Classification, RSA, Reconstruction Flashcards
What does semantic similarity has to do with MVPA?
Semantic similarity between close by areas of the brain allows for research about which area represent what. e.g FFA for faces
What are Voxel activation vectors?
Voxel Activation vectors are vectors that are computed on the levels of activation of multiple voxels in a specific area during presentation of specific condition, or alternatively on their GLM.
Which methods are used to address what areas are interesting for a MVPA
Region Of Interest, Searchlight Search, Whole Brain approaches
What are the temporal variants that MVPA can have?
– trial-wise BOLD signal
– run-wise GLM parameters
What is cross-validation? What are the folds?
Data is split into training and testing data. Training data is taken minus 2 occurrences from opposite conditions and then a classification algorithm is trained on the data. The algorithm has to classify properly the two training occurrences, this is one fold. This process is then repeated for (all) the occurencies of the training data and tested against the test data.
how does LDA work? What is a a Linear Discrimination Algorithm?
Linear Discriminant Analysis (LDA) is a Linear Discrimination Algorithm used in MVPA to find a linear decision boundary that optimally separates different classes in the feature space. Assumes multivariate normal distributions for the two
classes and estimates their class means (centroids; dots in slides) and common covariance (ellipses) to compute the posterior probabilities.It’s a generative model, that is, a model of the probability distribution of the data. Limited by the fact that there is a need to model within class covariance of data, which is not reliable for voxel spaces with high dimensionality.
How does Logisitic Regression Works? What is it?
Logistic Regression is a popular algorithm used for classification tasks. It is a statistical model that predicts the probability of an example belonging to a specific class based on its features.It is a discriminative model that does not make an explicit statement about the distribution of the data (doesn’t assume normal distribution), and not required to find within class covariance of data. IT directly models the posterior distribution using only a weight vector and bias (arrow with intercept in slides)
How does a Support Vector Machine works?
Nonprobabilistic approach to classification, based on the principle of margin maximization. Searches for linear discrimination boundary between two classes and then tries to find a hyperplane with the maximal possible distance from the nearby training data points. Trained classifier is defined only by those data points that lie at the edge of the margin, so called support vectors. Achieves
classification by searching for a separating area without data points (between colored lines in slides), which is as wide as possible (margin width; double-sided
arrow).
What is the Mahalanobis distance?
It is a measure used to discriminate data in MVPA. Mahalanobis distance directly quantifies the
distinctness of the two distributions by the distance of the centroids measured in units of the within-class standard deviation (ellipses in graph in slides)
What does it mean in MVPA if classification is possible?
information → explained multivariate variance explain this statement
What is the linear read out hypothesis? Is it true?
What is the idea behind Representational Similarity Analysis?
The idea behind is that a more accurate view of representations inside the brain can be obtained by comparing measurements of the same stimuli across several techniques through a dissimilarity matrix. This would allow to compare inter subject and inter species
How is the concept of second-order isomorphism is applied in RSA?
The idea is to compare between stimuli and representations thereof. Relating similarity between stimuli with similarity with their representations.
Was visual image reconstruction in the material?
