I spent the majority of today working on my neural network. I am having this problem with smoothing it out, and it is bothering me.

I have created a model of primary visual cortex (V1). It all started out as just a fun project for myself, but eventually developed into something I really care about. In fact, I care so much for it that I have turned it into my Masters project.

The plan is to graduate this year, and so I have to start testing my algorithm and writing up the thesis. I have a paper in the works that describes good ways to validate models, and using those techniques, I am trying to develop a good model of V1.

Now, according to some of those measures, my model is very good. However, there is one thing that my model lacks, and that is a sort of “smoothness” that other models and real data contain. My approach with this neural network model is that of simplicity. I want to show that by taking a simplistic approach and using biology as our motivation for creating the network, we can develop maps that look strikingly similar to what we see in reality.

I spent the entire day trying to think of good, computationally inexpensive ways to accomplish this, and so far I have come up fruitless.

Below is an indexed image representing a primate cortex from an Obermayer et al (2003) paper, and then my “best” map just after that. They do not look overly similar… yet. However, I am getting there. I am pleased that I am starting to get groups of neurons as I see in the real maps, and yes I have the “pinwheels” as well as the iso-orientation domains, but I do not have the same visual appeal that many of the real maps do. I’m getting there though. I’ll figure it out.

This is one of those problems that I just cannot let go until I solve it! I’ll get there, eventually…