A team o scientists fae da University o Glesga his unveiled a grundbrakin saftware model dat promises tae mak da analysis o geospatial data mair eco-freendly. Da model, named ‘GeoAggregator’, uises machine lairnin tae lessen da computin pouer needed tae haandle vast an intricate geospatial datasets.

As GPS an satellite data become mair common, da warld gaiders massive amoonts o geospatial information ilka day. Yet, makkin sense o dis data is nae sma feat. Traditional methods an even some AI models struggle tae grasp da complex spatial ties ithin da data. GeoAggregator tackles these hurdles wi a lichtweicht AI model dat deftly examines hoo nearby places affect ane anidder an hoo paitterns shift fae ane spot tae anidder. It is faster, mair scalable, an demands fewer resoorces than its predecessors, appenin doors for researchers an policymakkers alike.

Ae staundout feature o GeoAggregator is its Gaussian-biased local attention mechanism. Dis allous da model tae zero in on nearby data pynts while haadin da bigger picture in mind. Sic an approach boosts predictions for various spatial data issues, fae forecastin air pollution tae spotin hoosin price trends an studyin poverty spreid.

Da model’s Cartesian attention mechanism keeps it licht while ensurin heich accuracy. Even as datasets grow, GeoAggregator processes dem wi ease, a feat mony traditional AI models cannae match. Tests shaw dat GeoAggregator performs as weel as or better than ither leadin methods, uisin far fewer resoorces. Dis maks it a practical chyce for real-warld tasks whaar speed an efficiency are key.

Da researchers hae shared deir code as appen-soorce, urgins ither fowk tae uise an refine it. As interest in location-based data swells, tools like GeoAggregator can turn complex geospatial information intae meaningfu insichts aboot hoo fowk an places interact.

Dr. Mingshu Wang fae da University o Glesga’s School o Geographical & Earth Sciences, a co-author o da study, pynts oot da model’s potential: “GeoAggregator merks a muckle lowp in makkin complex data analysis mair efficient an accessible. We did aa oor data analysis on a single laptop, shawin juist hoo effective dis tool is.”

Da team is warkin on an appen-soorce Python package tae mak da model freely available. Dis effort is pairt o ongaan PhD research by Rui Deng, wha will continue tae enhance da model ower da neist twa years. Dr. Ziqi Li, an Assistant Professor at Florida State University, is a co-author o da paper.

This research is pairt o da University o Glesga’s growin expertise in AI an machine lairnin, follaein da laanch o deir Centre for Data Science & AI. Da School o Geographical & Earth Sciences his recently received fundin tae lead da Exascale computin for Earth, Environmental, an Sustainability Solutions consortium, whit will train new PhD students tae develop saftware for environmental applications on neist-gen computin systems. Da team’s paper will be presented at da AAAI Conference on Artificial Intelligence dis week.