search

ices-eg / DIG

ICES Data and Information Group
Creative Commons Zero v1.0 Universal
2 stars 1 forks source link

Lack of transferability of Machine learning training datasets between regions #517

Open sjurl opened 11 months ago

sjurl commented 11 months ago
Aspect Description
Challenges Processes developed in different regions may require new/different/expanded training sets or source data for machine learning to perform effectively across wider areas.

legacy entry

OliverWilliamsDataManager commented 6 months ago

I was a little unclear as to the original focus of this issue, it mentions 'for machine learning' but I decided to make it more generally about 'data enablement' for many purposes hence the title addition. This is a challenge only I think as data use / reuse is a much broader topic which is perhaps too obvious to require an opportunity?

I deemed impact as high because if all data are intercalibrated (or simply standardised at source) then data enablement is far more simple.

I deemed likelihood as medium, this is am less sure about because I am not 100% how much scope ICES has to influence data processes within working groups if members of these are reluctant or unable to comply with recommendations. On the other hand this is a vital task so in a sense has to be done.