Fuzzy classifiers and subjective impressions

[dubina]

I believe this question is grounded in math. It will seem more a psychometric question at first.

A person has a particular experience. It creates a subjective impression, but it could also be described in terms of attributes. Attributes might include "danger", "speed", "beauty", "comfort", "happiness", "admiration" and so on...semantic terms applicable to the experience. Attributes might have semantic or numeric values: eg. "speed" might be "40 miles per hour". Thus, the real experience could be characterized in terms of its abstract attributes and attribute values. (The fitness of the selected attributes and their values to describe the real exerience is not an issue here.)

Later, the person would have another experience. It would similar in some ways to the first experience, but different: some different attributes, some of the same attributes, some different attribute values for shared attributes, and so on. Even so, the second experience would immediatly remind the person of the first experience.

Now suppose the two relatively comparable experiences are described to an observer in literal terms rather than by direct personal experience. Furthrmore, the observer does not receive the two descriptions together in time, and his choice for association would be confused by other literal experiences, each of which would have more cognitive distance in their attributes and attribute values than the first two experiences described above. In theory, the second experience should still remind the observer of the first experience, but the observer's choice would now be more difficult due to multiple instances and random order of acquisition.

This exercise to be reminded of one thing by another would be similar to a WAIS Simlarities subtest question: eg. "How is an apple like a pear?" but more elaborate, testing for deep insight and pattern recognition abilites as much as for similarities and differences. I am trying to associate the problem of how one thing reminds a person of something else to the most fitting field / topic in math or some adjacent field. I think the foregoing example of pattern comparison and recognition might have something to do with categorical classifiers and algorithms that find regularities in data. I am not a mathematician, so I am seeking judgment and advice.

Thanks,

******

I found the following to further elucidate my question: If an observer's association of one thing to another is an example of insight, it can be explained to some extent in terms of layered processing units.

Creative Insight: The Redistribution Theory

For the purpose of explaining insight, the key point is that the subjective experience of seeing an object or a situation is the result of a number of rapid and unconscious but nevertheless real choices, constrained and biased by the prior experience encoded in the relative strengths and activation levels of the vertical and horizontal links. The existence of an outgoing link from processing unit U in layer N to unit W in layer N+1, the activation of that link, the activation of excitatory and inhibitory links within a layer and the activations of the feedback links from higher to lower levels are determined by prior experience as well as by current perceptual input. The biases residing in the relative strengths and activation levels jointly produce the visual system’s best guess as to the nature of the perceived situation. The final percept – the working memory content – is a projection of prior experience onto the situation at hand.

Stellan Ohlsson (Deep Learning: How the Mind Overrides Experience, 2011-01-31)

 

[dubina]

Here are two more instances that fit the example more or less

“Fuzzy classification is a method which aims at placing all the cases or specimens in one or other of the classes even if the "fit" is not perfect. The method allows the simultaneous use of several criteria for assorting, that is, several attributes of the objects or cases are taken into consideration. Every member of the class complies with most of them, but not necessarily with all of them. What is common to all the members of a class is no specific attribute but family resemblance, which means that several but not necessarily all of the attributes match. Characteristic of fuzzy classification is that there will be no surplus class for those specimens or cases which would not fit.”

Pentti Routio, Arteology, the Science of Products and Professions

******

Memories are much more than single isolated concepts. A memory of Jennifer Aniston involves a series of events in which she—or her character in Friends for that matter—takes part. The full recollection of a single memory episode requires links between different but associated concepts: Jennifer Aniston linked to the concept of your sitting on a sofa while spooning ice cream and watching Friends.

If two concepts are related, some of the neurons encoding one concept may also fire to the other one. This hypothesis gives a physiological explanation for how neurons in the brain encode associations. The tendency for cells to fire to related concepts may indeed be the basis for the creation of episodic memories (such as the particular sequence of events during the café encounter) or the flow of consciousness, moving spontaneously from one concept to the other.

Rodrigo Quian Quiroga, Itzhak Fried and Christof Koch, Brain Cells for Grandmother, February 2013

******

A distributed representation can be used to recognize many versions of the same object, and the same set of neurons can recognize many different objects by differentially weighting their outputs. Moreover, the network can generalize by correctly classifying new inputs from outside the training set. Much more powerful versions of these early neural network models, with over 12 layers of hidden units in a hierarchy like that in our visual cortex and using deep learning to adjust billions of synaptic weights are now able to recognize tens of thousands of objects in images. This is a breakthrough in artificial intelligence because performance continues to improve as the size of the network and number of training examples increases. Companies worldwide are racing to build special purpose hardware that would scale up these architectures. There is still a long way to go before the current systems approach the capacity of the human brain, which has a billion synapses in every cubic millimeter of cortex.

In 10 years a thousand times more neurons will be recorded and manipulated than is now possible and new techniques are being developed to analyze them, which could lead to a deeper understanding of how activity in populations of neurons gives rise to thoughts, emotions, plans and decisions.

Terrence J. Sejnowski, WHAT SCIENTIFIC IDEA IS READY FOR RETIREMENT? Grandmother Cells