What are your top concerns with Edge computing?

Question

Answer

The number one thing that I think they should be thinking about is security. Securing that data in ways that don't make you vulnerable. Some of it comes back to rethinking things a little bit and checking your current way of doing things at the door and being open to a different concept. You can read about the almost daily disclosures of breaches of personal information, credit card information, social security numbers etc. That is nothing if you consider it against the backdrop of IoT data which might be actual patient records,the algorithms that are controlling a drone, traffic signal information, or the pedestrian detection algorithms in your autonomous vehicle. Now you're moving into the scope and scale of life and death. So security is a deep conversation and the protection, safeguarding, and securitization of data on multiple threat vectors is a very important topic and probably the most important topic that's got to be addressed at the very outset of discussing an Edge product, solution, app, or whatever you're contemplating.

Answer

Security would of course be number one in my book and I believe it should be on device and built in. But from the perspective of Edge and looking at data, I would say the choreography of data from the first point to the second point to the third. Look at not technical data, but process steps. Go back and revisit your processes because what happened ten years ago, two years ago, or even yesterday is not the same process that you need for tomorrow. So in order to plan the flow of data, whether it's across a factory or multiple factories and multiple companies, you first have to look at how the process is run and where those data points between physical and digital actually happen.

Answer

I would like to comment on Machine Learning on edge in context of security concerns.

With ML on edge (using federated learning techniques) the concerns around security and PII data getting shared, gets addressed to some level. Today's smart phones and similar IOT and edge devices are capable enough to run localised machine learning models and instead of sharing data to a central GPU, they can share the hyper parameters that gives desired inference scores.

This addresses many concerns. Firstly data privacy, data stays locally on edge. Secondly, data transfer overheads (latency, cost) are not there. Hyper parameters information is fraction of actual data. Thirdly, the scale of tuning can be massive, which result in better training. Lastly, data drift situations can be better addressed as data is more recent and more "real" from field.

What are your top concerns with Edge computing?

Content you might like

If you are a current SAP customer, when do you plan to migrate to SAP S/4HANA?

What do you think about the rumors that Jony Ive and Sam Altman, the minds behind iconic Apple designs and OpenAI, are teaming up to create the "iPhone of AI?"

Has anyone found a AI solution to screen IT contracts to reduce the time it takes to onboard a new vendor solution?

Have you modernized your Problem Management process to include machine learning/AI capabilities in identifying problems and potential solutions? Taken IT Problem Management beyond IT into the business processes? Merged SRE practices into ITIL Problem Managment?

What's a quick win for optimizing spend? What's been your focus or "lowest hanging fruit"?