Posted: January 9, 2012 at 3:47 pm

Above is a reconstruction of approximately 20 frames of video test footage. (more…)

Posted: December 9, 2011 at 5:21 pm

I wanted to post my progress on segmentation before I start working on clustering and video sequences. Here are images that show the current state of segmentation (down to under 2s per frame) of a more realistic Vancouver scene captured on video: (reconstruction, mean-shift filtering, original image)

(more…)

Posted: November 26, 2011 at 3:40 pm

In playing with code for deep learning (DBNs in particular) I’ve had to install CUDA to do matrix processing on the GPU. So I went ahead and recompiled openCV to use CUDA. Following is a similar segmentation image to the one in previous posts. This one was computed in 2.7s on the GPU, and on a different machine. The non GPU version did it in 6s, and the old machine did it in 9s. This machine is already a few years old, and has an older 8600GTS that barely supports CUDA. A faster machine, or perhaps just a faster GPU, may get these numbers down to something more real-time.

Posted: November 11, 2011 at 12:32 pm

I’m continuing to listen to some of Franklin’s lectures about LIDA and cognitive modelling in general. Yesterday I got through the one explaining Global Workspace Theory (GWT). Little did I know, but I had already come across this theory during my Masters research, and discounted it due to its causal disconnection between consciousness and cognitive processes. While I continue to find this problematic as a model of sentient creatures, the notion of “functional consciousness” is certainly interesting and useful in the case of machines and AI.

(more…)

Posted: November 5, 2011 at 11:19 am

(more…)

Posted: November 4, 2011 at 5:02 pm

In attempting to think through a possible integration of LIDA and the current conception of the dreaming system, I’ve stumbled upon the above question. In the previous sketch, the content of memory (the current state of the “conceptual system”) is used to generate visual sensory data that is the dream content. In LIDA, it becomes somewhat unclear from which module(s) the content of the dream would arise. Clearly, it would involve longer term memory structures, such as those in episodic memory, which are only explicitly manifest in the workspace and then “broadcast” into the global workspace. So where does the dream occur? In the workspace or in the global workspace? Since the agent is only conscious of what is in the global workspace, then that appears to be the logical location. (more…)

Posted: November 1, 2011 at 4:39 pm

In this post I will describe the conceptual subsystem, as I have imagined it up to this point, before looking more closely into the LIDA architecture. The central feature of this conception is that each high level feature, for example location in space, or colour (or redness), is considered independent. This contrasts with many systems that concatenate all these features in order to create a cluster representation that encapsulates overall similarity. This is conceptualized as a highly shallow system where features are meant to be high level. In fact no clustering is needed at all, as only a distance function can be used to determine the distance between precepts in terms of each individual feature. (more…)

Posted: October 31, 2011 at 4:55 pm

This is an excerpt from my email I wrote regarding next steps:

1. Make a diagram of the conception of the system as it was before, including how I was thinking of links between high level neurons having different weights for each high level feature (which may be made up of a vector of lower level features, eg. histogram). This would be the context in which to situate discussions of machine learning systems.

2. Read more about LIDA to understand better how percepts and knowledge from memory stores are structured, and if a subset of the architecture can be used.

3. Wish list for machine learning / clustering (in no particular order)

• Unsupervised

• Ideally statistical (stores a distribution of patterns, not all the patterns themselves, but this is flexible)

• Output could be manifest in connection weights between neuron-like structures that represent higher level concepts. (concept as distribution of activation) Each item should belong to a concept to a degree, not Boolean.

• If it is deep learning and I can forgo the feature extraction, it must be elegant (simple enough for me to understand) and be feasible for high resolution colour input images.

• If it is deep learning it should be generative: can learn high level abstractions (concepts) from sensory data, and also construct sensory data from those high level abstractions.

Thank is all I can think of off hand. (more…)

Posted: October 28, 2011 at 12:42 pm

On one hand we have the deep machine learning systems discussed in the previous post, which are quite complex (from a mathematical perspective) and only roughly informed by neuro and cognitive sciences. On the other hand are cognitive architectures that are inspired by biological knowledge (BICAs). These are systems that attempt to balance high level (psychological knowledge) with low level (neurological knowledge) in a tractable system that models mind-brains. These systems can be purely symbolic, connectionist (sub-symbolic) or hybrids of the two. While they are biologically inspired they favour a high level abstraction of cognition over fine details of brain (neuron) function. (more…)

Posted: October 28, 2011 at 9:53 am

As part of stepping back to see the big picture I’ve turned a second look at deep machine learning systems and biologically inspired cognitive architectures (as suggested by my supervisor), the latter of which will be discussed in another post.

