GIS Meetup & Educational Seminar Summary #1
Presented on: July 31st, 2019
Presenters:
- Orhun Aydin (oayadin@esri.com),
- Lauren Bennett, (lbennett@esri.com),
- Alberto Nieto (anieto@esri.com)
Overview
Machine Learning (ML) is currently all the buzz, but it is important not to get distracted by shiny objects. We should solve problems with the simplest tool available and not treat ML as a tool looking for a problem.
There are three linked concepts from broad to narrow: Artificial Intelligence (AI), Machine Learning (ML), and Deep Learning (DL).
It is useful to distinguish between “traditional” ML and “spatial” ML. Spatial ML is concerned with shape, density, contiguity, spatial distribution, and proximity.
ML/DL is not new, but it has now become viable due to increases in processing power.
Esri's Focus
The majority of Esri’s research focuses on ML and DL to automate prediction, classification, and clustering. This talk focuses on the application of ML tools to spatial analysis.
-
Classification is the process of deciding what category an object should be assigned to based on a training dataset. ArcGIS supports the following classification techniques: maximum likelihood classification, random trees, support vector machine, and forest-based classification and regression.
-
Clustering groups objects based on similarities in their values and locations. ArcGIS includes the ML tools DBSCAN, HDBSCAN, and OPTICS for clustering. ML clustering techniques include spatially constrained multivariate clustering, multivariate clustering, density-based clustering, image segmentation, hot spot analysis and outlier analysis, and space time pattern mining.
-
Prediction is the process of predicting the value of a continuous variable. Prediction techniques available in ArcGIS include empirical Bayesian kriging, areal interpolation, EBK regression prediction, ordinary least squares regression and exploratory regression, geographically weighted regression, generalized linear regression, and forest-based classification and regression.
Active Research
Esri is actively researching DL for classification and object recognition e.g., landcover recognition from satellite imagery. ArcGIS integrates with a wide range of external frameworks including TensorFlow, IBM Watson, PyTorch, scikit-learn, Microsoft AI, and R through tools such as ArcGIS API for Python, ArcPy, R-ArcGIS Bridge, and ArcGIS Notebooks.
More Information
More information is available at:
https://spatialstats.github.io/
Reaction
This is a topic I am extremely interested in and I believe it fits well with my technical and programming background. Although this talk was a bit of a whirlwind of different tools, I was inspired to see some examples of what is possible. I would like to learn more about this area. Seeing this talk reinforces how much I still must learn in this area. However, the subject seems exciting to me and I feel comfortable with the technology. For me, the key issue is to find the right area to apply this technology and not fall into the trap of a tool looking for a problem to solve.