Day XIX:

Today we had another Space Weather group meeting. Dr. Dube explained his work on trying to see if a neural network can make a correlation between spikes in separate data streams happening at different times. He explained the x-correlation technique and how every point in one set is multiplied by every point in the other set and is then normalized. This produces a huge spike in the output data. Dr. Glenn showed his work on image processing and how each quadrant of an image is in a particular state is is compared to a baseline or "quiet state". When a CME occurs, each quadrant show a spike in the different between the quiet state and the actual state. The problem is to see if a neural network can take in the data from the image processing technique and find a pattern. This would ultimately turn everything on its head and result in the prediction of, as opposed to reaction to, Coronal Mass Ejections.

Day XVIII:

Today I started out doing a lot of the usual things. I work until about 2:30ish when I was notified that the Remote Sensing group was going to have their collect today on short notice. They greatly need help to set everything up and prepare for the fly over. I went outside and set up targets. We finished three minutes before the plane flew over.

Day XVII;

Today I was going to help the Remote Sensing Group with another data collection session today, but after we got most of the targets set up the plan was postponed. I wasn't too mad because it is fun being outside helping out, at least when it isn't raining.

When I got back to my work I started reading some of the papers posted on the RIT twiki site. I read one paper that caught me completely off guard. At this point in time I feel as though I have a pretty firm grasp on the Neural Network concept but I guess people have come up with even crazier ideas. This paper introduced this idea that was almost a composition of Neural Networks in that one overarching Network orders for the training and learning of minion networks. The outputs from these subnetworks would then be weighted to produce a final output. It is like a Neural Network electoral college with certain helper networks getting more of a say in the final output than others. I think that the overall point of making a Neural Network that is composed of smaller networks (as opposed to just making a bigger network) is that new information can be added with having to retrain the whole entire network. It is very cool but very confusing.

Day XVI:

Today I spent a lot of time looking at how to use the MatLab neural network tool. It seemed pretty straight forward until I tried to load the input and target data. It kept on giving me a Matrix Dimension error. I guess I need to learn more about MatLab syntax and, well, just more about MatLab in general. Ultimately I want to be able to combine both the abilities of a Neural Network and the tools available in MatLab (i.e. regression analysis, graphical representations, etc..) to better see the abilities of a Neural Network as different things change.

I was also going to help the Remote Sensing group with its second collect today, but apparently there were problems and it has been postponed until tomorrow.

This morning for a daily intern meeting, we all observed how our eyes adjust themselves to different light settings. First we stayed in a dark room for about ten minutes, letting out eyes get used to the dark setting. Then we covered one eye and preceded out into the light for a little walking, letting the uncovered eye adjust back to the normal light setting. We then returned to dark room and opened out closed eyes to find that we could see out of that eye but not the other. I have done this before so the result wasn't surprising but it didn't matter because it was fun nevertheless.

Day XV:

Today was very eventful. From 9 - 12 I helped the Remote Sensing group prepare for the flyover by setting up different colored cloths/fabrics. I got to go up on the roof of the building and pin down 2m by 2m "blankets". It was pretty windy but also very cool (double meaning). There were a lot of different experiments being done. Some of these experiment were dealing exclusively with Hyper Spectral Imaging and the techniques of detecting different materials, while others were using LiDAR or other means to detect variations in heat. The plane that flew over was mounted with all of these sensors. The plan made multiple flybys in an interlacing pattern (back and forth) so that the whole area we were working was captured twice.

For lunch I met up with Dr. David Lawlor. I have know Dr. Lawlor for years and, since he is a professor here at RIT, we had made arrangements to get together. We talked about a lot of things ranging from my work here to his recent vacation to the Caribbean. It was very a enjoyable and informative meeting.

Day XIV:

Today we had another Space Weather group meeting to touch base with everyone and see what different avenues of research should be explored further. We talked a lot about neural networks and what things should be done next in the testing and training process that would better suit our needs. After listening to Dr. Dube and his methods of dealing with a neural network, I realized I have a few flaws in my own techniques. For one, the network size I was trying to train was not large enough and two, my method of training was not sufficient. Before I was giving the network one input and expecting it to give its approximation of a sine wave. I found that it is better to instead provide the network with multiple ordered inputs of the sine wave and try to train the network to guess the next output. It is a much different from my way of thinking.

Day XIII:

I was back to doing many of the usual things today. I did more work analyzing the demo network and debugging my own. There are a lot of errors I have to fix and things I have to try. I really want to make this work.