Section
Computer Science
Federated Learning under Intermittent Connectivity
A federated training protocol tolerant of dropped clients maintains convergence guarantees under the connectivity patterns typical of rural deployments.
Karthik Nair · Vol 1, Issue 1 · pp. 15-26
Sparse Attention Mechanisms for Low-Resource Language Models
We present a sparse attention scheme that reduces inference cost for morphologically rich, low-resource languages without measurable loss in downstream accuracy.
Ananya Rao, Devashish Kumar · Vol 1, Issue 1 · pp. 1-14