Lauri Lovén

MELD allows autonomous agents to reconcile conflicting knowledge without losing information, achieving superior recall and efficiency compared to traditional methods.

Intelligence derived from edge data can now be treated as a first-class entity, transforming how we manage and share insights across devices.

Non-IID data traits can drastically influence Federated Learning convergence, with measurable impacts that practitioners can leverage for more stable training outcomes.

Behavioral identifiability of language model adapters reveals a consistent trade-off with distributional similarity, challenging assumptions about community conditioning in LLMs.

Forget incentive compatibility, can you even trust the marketplace operator?

Decentralized, market-based resource allocation in edge-cloud computing can actually *beat* centralized orchestration, even when faced with broker failures and network partitions.