Tuesday, November 16, 2010

Homework 10 Computing Futures

Homework 10


1.      Honestly I do not think that in the near future
present to +5 years that there will be a whole lot of development with
robotics. I do see in the next 15-40 years them taking a robot that has a
knowledge base such as the one I want to create and the ability to understand
and communicate with natural language and combine the two to make a sort of
android. One that would be able to help the human in certain knowledge areas.





2.      I feel that for the presentation, since it will
be an hour long for ours that we should include a power point somewhere. It
would discuss how Neural networks and Fuzzy logic work so that people would
have an understanding of what we are doing. Once we had gone over that we would
discuss the input and desired output. As well as, discuss the pre-processing of
data and delve into the decision process and long-term memory of the neural
network. Finally we will want to give a live demonstration of the program with
several inputs to show that it is working or not working (depending on the
outcome).





3.      I believe at this time it is important to
recognize that Matt and I have altered our project somewhat. We will still be
desiring the same output (Being able to automatically text-mine Type, Sub-Type,
Stat, and Stat Dimension). However, our means of attaining the output are
different. Initially we were going to mine tables in Science Direct files and
parse through them effectively creating our own database of facts with the
above four mentioned variables. This was mostly just parsing of semi-structured
data and proved tedious and minimally rewarding. I found it more challenging to
instead take a sentence as input and be able to intelligently mine through for
the same four variables. This will give a exponentially greater repository to
mine. If successful it should also signify a greater breakthrough to the
text-mining community. We plan on using a data structure very similar to a
neural network. The major difference is the way the layers are connected. Layer
1 connects to both the layer 2 and layer 3. This is contradictory to the
traditional and mainstream feed forward networks. We will also be adding a
genetic algorithm to the back propagation so that if the training gets in a
bind it will automatically spring out due to the mutation in the values caused
by the genetic algorithm.

No comments:

Post a Comment