Representational Similarity Learning with Application to Brain Networks

Authors: Urvashi Oswal, Christopher Cox, Matthew Lambon-Ralph, Timothy Rogers, Robert Nowak

ICML 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We show, in theory and f MRI experiments, how Gr OWL deals with strongly correlated covariates. ... Before applying our framework to real f MRI data, we consider a simulation study that allows us to compare results against a known ground-truth. ... We next consider the application of group lasso and Gr OWL to the discovery of similarity structure in neural responses measured by f MRI across the whole brain while participants perform a cognitive task.
Researcher Affiliation Academia Urvashi Oswal UOSWAL@WISC.EDU Christopher Cox CRCOX@WISC.EDU Matthew A. Lambon Ralph MATT.LAMBON-RALPH@MANCHESTER.AC.UK Timothy Rogers TTROGERS@WISC.EDU Robert Nowak RDNOWAK@WISC.EDU University of Wisconsin-Madison, Madison, WI 53706, USA University of Manchester, Manchester M13 9PL, UK
Pseudocode No The paper describes mathematical formulations and algorithms conceptually but does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any statement or link indicating that the source code for the described methodology is publicly available.
Open Datasets No The fMRI dataset used for the real data application was collected as part of a larger study from 23 participants at the University of Manchester. No public access link, DOI, or specific repository name is provided for this dataset.
Dataset Splits Yes For each participant, training data were divided into 9 subsets containing 4-5 stimulus events each. One subset was selected as a final hold-out set. Models were then fit at each of 10 increasing values of each λ and λ1 parameter (grid points) using 8-fold cross validation. At each fold we assessed the model using the Frobenius norm of the difference between the target Y entries and the predicted b Y = X b B entries for hold-out items (henceforth the model error).
Hardware Specification No The paper mentions that fMRI scans were collected, but it does not specify any computing hardware (e.g., CPU, GPU models, memory, cluster type) used for data analysis or model training.
Software Dependencies No The paper mentions 'standard pre-processing' and 'deconvolution procedure with a standard HRF kernel' but does not specify any software names with version numbers for dependencies.
Experiment Setup Yes We fit models by searching a grid of parameters (λ, λ1), including λ1 = 0 as the special case of Gr OWL that is group lasso. ... Each column corresponding to a voxel was normalized to be of standard deviation equal to one and a column of ones was added for bias correction. ... Models were then fit at each of 10 increasing values of each λ and λ1 parameter (grid points) using 8-fold cross validation.