A Statistical Recurrent Model on the Manifold of Symmetric Positive Definite Matrices

Authors: Rudrasis Chakraborty, Chun-Hao Yang, Xingjian Zhen, Monami Banerjee, Derek Archer, David Vaillancourt, Vikas Singh, Baba Vemuri

NeurIPS 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We perform numerical experiments demonstrating competitive performance with state of the art methods but with significantly fewer parameters. We also show applications to a statistical analysis task in brain imaging, a regime where deep neural network models have only been utilized in limited ways.
Researcher Affiliation Academia University of Florida, Gainesville, USA University of Wisconsin Madison, USA
Pseudocode No The paper describes mathematical equations for its model but does not include a clearly labeled pseudocode or algorithm block.
Open Source Code Yes Our code is available from https://goo.gl/Sf Aez S.
Open Datasets Yes We used the Moving MNIST data as generated in [57]. For this experiment we performed 2 and 3 classes classification experiment. In each class, we generated 1000 sequences each of length 20. ... We performed an action classification experiment on UCF11 dataset [43]. ... The data pool consists of d MRI (human) brain scans acquired from 50 PD patients and 44 CON healthy controls.
Dataset Splits Yes In each class, we generated 1000 sequences each of length 20... We report the mean and standard deviation of the 10fold testing accuracy. ... All the models achieve > 90% training accuracy. ... Then we perform permutation testing for each tract as follows (i) randomly permute the class labels of the subjects and learn m PD and m CON models for each of the new group.
Hardware Specification Yes All images were collected using a 3.0T MR scanner (Philips Achieva) and 32-channel quadrature volume head coil.
Software Dependencies No The paper mentions using 'FSL [8] software' but does not specify a version number for it or any other ancillary software.
Experiment Setup Yes For the convolution layer, we chose the kernel size to be 5 5 and the input and output channels to be 5 and 10 respectively... The number of output units for LSTM is set to 10 and the number of statistics for SRU is set to 80. ... SPD-SRU takes 75 epochs to converge to the reported results... The parameters of the diffusion imaging acquisition sequence were: gradient directions = 64, b-values = 0/1000 s/mm2, repetition time =7748 ms, echo time = 86 ms, flip angle = 90 , field of view = 224 224 mm, matrix size = 112 112, number of contiguous axial slices = 60 and SENSE factor P = 2.