User:Nbc91/sandbox

=Information Integration Category Learning=

It is important as humans to have the capabilities to categorize as it is such a vital component of our lives. Having the capacity to categorize and the knowledge of the use of the available memory systems for categorization, will allow us to further understand this skill. Categorization is the ability to attend to objects or events differently by their qualities or classes, and to separate them accordingly. Categorization is not only part of our daily lives but also it is becoming increasingly emphasized in cognitive science as an important concept for humans. Research on category learning has become increasingly widespread, dealing with how individuals go about categorizing stimuli of unfamiliar classifications. Learning to associate a new stimulus or object to a category requires the use of creating novel representations within an existing category. The way humans categorize is by use of a decision strategy and humans have the capabilities to categorize in multiple different ways.

Information-integration categorization comes from general recognition theory (GRT), which investigates stimulus identification through use of perceptual and decisional processing. When dealing specifically with categorization behavior, it is often called decision bound theory. Information-integration category learning task is the process of integrating information from at least two stimulus dimensions before making a decision. It is important to reiterate that integration is made before deriving a resolution. In most information-integration tasks, the optimal decision bound may be at a diagonal, which would indicate that the participant would be required to attend to two separate dimensions, for maximal precision. The feedback given in information-integration category learning is essential in order to determine whether or not a different strategy may be required. Information-integration is mediated by an implicit procedural system which relies on the striatum. As individuals get older, the volume of the brain tends to decrease, which could be a reason for why older participants perform worse on information-integration category learning tasks. In general, it is more difficult and time consuming for humans to categorize this way.

Similarities and Differences
Information-integration categorization is difficult for an individual to verbalize due to the fact that it occurs outside of conscious awareness. On the other hand, rule-based categorization deals with participants attending to either spatial frequency or orientation and does not require the participant to attend to both, thus making it easier for the participant to verbalize and be aware of. Again, with respect to rule-based tasks, the answer is obvious and easily verbally identifiable. For the decision to be verbal, the participant must first split up each feature and make decisions for them separately. Once this is complete, then the participant is required to combine the two decisions together using operations such as “and” and “or”. Note that the decisions from each dimension are made prior to the information being combined, unlike information-integration categorization where information from each dimension is combined before any decisions are made. Information-integration learning is oriented towards gaining response feedback, which is relatively time consuming, while rule-based learning is oriented towards labeling categories. Paul, Boomer, Smith and Ashby discuss how rule-based tasks and information-integration tasks are solved using two different systems. They argue that rule-based tasks are strongly dependent on declarative memory systems such as working memory, while information-integration tasks are dependent on procedural memory. In information-integration categorization, because of the fact that feedback is automatic, this type of categorization does not require any attention or effort. The competition between verbal and implicit systems, also known as COVIS, assumes that the explicit system is emphasized by rule-based categorization learning while the implicit system is dominated by information-integration learning tasks. The explicit system learns through hypothesis generation and testing – meaning that once the response is given and if it is incorrect then another rule must be applied until a correct response is delivered. The implicit system is dependent on a reward signal, which will indicate the appropriate category automatically.

Theories of Information-Integration Category Learning
Theories of information-integration category learning can be separated into two classifications – parametric and non-parametric. Parametric classification deals with a few assumptions – that the categories are specifically structured or that the boundaries of categories have a specific form (ex. linear). Non-parametric classification on the other hand, does not assume anything to be true regarding the structure or boundaries of categories.

An example of a parametric classification deals with prototype models in which the assumption is that there is a linear decision boundary, while exemplar models are an example of a non-parametric classification since the assumption is that each prototype is stored in memory with its category label and all its categorical information. In prototype models, people classify objects into categories by comparing their similarities to other objects in a category. This type of classification is not as structured as exemplar models. Exemplar models are more structured such that it is a more concrete way of categorizing.

COVIS
Competition between a Verbal and an Implicit System, otherwise known as COVIS represents the class of models of multiple systems – there is an explicit system that generates hypotheses and an implicit system that is useful in procedural learning. It is assumed that COVIS is motivated by the functions of neurobiological structures of the brain where learning in the explicit system is central in the anterior cingulate, prefrontal cortex, and the top of the caudate nucleus while learning in the implicit system is based at the back of the caudate nucleus. In other words, the back of the caudate nucleus would be activated during information-integration category learning tasks. It is suggested that implicit learning is dependent on the connections between  visual cortical areas and the caudate. Such connections are likely to help with creating new representations in the cortex by use of feedback signals. Again, this brings us back to the importance of receiving immediate feedback in information-integration category learning.

