1. 9

2. no

3. 7

4. 4

This proposal is somewhat clear, it is fairly concise, and it is well-organized. The sections are laid out in a useful manner and the sentences aren't overly wordy. For the most
part, this proposal addresses everything it needs to as specified by the proposal guidelines. There are just a couple things it misses. There isn't much direct citation in the
related work. It talks a lot about general applications where neural networks have been used in the past, but there aren't any specific examples (with citations) talked about. The
previous experience doesn't really talk about the previous experience of the researchers, it just talks about what each researcher is going to do. It would be better if this
information was somewhere else and the actual experience was here. Also, the demonstrated need is not very compelling.

This project fits somewhat within developmental robotics, though it is a better fit for machine learning (admittedly they are similar fields). This project doesn't really use
robots in any way, and it doesn't use data grounded in sensory input. The best part about this project is that it is inspired by Jeff Hawkin's neo-cortical model. This model is
very interesting and holds a lot of potential. The part I like least is that the actual project isn't very related to Hawkin's model. It appears to be just a standard neural
network, nothing about learning spatio-temporal patterns.

My main concern with this project is that it is too simple and doesn't really do anything that has not been done (more than a few times) before. This project should easily be
finished on time. In fact, approach 1 could probably be completed in about 20 minutes, give or take a few depending on the researchers' experience with Weka. Overall, this project
is too easy with very little improvements over current systems. Also, the use of human created feature vectors (square, red, etc. for an object) rather than actual input data from
a robot may yield impractical results.

This project leaves out a couple critical details. Approach 1 doesn't detail any specific algorithms besides SOM that will be used. Neither approach really gives a good formal
definition of the algorithms or data structures to be used. Having this would have made the approaches much clearer.

This project can be improved in a couple ways: First, using actual data gathered from a robot or robot simulation would produce much more useful results. There are plenty of data
sets from the developmental robotics lab that can be used, just ask Jivko about them. Second, approach 1 can be redesigned so that it has more complexity, such as performing
feature extraction on the data first, combing the classifiers into an ensemble, or trying novel ways of organizing the networks. Overall, this project is not very complex, but
applying a couple small changes could improve it significantly.