Do you sometimes find yourself getting worse at writing? I don’t know if it’s down to burnout, or whatever else – but I was just reading some of the things I wrote a year ago and I must say, I wish I was that clear about things now. Isn’t research supposed to clear things up? It seems to me that right now I feel like I am just starting, whereas when I started I had a pretty clear picture.
Specifically, I am currently trying to write my literature reviews for my Transfer from MPhil to PhD and I’m finding it so difficult to summarize the papers I’m reading that I’m resorting to “how to write a literature review” guides. Then I looked back at the lit reviews I had done for my proposal (yes, that was before my research) and they are great! They summarize all the key points, compare and contrast them to my project and cross reference the most important works! It’s ridiculous!

Natal and VR

Microsoft presented project NATAL at E3 this year. This has rekindled my interest in the matching VR display problem (we want that holodeck!). Anyway, my main interest is in multiview displays or panoramagrams.

The idea is that a display presents multiple images at different angles, this generating a different image for each eye from whichever angle you look at it – generating stereoscopic image.

I think the future solution to this will be something like nano-pixels, where a group of light sources (lED or something smaller) form a half sphere. The seperate rays are deliniated by tiny tiny tubes (maybe not quite nano) that prevent the eye from seeing the others from a given angle.

This, together with a motion detection tech similar to NATAL and a VR setup like the CAVE would pretty much check all the boxes for a holodeck, although we still cant hold project objects onto our hand that way (we’d need a display on our hands with the current solution, which seems infeasable). So it’s not perfect, but definately a step forward as it works without periferals and with multiple participants: both the motion detection and the multiview screen techs are user-number independent.

Number conversion

The idea of number conversion from say decimal to hex intrigued me lately. If dealing with the same basic operands in two systems reaps such different results, then maybe numbers like pi are simply based on a different system, thus not allowing us to determine it using our number system.

Mirror Neuron initial thinking

The idea of “mirror neurons” seems false to me when opposed to the idea that standard neuron network can
produce the same mirroring effect:

If, say, taring a paper is a combine sensory experience of touch resistance, sound and visual stimuli,
the brain will have a network that stands for this concept.

If this network is highly trained/associated, then only one part has to become active (ie hearing a paper tare,
or seeing another person tare a piece of paper) to activate the associated neurons in the network.

Recent study (Dinstein I, 2008, 2008) propose that mirror neurons don’t exist and that firing in those regions
due to other processes.

The human infant brain uses “mirroring” to an effect before 12 months of Age. This indicates to me that
the connections that form “concepts” in our brain are still developing at that age and can change and modify
quickly during that period.

More and more I seem to believe in an almost 2 stage learning process in the brain. The first is the
growth of primary “loose” connections between neighbouring neurons. The second is the growth of longer

# ^ Dinstein I, Gardner JL, Jazayeri M, Heeger DJ. J Neurosci. 2008 Oct 29;28(44):11231-9.
# ^ Dinstein I, Hasson U, Rubin N, Heeger DJ. J Neurophysiol. 2007 Sep;98(3):1415-27. Epub 2007 Jun 27.
# ^ Terje Falck-Ytter, Gustaf Gredebäck & Claes von Hofsten (2006), Infants predict other people’s action goals[4], Nature Neuroscience 9 (2006)

Structured Neural Networks?

Back propagation networks often initialize their weights randomly. What effect do different random settings have on the ultimate behavior? Further, would it be more efficient to have structured weights, based on the behavior from the start? Is this similar to the set up of the brain, where connections are initially based on the growth of nerves? Is the growth of these nerves (during forming of the fetus led by the inputs or predetermined?

Damn you ADD!

1. Code a predictor in Flash

2. Realize that making a predictor might be redundant and I should really create the complete architecture.

3. Write List of Components in BB’s Book

4. Write Abstract for list

5. Write about BB being ahead of his time with Neural Networks

6. Find a reference for the first time dependent (meaning it takes time to activate a neuron) Neural Network design.

7. Finds a list of NN implementations and starts reading

8. Realizes that I am now completely lost and need to speek to someone. Being alone, decides to blog it.

9. Decides to Rename his blog to “A PhD with ADD” – which sounds like a famous book title and hope he will be famous for it.

10.  Realizes now WHY he felt inclined to read a bout NN’s, because He just found that Backpropagation NN’s just about try to do the same thing as his predictor algorithm.

11. Adds the Neural Network summaries from


to his learning Schedule


I think Hierachical Temporal Memory could be used to teach a computer creativity in drawing. A learnt image is represented in the weighted network structure of the HTM. If we can develop a mechanism for creating (drawing) an image that follows the trends in the network we should get approximised and idealized versions of what the mind knows