Applying Machine Learning to Intelligence Problems (Part 4)

All Machine Learning algorithms require large amounts of training data (“experiences”) in order to learn. They recognize patterns in the training data and develop a “model” of the world being described by the data. Reinforcement learning is slightly different from other techniques in that the training data is not given to the algorithm, but rather, is generated in real time via interactions with and feedback from the environment. But in all cases, as new training data comes in, the algorithm is able to improve and refine the model. This is particularly well suited for solving three types of intelligence problems, namely: classification, recommendation/ prediction and clustering.

Image: Three types of problems that can be solved by Machine Learning

First, tackling classification problems involves making observations, such as identifying objects in images and video or recognizing text and audio. From an intelligence perspective this could help tremendously in reading and analyzing large volumes of text to determine the “who”, “what”, “when” and “where”.  When searching for information, ML can for example identify “trending themes” or the most widely “talked about” subjects in a body of text. Human analysts can then use this intelligence to determine the “why” and decide on the appropriate course of action.  

Second, Machine Learning can also be used for recommendations and predictions such as estimating the likelihood of events and forecasting outcomes. This could, for example, be very powerful for an analyst wanting to predict the likelihood of a competitor launching a new product at a particular point in time.

Third, Machine Learning can be used to segment data into clusters according to association. An intelligence application of this is the ability to automatically detect and extract sentiment on a more accurate level, e.g. is this news positive or negative, and is this event a strength or a weakness for us?

Applied to a Competitive Intelligence environment, this can have extremely positive outcomes. ML can automate many areas of data synthesis and can significantly help in arriving at faster Times to Insights over traditional methods:

·      Automation classification analytics surrounding vast quantities of data, performed real time, within on-line tools and databases

·      Templates / analytics algorithm foundations can be used as a basis for ML data population. Again, automating this process can allow it to be as close to real time as possible

·      Clustering techniques for pattern recognition amongst various data promotes quicker development of content and deliverables

Tool structure is key here – utilizing a robust IT based internet-interactive ML capability will automate, provide better content, and yield better quality and faster results. Additionally, learning algorithms can be built into the tool design so the functions can essentially be continually improved, automated and more efficient. 

More to come...

This was part FOUR in our series, based on the article “Machine Learning Implications for Intelligence and Insights”, written by Jesper Martell, Comintelli, and Paul Santilli, Hewlett Packard Enterprise.

WEBINAR: Machine Learning Implications for Intelligence and Insights

Presenters: Jesper Martell, Comintelli & Paul Santilli, Hewlett Packard Enterprise

Thursday, October 26, 2017, 10:00am Eastern, Hosted by    SCIP

This interactive webinar describes how Machine Learning (ML) can be applied to solve intelligence problems.

1. What is a Machine Learning algorithm?
2. How can new ML/AI technologies augment our intelligence capabilities?
3. What are some of the challenges and risks of ML?

Read more and register here!