New Addition
A new addition to the COVIS model is the Medial Temporal Lobe (MTL), which works with the prefrontal cortex and the caudate for verbalizing rules of categorization. With the use of functional magnetic resonance imaging (fMRI), this new multiple-systems model of learning can be tested with rule-based and information-integration categorization tasks. A study was done to investigate the use of MTL where participants were required to integrate information from both position (orientation) and rate of occurrence (frequency) of sine waves to create categories. After each condition, brain activity was recorded. Correct categorizations were compared to incorrect categorizations and each was identified with specific brain regions. fMRI results support the hypothesis. The brain region activated in information-integration categorization is the caudate area.

Current Research
Maddox and Filoteo argued that since COVIS defines information-integration category learning by use of an implicit system, then it must therefore make sense that adding extra decision bounds should not affect learning difficulty. They stressed that it was important to control for extraneous variables and by doing so, they simply increased the distance between perceptual category clusters to achieve the greatest accuracy. On the other hand however, Roger Stanton and Robert Nosofsky supported the idea that adding extra decision bounds would negatively affect learning difficulty, and therefore did not account for this increase in parameters.
 * 1) They used four perceptual clusters of a bivariate normal distribution, each cluster being a category in itself. Participants were given either two or four categories in their task – both tasks displaying a stimulus for which the participant was required to respond to. The correct category label was then presented and whether the participant was right or wrong would be indicated.
 * To statistically analyze the data, conducting an analysis of variance (ANOVA) to determine the differences of means is necessary. After conducting a three way ANOVA, the results showed that participants who were given four categories performed worse than those who were given only two categories, impacting the learning of information-integration categorization and thus going against Maddox and Filoteo’s predictions.
 * It is therefore evident that when you do not extend the distance between perceptual clusters, performance decreases in four categories versus two.

The Role of Procedural Memory in Information-Integration Category Learning
Procedural memories are learned through experience – the idea being that they are helpful in tasks, which are not easily learned logically. Research supports the idea that rule-based and information-integration tasks are used in different streams – an explicit, working memory system and an implicit, procedural system. There is evidence that information-integration tasks are guided by procedural memory. To test this theory, the most common way is to use of a serial reaction time (SRT) task. In a serial reaction time task, participants are required to respond to various stimuli on a screen by pressing specific keys as fast as possible. If you were to change the location of the keys however, this would interfere with learning. If you were to switch hands that press the keys, this should not interfere with learning. Results have proven this theory to be correct.

Procedural learning tasks again appear to be linked to information-integration learning tasks. Both types of learning are effective most when feedback of response is immediate. When the feedback is delayed however, this interferes with information-integration learning. An experiment was done to demonstrate this idea. Participants were given two categories and were instructed to learn about them accurately. There were four conditions: a 10 second delay condition, a 5 second delay condition, a 2.5 second delay condition and an immediate condition. Results indicated that indeed the effect of delay was significant. The results demonstrated that participants performed better in the immediate feedback condition compared to the delayed conditions - specifically in the information-integration task.

Neuropsychological Patient Groups
There has been much research and focus on brain-damaged patients, especially regarding information-integration category learning. Specifically, research has focused on patients with Parkinson’s disease and Huntington’s disease. Procedural memory is dependent on the basal ganglia. So, if this area of the brain is damaged, then it would naturally make sense that learning in information-integration tasks would be affected. There is also evidence however to contradict such findings. Other experiments have shown that patients with Parkinson’s disease are more impaired in rule-based tasks than information-integration tasks. Parkinson’s disease kills dopamine-producing cells in different areas of the brain. The dopamine-producing cells are damaging to the caudate nucleus, which is a region responsible for implicit learning. It therefore makes sense to assume that individuals with Parkinson’s disease would have difficulty with information-integration category tasks. What appears to be the case however is that individuals with Parkinson’s disease are more impaired on simpler rule-based categorization learning tasks. Evidently, due to conflicting findings it is clear that more research with regards to Parkinson’s disease is needed.

Conclusion
Information-integration category learning is a rather complex phenomenon. Compared to rule-based learning, in which the rule is easily verbalized, information-integration calls for the combining of facts from two separate dimensions in order to provide a rule for categorizing. Having information come from two separate dimensions makes the categorization rule difficult for an individual to explain verbally. Information-integration category learning is focused on getting feedback in order to understand what went wrong and how to fix it. While it is a time consuming process, it does aim to reach maximum precision, which of course is a good thing.

Future Directions
Perhaps in the future, researchers should dive deeper into the effects of information-integration categorization in humans with Parkinson’s disease since so far, research is showing conflicting findings. As well, there is room for research into the field of non-human species, such as animals.

All in all, in order to attain an optimal strategy in information-integration tasks, a form of procedural learning is necessary, as well as an immediate feedback response.