrophic Forgetting in Parallel Split Learning when there is high data heterogeneity. Outcome: (i) accepted in AAAI 2026 [C4], and (ii) accepted by Telefonica's Patent Office
supervisors: Dimitra Tsigkari, David Solans Noguero, Nicolas Kourtellis
PhD Visit, TU Delft , The Netherlands
April - August 2023
This research visit is part of the PhD. We studied Parallel Split Learning from a more theoretical perspective. In detail, inspired by the parallel machine problem, we built a new model that fully describes the system. Outcome: The publications (INFOCOM 2024) [C2] and (TMC 2025) [J1].
supervisors: George Iosifidis, Dimitra Tsigkari, Dimitris Chatzopoulos
Software Engineer Intern, 3DEXCITE -- Dassault Systems, Munich Germany
July - December 2021
Topic of the project: Deployment of DStellar in Outscale and analyzing performance.
Automatic deployment in cloud using AWS and Ansible. Also, I built and gathered results using Buildbot. Learnt working in an Agile scrum team.
Education
Phd in Computer Science
University College Dublin, Ireland
January 2022 - Decemeber 2025
Topic: Decentralized and distributed Machine Learning (ML) for resource-constrained devices, such as mobile and IoT devices. Throughout my PhD, I have studied the challenges of Federated and Split Learning, built frameworks for supporting such operations, and developed optimization algorithms that improve the system's performance and training. In general, my interests are in building systems for distributed ML and optimizing ML training under challenges.
Supervisor: Ass. Profesor Dimitris Chatzopoulos
Integrated Master in Electrical and Computer Engineering
University of Thessaly, Greece
September 2016 - September 2021
Bachelor's and Master's in Electrical and Computer Engineering. Main focus on Computer Engineering with core knowledge of Software and Hardware.
Indicative subjects: Programming I&II, Concurrent Programming, Computer Organization and Design, Distributed Systems, Networking, High Performance Computing Systems, Database Systems, Machine Learning, NeuroFuzzy Programming.
Total Mark: 8.9/10, Graduated with Honors
Thesis Title: Support for Parallel Drone-based Task Execution at Multiple Edge Points thesis link (english version)
List of publications
Conferences/Workshops
[C4]: Tirana, J. Tsigkari, D., & Noguer S. D., Kourtellis, N. (2026, January). Data Heterogeneity and Forgotten Labels in Split Federated Learning. In Proceedings the AAAI Conference.
[C3]: Tirana, J. Lalis, S., & Chatzopoulos, D. (2025, March). Estimating the Training Time in Single-and Multi-Hop Split Federated Learning. In Proceedings of the 8th International Workshop on Edge Systems, Analytics and Networking (pp. 37-42).
[C2]: Tirana, J., Tsigkari, D., Iosifidis, G., & Chatzopoulos, D. (2024, May). Workflow optimization for parallel split learning. In IEEE INFOCOM 2024-IEEE Conference on Computer Communications (pp. 1331-1340). IEEE.
[C1]: Tirana, J., Pappas, C., Chatzopoulos, D., Lalis, S., & Vavalis, M. (2022, July). The role of compute nodes in privacy-aware decentralized ai. In Proceedings of the 6th International Workshop on Embedded and Mobile Deep Learning (pp. 19-24).
Journals
[J2]: Tirana, J., Lalis, S., & Chatzopoulos, D. (2026) Implementation and Evaluation of Multi-Hop Parallel Split Learning. IEEE Access, 14, 9419-9434
[J1]: Tirana, J., Tsigkari, D., Iosifidis, G., & Chatzopoulos, D. (2025). Minimization of the Training Makespan in Hybrid Federated Split Learning. IEEE Transactions on Mobile Computing, (01), 1-18.
Book Chapters
[B2]: Byabazaire, J.,Tirana, J., Chouliaras, A., Koutsos, V., Aslanidis, T., Panagiotidis, I., & Chatzopoulos, D. (2025). Deep learning and the Internet of Things: Applications, challenges and opportunities. Internet of Things A to Z: Technologies and Applications, Second Editions. In Press. Wiley and Sons.
[B1]: Tirana, J., & Chatzopoulos, D. (2025). Split learning and synergetic inference: When IoT collaborates with the cloud-edge continuum. In Advances in the Internet of Things (pp. 203-227). CRC Press.
Dissertations
[D2]: Tirana, J. (2026). Towards Split Federated Learning Optimization: Architectures, Efficiency, and Non-IID Challenges (PhD thesis in Computer Science @ UCD) -- -- supervisor: As. Prof. Dimitris Chatzopoulos
[D1]: Tirana, J. (2021). Support for Parallel Drone-based Task Execution at Multiple Edge Points. Mater's Thesis for University of Thessaly (Dimploma for Electrical and Computer Engineering @ UTH) -- supervisor: Prof. Spyros Lalis
Research & Coding Project
• Tools: PyTorch
Code (8 stars GitHub)
In this work, we propose SplitPipe, a Machine Learning as a Service (MLaaS) modular and extensible framework for collaborative and distributed training. SplitPipe processes high-level tasks (e.g., with the model’s description that will be trained) and orchestrates the training process based on a novel Split Learning (SL) protocol. Additionally, SplitPipe supports multihop SL-based training that enhances data privacy and relaxes memory demands.
• Tools: C++ and LibTorch, devices: Raspberry Pi and Jetson
Code (16 stars GitHub)
In this work, we consider a parallel SL system with multiple helper nodes. Specifically, we focus on orchestrating the workflow of this system, which is critical in highly heterogeneous systems. In particular, we formulate the joint problem of client-helper assignments and scheduling decisions to minimize the training makespan. We propose a solution method based on the decomposition of the problem by leveraging its inherent symmetry.
• Tools: Python, Gurobi, cvxpy
Code (23 stars GitHub)
Developed a distributed system consisting of a server in the cloud and multiple servers on edge nodes. Each edge node is located near a group of drones, with direct access to them. Edge-nodes can process the generated data in parallel and independently of each other. The system offers users a shell interface through which one can initiate tasks to specific edge nodes and afterwards combine the results. The communication between the server and the edges is done without any user intervention. Also, created an estimation model using metrics that were extracted from experimental testing.
• Tools: Python, Docker, ardupilot
Code (6 stars GitHub)
Build multiple distributed computing systems during Bachelor's and Master's studies. Some indicative examples are: a distributed computing environment with transparent migration and load balancing, distributed system for Uniform Reliable multicast communication with synchronous view.
• Tools: Java, Unix libraries for networking