We believe the future of machine learning is decentralized — where intelligence grows collaboratively, without data ever leaving its source.
Our mission is to accelerate decentralized ML and make it the de facto paradigm for building intelligent, efficient, and privacy-preserving systems.
No central authority — just collective learning, where everyone contributes — safely, privately, and efficiently.
We push the boundaries of learning and optimization at the edge, with research to create scalable, efficient, and multimodal decentralized learning systems.
- ⚡ Federated & Split Learning — frameworks for decentralized collaboration across heterogeneous devices and institutions.
- 🧩 Foundation Model Fine-Tuning — adapting large-scale (multimodal) Foundation Models (FMs) to decentralized and resource-constrained environments.
- 🔒 Privacy-Preserving Mechanisms — integrating differential privacy, encryption, and secure aggregation into multimodal FMs.
- 🛰️ Edge & On-Device Intelligence — enabling lightweight, self-improving models that learn directly where data is generated.
- 🔄 Decentralized Optimization & Aggregation — redefining how distributed models synchronize, exchange knowledge, and evolve without central coordination.
We’re developing a growing ecosystem of open-source projects — spanning Efficiency, Adaptivity, Privacy, and Edge FMs — to accelerate the future of decentralized intelligence.
| 🌐 Project | 🧾 Description | 🎯 Focus Area | 🏛️ Venue |
|---|---|---|---|
| FedSTAR | Semi-supervised FL with adaptive reliability. | 🧭 Adaptivity | ICASSP 2022 |
| FedLN | FL under label noise. | 🧭 Adaptivity | NeurIPS 2022 Workshop |
| FedCompress | Task-adaptive model compression for efficient FL. | ⚡ Efficiency | ICASSP 2024 |
| EncCluster | Scalable FM secure aggregation through weight clustering. | 🔒 Privacy | NeurIPS 2024 Workshop |
| DeltaMask | Communication-efficient federated FM fine-tuning via masking. | ⚡ Efficiency / 🛰️ Edge FMs | ICML 2024 Workshop |
| MPSL | Multimodal FM fine-tuning via parallel SL. | 🛰️ Edge FMs / ⚡ Efficiency | IJCAI 2025 Workshop |
| MaTU | Many-task federated FM fine-tuning via unified task vectors. | 🧭 Adaptivity / 🛰️ Edge FMs | IJCAI 2025 |
| EFU | Enforcable Federated Unlearning. | 🧭 Privacy | CIKMI 2025 |
✨ Accelerating the future of decentralized intelligence — together. ✨