Deep machine learning systems attempt to resolve an issue with “shallow” learning which has increasingly become the following process: Input → Feature Extraction → Machine Learning. An argument against this approach is that the “intelligence” shifts from the machine learning system to the human-centred, and domain specific, art of feature extraction. Deep learning systems excise the middle man, allowing Input → Machine Learning, without the intermediary. (more…)

Posted: October 20, 2011 at 1:41 pm

After writing the previous post I had a meeting with my supervisor yesterday. He suggested that the answers to these questions of feature abstraction should be contextualized by the machine learning method used to organize these perceptual units. So I’m putting further development on the back burner to look at the big picture and do more reading. Following is a figure of the overall structure of the whole system, as I currently imagine it. (more…)

Posted: October 20, 2011 at 11:55 am

Since it’s time to explore some clustering methods for these perceptual patches, I first calculated a RGB histogram for each patch, making use of the mask. In attempting to make sense of the structure of the openCV hist (1 channel, but 3 dimensions) I realized that perhaps the hist is not appropriate for this project. The central question is: what is important about colour for the purpose of associating perceptual units? Perhaps the histogram should be abstracted into a set of concepts (for lack of a better term) representing major colour clusters. (more…)

Posted: October 5, 2011 at 3:01 pm

I’ve finally managed to get the segmentation data into a useful form. Following is a reconstruction of the original image from extracted patches, and a corresponding image that shows the segments, filled in random colours.

(more…)

Posted: September 1, 2011 at 6:05 pm

Now that I’ve passed my comps I’ve been getting back into working on image segmentation.

Here is the current state of segmentation (using the same test image as in previous posts), where every region, no matter now small, is coloured in a random colour:

(more…)

Posted: July 29, 2011 at 10:36 am

I successfully defended my comprehensive examination yesterday. The exam consists of three questions, two written and a third orally presented. Here are my answers:

1. Breadth Question: Propose your own typology of generative arts and choose relevant examples to illustrate it.
2. Depth Question: Present, discuss and contrast two current models of dreaming and two current models of mental imagery and/or imagination. For each of the four models, discuss their relevance to computational creativity (the field concerned with using computers as creative means).
3. Methodology Question (limit 30 minute presentation): What methods are available to evaluate generative art systems inspired by cognitive sciences? Present and compare at least three methodologies.

Posted: July 14, 2011 at 4:03 pm

My written questions have been submitted, and I’ll be defending on July 28th.

Posted: June 30, 2011 at 4:35 pm

This is a sketch of the system as I’m currently thinking about it. I’m not entirely happy with it, and some aspects are unclear. For example the use of object location in space and time is not known yet. Also there needs significantly more detail in the description of the Neuron-Like Network. (more…)

Posted: June 30, 2011 at 3:50 pm

One of the points of discussion with the philosophers was whether the dreaming machine will have analogues of all the characteristics of human sleep. Some of these don’t make sense to include, as in the alteration of self-consciousness (as the machine has no consciousness), but one in particular, the stages of sleep ranging from REM sleep to Slow-Wave-Sleep, could be relevant. (more…)

Posted: June 28, 2011 at 11:32 am

I tried doing the segmentation in CIE Luv colourspace, as its perceptually oriented, but it appears to actually do a worse job of extracting components than RGB space:

Posted: June 24, 2011 at 4:44 pm

I’ve been keeping my rough notes on DM3 for quite a while now, since doing the directed readings with Steven, through IAT 888, and now for the “actual” development. I’ve also been keeping notes through the early IAT888 development. I’ve gone through the documents and commented and/or striked out stuff that is no longer relevant, is too dependent on development, or is just out of scope with the time limitations of the project. I wanted to post an archived version before I remove the striked out text. I’m using the information contained in it to start sketching out a system design. I hope my next post will be a start of that document, and a diagram of the system as I’m currently thinking about it.