What is the machine learning technique used by IBM Watson? IBM says that 'IBM Watson is a technology platform that uses natural language processing and machine learning...' But what is the name of the machine learning technique they are using. I assumed it was artificial neural networks but I can't find any reference regarding the same.
 A: In this Quora question, an IBM researcher answers that the statistical modeling done by Watson is mostly logistic regression, although this is built into a very complex hierarchy. They also state that the statistical modeling is not necessarily the most challenging part of Watson. 
I have also independently heard the same thing reported by another Watson researcher. 
A: Here is a great article by Robert L. Blum: http://www.bobblum.com/ESSAYS/COMPSCI/Watson.html
Among other things it contains useful pointers:

The best high level article on Watson was written by Ferrucci's IBM team (mirror: http://www.aaai.org/ojs/index.php/aimagazine/article/view/2303), and appeared in AI Magazine in 2010.  They describe the evolution of their DeepQA architecture from earlier QA systems including IBM's Piquant and NIST's Aquaint. I also recommend KurzweilAI's Amara Angelica's interview with IBM's Eric Brown.
The best tv show on Watson was this PBS NOVA, Smartest Machine on Earth. These brief videos about Watson on IBM's website are also excellent. And, if you missed the Jeopardy shows, you'll want to see this clip of the champs in a preliminary bout.

Also, from https://ai.stackexchange.com/questions/17/what-things-went-into-ibms-watson-supercomputer


*

*Author: iliasfl

*Licensed under cc by-sa 3.0 with attribution required



There was a special issue (http://ieeexplore.ieee.org/xpl/tocresult.jsp?isnumber=6177717) that discussed lot of internals of Watson. I will try to give a brief overview.
  In general the overall architecture was a ensemble of independent "experts" that all were trying to answer the same question in parallel. Each expert was trained on different datasets and/or was based on different algorithms. The overall decision was made by a combination of the responses of the independent experts, like regression where inputs are the experts' responses. This allowed a structured and incremental development of the machinery: researchers proposed new experts, these were added to the system, if overall accuracy was increased they were kept otherwise they were discarded. Some NLP preprocessing was taking place (correct me if I'm wrong) to make questions digestible by the experts.

