b'Data trendsData trendspattern recognition in images and signalschecked), what about surveys where since the wonders of digital computersonly 95% or less of measurements are allowed the first steps towardsconsidered quality compliant. Is one automated processing of gravity infifth of the resulting map suspect? 1967 (Hansen et al., 1967). Little surpriseIf so, which parts? Geostatistically a that ML is full of the same words andconfidence measure on the input and processes. Regularisation, matrices andoutput matters, as geological models error minimisation on super computersare spun by methods that have more crunching kernels and optimisation.in common with computer science Rename the work function to objectivethan geology. Each pixel and polygon function, and you are almost at a Dougcould have an individual measurement Oldenburg ASEG talk. for various survey aspects such as flight speed, height, and the variousthe impending AI apocalypseinstrument related measures, which Tim Keepingis just what industry has beenML could probably be trained to Associate Editor for geophysical datagenerate.management and analysis ordering for email@example.com This (sort of) ties into numbers, and But what can it do for meor forthe methods we think we can use. you? I looked at Googles introductoryI am currently being extra wary of Machine learning Tensorflow lesson (Google, 2019), whichtreating supposedly similar numbers resolves 60 000 pictures of clothing intowith the same techniques. I am a user The renaming of statistics to data10 classes with 87% accuracy. It fell shortof the open source Python package science was easy to follow, butof the psychologically classic 95% mark,Fatiando, whose author has just why a separate name for Machinebut it could still help us get through theannounced he has submitted a paper Learning (ML)? Technology bloggersdry parts faster. I could train the softwareto Geophysics questioning the gravity have described machine learning aswith radiometric line data, and then useanomaly as a harmonic function programmes that use data to writeit to identify odd readings in a survey.and the mathematical methods that themselves, accumulating rules-of- It could consume various layers withinentails. Go to Leonardo Uiedas home thumb knowledge in the form ofa multi layered storage format, such aspage https://www.leouieda.com/statistical models to apply to data,the Multi Resolution Raster featured apapers/use-the-disturbance.html to especially visual data. Legend says afterfew issues ago, and form opinions byread more.Googles AlphaGo beat the worlds bestpredicting holes in data, or flag inferred Go players, a version stripped of itscontradictions when new data is added.Referencesknowledge, but armed with machineML supposedly promises all this without learning hardware, supposedly learnt(much) programming. Hansen, D.A., Heinrichs, Jr. W.E., Holmer, to play from scratch and could beat itsData science also promises to findR.C., MacDougall, R.E., Rogers, G.R., older brother in under a week. Cue theyet unknown relationships, but aSumner, J.S., and Ward, S.H. 1967. spooky music. concern for users of said models willMining Geophysics Volume II, Theory. Lucky for geophysicists andbe the cumulative effect of the variousdoi:10.1190/1.9781560802716geostatisticians, the impending AIconfidence measurements of eachGoogle. 2019. Train your first neural apocalypse is just what industryhasdata set used. If Tensorflow operatesnetwork: basic classification. https://www.been ordering for years. The industrywith 87% success on a 100% labelledtensorflow.org/tutorials/keras/basic_has been at the front line of applieddata set (i.e. the answers can beclassification (accessed 06 03 2019).37 PREVIEW APRIL 2019'