Limitations

7.2 Limitations 171

Maebashi Institute of Technology, Doctor Dissertation of Engineering

Yang Yang: A Brain Informatics-Based Study on Human Cognition, Emotion, and Their Relationship

strategy when performing calculations was indeed discussed. In future study, it is possible for me to explore the role of memory in mathematical study of children, given that individual growth and functional maturation might result in a shift from the memory-based frontal activation to the quantity-specific parietal activation in arithmetic processes.

According to different research purposes, a number of paradigms are available for studying mental calculation. For instance, different tasks can be used for examining exact arithmetic or approximate arithmetic. Methods of feedback collection are also multifarious. Responses can be made by choosing the correct answer out of a few puzzling options, or judging the validity of a proposed solution as adopted in our experiments, or in other forms. However, merits and demerits coexist for each feedback method. In the case of multiple puzzling options, subjects need to control more fingers to manipulate the controller for answering questions, which is likely to elicit more cognitive compounds, and risk pressing wrong buttons. The manipulation will turn to easier if subjects are asked to make a right-or-wrong response, however, the rate of correct answer will increase even if the answers are reliant on guess. We have to evaluate each design. Basically, the more complex an experimental design appears, the more conditions need to be controlled for the experiment. As a (preliminary) main experiment for studying mental arithmetic, we chose the right-or-wrong judgment as described in this thesis. In my opinion, comparison of results from experiments with identical contents but different feedback methods is likely to be an interesting research.

Top-down and bottom-up processes were mentioned in this thesis frequently.

Generally speaking, they are two approaches to understand the flow of information in processing within human brain and pervasively utilized in neuroscience and psychology.

Typically, sensory input is considered "down", and higher cognitive processes, which

7.2 Limitations 172

have more information from other sources, are considered "up". In the bottom-up direction, process starts at the sensory input, i.e., the stimulus. Thus, this approach can be described as data-driven. In the top-down direction, process is characterized by a high level processing by more cognition, such as goals or targets. In a micro-scaled perspective, it is known that information about the stimulus is encoded in the pattern of action potentials and transmitted into and around the brain. Neurons are capable of propagating the signals in a remarkable speed over large distances. Thus, switches between top-down and bottom-up transmissions are probably proceeding transitorily.

One-way transmission of information during a relatively coarse temporal period is almost impossible. Nevertheless, meso-scaled studies regard human brain as a holistic system, and aim at constructing a cognitive architecture composed by several spatially distributed subsystems (brain regions). The key to achieving such a goal is to conceptualize the trend how information flows are processed. Serving as a main approach dedicated in meso-scaled studies, fMRI is insufficient to disclose the signal transmission at the level of neuron, but is able to offer millimeter-based clues of neural mechanisms to establish cognitive models. In our future work, the observation of brain will be extended to micro scale, and the phrasing must be changed at that time.

In the present thesis, all the experimental results were obtained based on group analyses by averaging results from multiple subjects doing the same task. Apparently, results concluded from one individual subject will be biased by the individual difference extremely. The objective of our researches is to figure out some universal principles to disengage the puzzle of human brain. As a result, we must recruit a group of subjects for representing a population in each experiment to meet the statistical criteria. However, one problem is how to unify the original state of all the subjects before the experiment

7.2 Limitations 173

Maebashi Institute of Technology, Doctor Dissertation of Engineering

Yang Yang: A Brain Informatics-Based Study on Human Cognition, Emotion, and Their Relationship

information and education background (e.g., almost same age, from same college, etc.) to reduce individual differences; (2) performing identical experimental procedures and using same instructions on each subject; (3) training subjects on how to participate in tasks until all the subjects achieve almost same level of proficiency; (4) starting the on-task session of fMRI scanning after the first resting session to ensure that all the subjects have adapted to the environment inside the scanner. Although it is impossible to eliminate differences in the original states between subjects completely, we have done our best to control the differences.

Another problem is how to apply the results generated from group-based analyses to explain or predict an individual subject. Due to the differences in the shape of brains, involved number and even type of neurons confined in a certain brain region that can be specified by coordinates and size may be distinct across different subjects. fMRI has long been expected to play a part in clinic, such as supporting the diagnosis of mental diseases for a single patient. However, this goal is still unreachable at present. Statistics is a core technique supporting analysis of fMRI data. The weakness of fMRI aforementioned is related to the statistics characterized by collecting and analyzing numerical data in large quantities rather than describing each single data in detail. But even if the group-based results cannot be used for personalized service for now, they are still helpful to uncover general mechanisms of brain activities. Furthermore, as the sample size in an experiment increases, the accuracy of the results for predicting each single person will grow as well, as long as the sampling is reasonable. Computer science will accelerate the application of fMRI techniques. Dosenbach et al. (2010) trained a computer to recognize patterns in resting-state data. They collected data from nearly 240 people aged 7–30 years to build up maps of brain connectivity at different ages. By using the maps, they could take a single brain scan from a different person and,