FLock.io used federated learning with Sarawak AI Centre to train local models while preserving public-sector data privacy.
FLock.io, a London-based AI start-up, says it has demonstrated sovereign artificial intelligence capabilities in Sarawak, Malaysia.
The company, in a project with Sarawak AI Centre (SAIC), using federated learning to train models on local data without sharing raw datasets, set out to show that public institutions can develop AI locally without relying on foreign tech giants and while keeping data within jurisdictional control.
“Privacy is non-negotiable in the public sector. Governments hold highly sensitive data: health records, tax history, welfare, police and criminal files. Data breaches can be disastrous. There is significant appetite among governmental bodies – such as in Sarawak – to use data to improve public services through privacy-preserving techniques like federated learning," said FLock CEO Jiahao Sun.
FLock.io said it used two technical approaches: federated learning to collaboratively train models without exchanging data, and distributed inference that shards large models across smaller GPUs in networked local hardware to avoid centralised data centres.
The AI experiment trained the models on the local language Sarawak Malay, and ran on distributed state hardware at SAIC’s lab in Kuching.
Related reading
Professor Patrick Then, CEO of Sarawak AI Centre, added: “We have so many indigenous languages which are not captured at all in most of these large language models – it’s a huge challenge.”
“Now, the vision in the next five years is for Sarawak to be a fully AI nation.”
The recap
FLock.io deployed federated learning with Sarawak AI Centre.
Trained models on Sarawak Malay using distributed local hardware.
SAIC targets developed, high-income, sustainable state status by 2030.
