preview

Model-Based Prediction Theory

Decent Essays
Open Document

2.1 How does model-based prediction work The way of the model-based prediction works can be described in two stages: 2.1.1 Preparations and model training According to the certain case, different preparations are needed. Like in the study of Mitchell et al. (2008), the meaning of a word as a vector of intermediate semantic features needed to be encoded from the occurrences of stimulus of this word within a large text corpus, and the fMRI responses of each voxel also needed to be recorded. As in the case of Kay et al. (2008), they also need to record the fMRI data and in their case, especially from visual areas V1, V2, and V3. After the preparations, the important step of estimating the model begins. They will use the tuning from …show more content…

So the next step for the MVP classification is to compare the predicted experimental condition with the real experimental design, while the model-based prediction needs to compared the predicted brain activity pattern with the real observed pattern to determine the quality of the prediction. 2.2.2 Job of the model (algorithm) The two kinds of method differed a lot in the jobs that the algorithm really does. In the MVP classification analysis, the algorithm in the training period learns a function which can map between the voxel activity pattern and the experimental conditions (classification), while in the model-based prediction, the algorithm is actually a multiple regression algorithm of the features and the observed fMRI responses (Mitchell et al., 2008). 2.2.3 Level of modeling (input) In the MVP classification analysis, when the classifier is trained, the input data is a subset of the data that has labeled pattern after the pattern assembly, which involves series of voxels. However, in the model-based prediction, each voxel is modeled individually with a multiple regression algorithm, which actually did not involve other voxels in this …show more content…

So this method is often used in a group of voxels that most related with the classification (e.g. VT cortex in the visual object classification), and if needed to do it in the whole brain, other methods such as the searchlight method will be used to reach the whole brain level. But it is a different case of the model-based prediction method. Because this method actually modeled the individual voxel, it is possible for them do model through all the voxels in the whole brain (do not have a hypothesis of whether the encoding will localize in particular cortical regions) (Mitchell et al., 2008), or they can also select the cortex that makes the most senses (Kay et al., 2008). As a summary, MVP classification analysis and the model-based prediction differ in the level of modeling (multi-voxel vs. individual voxel), the job of the algorithm (pattern classification vs. multiple regression), the output and the indicator (experimental condition vs. brain activity), and the range of voxels analyzed (a group of voxels in general vs. whole

Get Access
Related
Get Access