Open Call Winners

TERMINET Open Call Winners!

TERMINET proudly announces the 4 winners of TERMINET-OC2022! On behalf of the TERMINET Open Call Project Committee, we would like to thank all the participants for submitting their proposals! In total 18 Open Call proposals were received from 9 countries.

Congratulations to the 4 winners!

Topic 1 winner

Title: Privacy-Preserving Federated Learning Application for diagnosis of coMmunication disordErs iN Child development

Acronym: FLAMENCO

Country: Romania

Approach & Objectives:

The project aims to design and develop novel federated learning models, model architectures, algorithms, deployment strategies, and optimization techniques that can provide advanced and optimized FL solutions tailored to the environments and requirements described by TERMINET.

The FLAMENCO project will apply Federated Learning to an existing software application suite, which is used to diagnose communication skills development in children, detecting potential deficiencies timely and accurately. The software suite collects data from a child’s responses to an animation game, along with heart rate readings obtained from a smartwatch. The data is then sent and stored into a cloud data hub and analyzed using AI-based classification techniques. The outcome is a risk indicator that suggests the likelihood of a child developing any learning and communication disorders.

As the data collected is sensitive and personal, and the predictive model requires continuous and incremental training, more sophisticated techniques are required. The FLAMENCO project will employ several algorithms to ensure users’ privacy and prediction accuracy. First, the project will use Fully Homomorphic Encryption during the model aggregation step to protect user data from potential breaches. Second, client selection techniques will be utilized to remove users with corrupted or missing data to improve the model’s predictive accuracy. Third, state-of-the-art aggregators will manage data imbalance, ensuring that the Federated Learning model can converge effectively.

The project outcome will simulate a Federated Learning process using real-world data from IoT edge devices and incorporating the proposed extensions. The findings will demonstrate the potential of this approach to pave the way towards personalised healthcare solutions while respecting patient privacy.

Further info/links:

• https://opsys.ro/
• https://duth.gr/en

The FLAMENCO project (Privacy-Preserving Federated Learning Application for diagnosis of coMmunication disordErs iN Child development) adopted TERMINET’s FL approach, developing more advanced Machine Learning models for predicting the risk of communication skills’ development in children. More information about the FLAMENCO project can be found in TERMINET’s D8.5 Validation, Evaluation and Lessons Learned of 3rd Parties.

• FLAMENCO open-source repository: https://github.com/nikopavl4/FLAMENCO-Project
• FLAMENCO public dataset in Zenodo: https://zenodo.org/records/10492892
• FLAMENCO public dataset in IEEE DataPort: https://ieee-dataport.org/documents/flamenco-learning-disabilities-dataset
• Research Paper: Towards Energy-Aware Federated Traffic Prediction for Cellular Networks
• Research Paper: Federated Learning for Early Dropout Prediction on Healthy Ageing Applications
• Interview of winner of TOPIC 1 of TERMINET’s open call: Privacy-Preserving Federated Learning Application for diagnosis of coMmunication disordErs iN Child development: https://youtu.be/i3duDfYPN3Y
• FLAMENCO project provides an overview of their project: https://youtu.be/Qey_2KlzHX4?si=e-Ss2aSiq3tCx8c7
• FLAMENCO project provides a more technical presentation of their project – Technical Overview: https://youtu.be/mOL3DnduYxw?si=kfan8ALhlBsW4WJ3
• FLAMENCO project demonstrates the impact of their solution to the end user: https://youtu.be/6YA98WPy6EI?si=JMwO0H_eAIGnQ6FK
• FLAMENCO poster