🔒 Ready to lock down your IoT edge? Discover how federated learning lets devices train AI while keeping data private. A security game‑changer! #AI #IoT #FederatedLearning #EdgeComputing #Security 🔗

This line of research was produced at @csunivie.bsky.social, @univie.ac.at with support from @fwf-at.bsky.social and CHIST-ERA.

#NeuromorphicComputing #SpikingNeuralNetworks #EdgeAI #FederatedLearning #DistributedSystems

半教師あり連合型ASRのための実践的レシピ：オンライン疑似ラベルとサーバー更新安定化

ASRにおける半教師あり連合学習の課題を克服し、実践的な安定化手法と疑似ラベル生成戦略を詳述。

#FederatedLearning #ASR #Semi-supervisedLearning #MachineLearning #Privacy

Hybrid quantum-classical federated learning achieves 87.57% accuracy using only 12 trainable parameters versus 1,002+ for classical alternatives, enabling privacy-preserving multi-party ML without centralizing raw data.

#QuantumMachineLearning #FederatedLearning #Research

Roger Federated's runtime wrapper now closes the loop: each day it pulls your federation's aggregated update and attaches it as a LoRA adapter (--lora/--enable-lora) - model file untouched, no copy stored. #federatedlearning #opensource #selfhosted https://github.com/roger-federated/roger-federated

5 pharma firms trained an AI on 20,167 private protein structures, without sharing one. Accuracy: 35.6% → 52.1%. Analysis: #AIDrugDiscovery #AlphaFold #FederatedLearning #CompBio #OpenFold3 #AISB-1-Fed #MELLODDY

artificialscience.org/2026/09/fede...

Exploring the realities of Federated Learning in prostate cancer diagnosis
📊 14 datasets analyzed from 10 countries
🔍 FL aids mostly with moderate data, limits when data is scarce or abundant

#FederatedLearning #Healthcare #DataPrivacy https://tnyp.me/USDeKGGt/s

MoE federated learning gotcha: naive FedAvg merges ALL expert weights, even ones a client never activated - corrupting knowledge in untouched experts. FLEx (arXiv 2506.00965) fixes it: aggregate only non-expert params, graft a personalized expert from local activation. #FederatedLearning #MoE #FOSS

FederatedSkill (arXiv 2606.03143): trading raw trajectories for shared skill diffs hit 44.4% higher success, 37.5% less compute, privacy intact. Roger Federated's federations already trade adapters, not rollouts. #federatedlearning #opensource https://github.com/roger-federated/roger-federated

We the adopters of #FederatedLearning would like to acknowledge the people who are all too familiar with the light that bounces off their chrome tuning keys, whether they're on a bass or guitar.