Distributed Weight Consolidation: A Brain Segmentation Case Study
Authors: Patrick McClure, Charles Y. Zheng, Jakub Kaczmarzyk, John Rogers-Lee, Satra Ghosh, Dylan Nielson, Peter A. Bandettini, Francisco Pereira
NeurIPS 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluated DWC with a brain segmentation case study, where we consolidated dilated convolutional neural networks trained on independent structural magnetic resonance imaging (s MRI) datasets from different sites. We found that DWC led to increased performance on test sets from the different sites, while maintaining generalization performance for a very large and completely independent multi-site dataset, compared to an ensemble baseline.In Table 2 we show the average Dice scores across classes and s MRI volumes for the differently trained networks. |
| Researcher Affiliation | Collaboration | Patrick Mc Clure National Institute of Mental Health patrick.mcclure@nih.gov Charles Y. Zheng National Institute of Mental Health charles.zheng@nih.gov Jakub R. Kaczmarzyk Massachusetts Institute of Technology jakubk@mit.edu John A. Lee National Institute of Mental Health john.rodgers-lee@nih.gov Satrajit S. Ghosh Massachusetts Institute of Technology satra@mit.edu Dylan Nielson National Institute of Mental Health dylann.nielson@nih.gov Peter Bandettini National Institute of Mental Health bandettini@nih.gov Francisco Pereira National Institute of Mental Health francisco.pereira@nih.gov |
| Pseudocode | No | The paper provides mathematical formulations and descriptions of the methods but does not include any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide any concrete access information for source code, such as a repository link, an explicit code release statement, or mention of code in supplementary materials. |
| Open Datasets | Yes | We use several s MRI datasets collected at different sites. We train networks using 956 s MRI volumes collected by the Human Connectomme Project (HCP) [30], 1,136 s MRI volumes collected by the Nathan Kline Institute (NKI) [22], 183 s MRI volumes collected by the Buckner Laboratory [2], and 120 s MRI volumes from the Washington University 120 (WU120) dataset [24]. |
| Dataset Splits | No | A 90-10 training-test split was used for the HCP, NKI, Buckner, and WU120 datasets. The paper does not specify a separate validation dataset split. |
| Hardware Specification | No | The paper does not provide specific hardware details such as GPU or CPU models, memory amounts, or detailed computer specifications used for running the experiments. |
| Software Dependencies | No | The paper mentions software like Freesurfer and optimizers like Adam, but does not provide specific version numbers for any key software components or libraries required for reproduction. |
| Experiment Setup | Yes | All networks were trained with Adam [13] and an initial learning rate of 0.001. The batch-size was set to 10. Weight normalization [28] was used for the weight means for all networks and the weight standard deviations were initialized to 0.001 as in [19] for the variational network trained on HCP. For MAP networks and the variational network trained on HCP, p(w) = N(0, 1). |