The Project Aims to explore Caution AI/ Explainable AI/ General AI

The scope of the project is limited currently to the following visual comparisons as required

• A visual representation of interactions of various nerual networks at a lower dimensions .
• Analaysis of how those models might have a concept of "limits" of a model .
• A formal definition of what a limit might be and how weights, biases and limits can be defined .

Environment and Setup

• The project uses conda environments and jyuptier notebooks heavily to make reports and analysis
• A toy model with 2 input neurons , three hidden neurons and finally three ouput neurons are the center of all these experiments.
• First pure random number with fixed ranges are generated and finally that is trained over the nueral network.
• Similary the problem is transfered over to a visual task of selecting color of pixels.A data collector is custom made for this task. This allows us to make a comparison to "true" intelligence or human intelligence.
• Then with the findings.Its applied on models designed for the MNIST data set to find those points.
• Evaluation all models gives us a domain wide idea r

WIP

• MNIST utils.