After two silent grant rejections, researcher posts federated learning paper on arXiv
wandedob · x · 2026-11-03
Researcher Anders Eklund says his grant application was rejected in March and a longer version rejected again in April — with zero feedback either time — so he posted the paper, Federated Learning for Unpaired Multimodal Data, on arXiv for public comment.
The paper proposes training a global multimodal transformer across decentralized nodes holding disjoint, strictly private modalities (e.g., sensor data on one node, text logs on another). Key ideas: a small public anchor set with Gram matrices and centered kernel alignment to align modalities without transmitting private samples; subspace-stabilized fine-tuning for huge transformers; and precision-weighted averaging that downweights uncertain nodes via uncertainty estimates.
More from Research
- Counting mRNAs in living bacteria: the 2005 Golding paper that overturned the Poisson model of transcription — NikoMcCarty · 2026-11-03
- AAAI 2027 to host Neuro for AI & AI for Neuro workshop in Montreal, papers due in 3 weeks — naturecomputes · 2026-11-03
- DeepSeek V4.1 Flash 1017 doubles ARC-AGI-3 score to 5.81%, leads open models — teortaxesTex · 2026-11-03
- Critics tear apart "Weight-Space Programming": Atari test spoon-feeds answers, Qwen case fails on Gemma — anshulkundaje · 2026-11-03
- Cohere paper: adjacent tokens share 20-45% of experts, making speculative decoding cheap even at batch size 1 — rohanpaul_ai · 2026-11-03
- Editing 131k weights to fix looping: activation steering yields free benchmark gains — xennygrimmato_ · 2026-11-03