← Volver a la portada
Oceanía · Actualidad general · iTWire

Why Australian industry's next AI wave will run on the device, not in the data centre

Publicado por iTWire · Australia hace 1 h
Guest Opinion
Guest Opinion · iTWire

Resumen detallado

Most conversations about artificial intelligence in business still start in the data centre. Train a model in the cloud, send data up, get a decision back. That works well for...

GUEST OPINION: Most conversations about artificial intelligence in business still start in the data centre. Train a model in the cloud, send data up, get a decision back. That works well for an office in Sydney. It works far less well for a mine site in the Pilbara, a cattle station in Queensland or a pumping station at the end of a long rural line.

Much of Australian industry operates a long way from reliable infrastructure. That geography is about to shape how AI gets deployed here, and it points away from the cloud and towards the device itself.

Cloud AI carries three hidden assumptions: a fast connection, a cheap connection and a connection that is always there. Remote Australian operations can rarely count on all three at once.

Latency is the first casualty. A safety system that spots a worker entering a danger zone has to react in milliseconds, not after a round trip to a server in another state. Bandwidth is the second. Streaming high-resolution video from dozens of cameras over satellite or rural links is expensive, and most of those frames show nothing worth analysing. Reliability is the third. When a storm or a cut cable takes the link down, a cloud-dependent system simply stops making decisions.

Then there is data sovereignty. Many organisations, from health providers to critical infrastructure operators, would rather not ship sensitive footage or operational data offshore at all.

With on-device AI, the model runs on the camera, the sensor gateway or the machine controller itself. Decisions happen locally, in real time, whether or not the link to head office is up. Only results, alerts and the occasional sample image need to travel, which cuts bandwidth to a fraction of a raw video stream. Sensitive data can stay on site.

The gains are not theoretical. In one project by engineering firm Yalantis, a vision model running at the edge filters the output of a PCB manufacturer's automated optical inspection line in real time, cutting false rejects by 35 per cent.

Movi…

Inicio del artículo, extracto de iTWire. Para leer el resto, abra el artículo en el sitio del medio.

Leer el artículo completo en iTWire Inicie sesión para guardar

MarketNews muestra el titular y el resumen publicados por el medio. El artículo completo se lee en el sitio del editor. (itwire.com)

También te puede interesar