By Mariam Awad
The purpose of this series of talks was to discuss how we can utilize machine learning for creating situational awareness of both intentional and naturally occurring biological incidents. One of the current hurdles in conducting biosurvillance for Bacillus anthracis and pandemic influenza include lack of tools that can rapidly structure, integrate and analyze large, disparate data with little human exposure and intervention. The speakers focused on how to apply advancement in artificial intelligence and machine learning towards analysis of data for decision makers. Topic areas ranged from detection, tracking, and forecasting events, as well as analysis of genomic data as it relates to understanding and characterizing a biological event.
One speaker discussed how we can integrate data from multiple streams. One cheap stream is scanning social media accounts to track the spread of disease. This data, while unstructured, can be useful by incorporating a program that would give the raw data parameters such as location, population and duration. One limitation for this potential tool is its inability to predict unknown diseases. How can we conduct surveillance for a bug we don’t know? How do you know what data is significant?
A group from Carnegie Mellon University has designed a program that detects new novel symptoms through a “key-word” scan. The program assigns a significance value to new symptoms as well as old. If those new symptoms are occurring over and over in the data, then scientists are able to pay attention to them and assess whether or not they are significant. For example, if the program finds “tainted coffee” while scanning the data, it will assign it a significance value and will be added to the “memory” for future encounters. This tells us what to search from within the data. The potential impact of this tool in biosurvillance and early detection is huge.