Over 85% of data science projects fail to go beyond testing and into production. If everyone is starting a Machine Learning / Artificial Intelligence project, where is it going wrong?
With this post it should help you to focus on the types of problems that AI is good at solving. In order to actually benefit from using AI to increase automation, you will need to have the right data, have enough data, have a methodology that can be defined with data points, and be creative in understanding how to apply or craft solutions for parts of your team’s workflow
The aim is to extract the most relevant sentences from a larger piece of text. Extractive models select whole sentences to include in a summary; and abstractive models select parts of sentences that get combined with computer generated words and parts of other sentences.
Given a set of documents (or articles, pieces of content, patents, customer profiles, etc.) identify similar content within the database you are searching.
You may not require A.I. for building a better estimation model unless you have enough pieces of data that your opening Excel crashes your computer because your spreadsheet has more than 100,000 rows and many variables. Machine Learning can optimize for hundreds of dimensions when you aren’t sure of the importance of all the variables.
Think of modeling all the factors that affect an individual’s health using a lifetime of population data vs. modeling home prices based on zip code, # of bedrooms and size.
Think Cyber Security. Your IT department has a sense of the normal activity of all of your company’s employees, and needs to be alerted when there are actual risks to your company. Example: hackers are infiltrating your network and stealing your company’s IP.
While it may be impossible to anticipate in advance what the actual form of attack will look like, an anomaly detection model can be deployed to look for deviation in behavior like a 10,000% spike in log-ins or outbound traffic going to a server located internationally.