Machine Learning is going to kick ass - from both a tech 'wow!' standpoint and from a competitive 'pow!' standpoint. If you are a business owner who is interested in or, better yet, has started to explore the technology then good for you. If you are, on the other hand, a professional who spends your day digitising parcel boundaries from aerial imagery or updating labels on a map then it's time to pay attention.
Machine Learning or 'cognitive computing' is, along with Artificial Intelligence, the latest buzz term, which is becoming increasingly popular thanks to the immense processing power of 'supercomputers' such as Google's DeepMind, IBM's Watson, Microsoft's Project Oxford and the lesser known Chinese Baidu.
Although Machine Learning may seem like the stuff of sci-fi movies, it is not impossible to understand. The technology is simply the next era of computing (which began with tabulating machine in the 1800's and evolved into the modern programmable era in the 1930's). Supercomputers consist of software frameworks which are written on standard programming languages. This software runs on a computer operating system to provide distributed computing capabilities. Their hardware allows for massive parallel processing thanks to a large number of servers clusters and many terabytes of RAM.
Geobreadbox tries to bring something new to a geospatial industry which now relies on social media as its main communication platform. It's about my favourite 3 subjects: geography, technology and learning and this should explain why geospatial is mentioned so often. Geography is a powerful subject in the globalised world of 2017 and it helps us to make sense of lots of information. So, in order to keep it fresh and secure, I thought that a breadbox (which lies at the heart of most homes) would be useful.
The Era of Unstructured Data
It is a supercomputer's information processing capabilities which distinguishes it from its computing predecessors. While programmable systems rely on well structured data sources and file formats, supercomputers possess an almost human-like flexibility and intuition when it comes to working with unstructured data. This explains why there is a shift in the world of data management away from a reliance on relational database systems towards the use of schemaless NoSQL databases which are capable of managing today's deluge of Big Data.
Missing a comma in your CSV file? Using incorrect grammar or abbreviations in your Tweet, or even slang terms in your weekly vlog? Not not worry. Today's machines are much better placed to understand their often complicated and unpredictable human counterparts.
How does it work?
In order to understand how machine learning works you need to understand the 4 stages of human reasoning. The supercomputer first of all observes visible phenomenon in bodies of evidence presented before it. It then draws on its own knowledge or 'evidence' in order to interpret and generate hypotheses about what it is seeing. Next, it evaluates which hypotheses seem the most likely answer to the question at hand. Finally, considering all of the above, it decides on the best course of action.
In order to get to the point where a supercomputer can make decisions and providing unique insights a whole lot of work needs to be done by humans. First, the supercomputer needs to 'consume' vast quantities of data. The more information it has, the better it can evaluate problems and reason later on. Next, the supercomputer is 'trained' by humans on how to discern meaning from this information through the use of question and answer tests and other interactions and challenges. This ongoing consumption, exercising and feedback process helps the computer to better manage, evaluate and derive insights and patterns from new and updated information. After a certain point supercomputers are able to educate and train themselves - just like an adult human being can.
Machine Learning and Earth Observation
While there are countless possibilities to explore through this technology, it is not in the scope of this week's post to go into too much detail. However, the most obvious benefit to be harnessed by geospatialists from supercomputers is in creating value from today's vast quantities of Earth Observation data. This includes developing applications for agri-tech and crop management, weather forecasting, water usage monitoring as well as for Smart Cities. Machine Learning, combined with powerful remote sensing detection technologies such as spectral imaging and thermal/lidar systems will allow for detection of changes in landuse and ground movement, atmosphere, water and surface temperature, as well as a range of other 'observations' which can be placed within a wider contextual understanding of the world.
Machine learning will undoubtedly bring disruption to the geospatial and other industries. However, it will also allow these industries to solve their problems thanks to more evidence-based decision-making.
The worlds of Machine Learning, Big Data, Smart Cities and Earth Observation are converging at precisely the right time.
It's time for geospatialists to get involved in the conversation.