Learning Deep Attribution Priors Based On Prior Knowledge
Authors: Ethan Weinberger, Joseph Janizek, Su-In Lee
NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We apply our method to a synthetic dataset and two real-world biological datasets. Empirically we find that models trained using the DAPr framework achieve better performance on tasks where training data is limited. |
| Researcher Affiliation | Academia | Ethan Weinberger Paul G. Allen School of Computer Science University of Washington Seattle, WA 98195 ewein@cs.washington.edu |
| Pseudocode | No | The paper does not contain explicitly labeled 'Pseudocode' or 'Algorithm' blocks. |
| Open Source Code | No | The paper does not provide an explicit statement about releasing source code or a direct link to a code repository for the methodology described. |
| Open Datasets | Yes | The data for this experiment comes from multiple datasets available on the Accelerating Medicines Partnership Alzheimer s Disease Project (AMP-AD) portal1; in particular we make use of the Adult Changes in Thought (ACT) [24], Mount Sinai Brain Bank (MSBB), and Religious Orders Study/Memory and Aging Project (ROSMAP) [1] datasets. [Footnote 1: https://adknowledgeportal.synapse.org/]. Our data comes from the Beat AML dataset...[39]. |
| Dataset Splits | Yes | For each dataset we use 20% of the dataset for model training, and divide the remaining points evenly into testing and validation sets, giving a final 20%-40%-40% train-test-validation split. (Section 4.1) |
| Hardware Specification | No | The paper mentions training models and using attribution methods but does not specify any particular GPU model, CPU, or other hardware details used for the experiments. |
| Software Dependencies | No | The paper mentions using 'Adam' for optimization and 'Enrichr' library, but does not provide specific version numbers for these or any other software dependencies. |
| Experiment Setup | Yes | We optimize our models using Adam [14] with early stopping, and all hyperparameters are chosen based on performance on our validation sets (see Supplement for additional details). ... For a given number of features p the first hidden layer has p/2 units, and the second has p/4 units. ... All prediction model MLPs have two hidden layers with 512 and 256 units respectively, and for our DAPr models we use MLPs with one hidden layer containing four units. |