How to address monotonicity for model risk management?

Authors: Dangxing Chen, Weicheng Ye

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental As a result of empirical examples, we demonstrate that monotonicity is often violated in practice and that monotonic groves of neural additive models are transparent, accountable, and fair.
Researcher Affiliation Academia 1Zu Chongzhi Center for Mathematics and Computational Sciences, Duke Kunshan University, Kunshan, Jiangsu, China. Correspondence to: Dangxing Chen <dangxing.chen@dukekunshan.edu.cn>.
Pseudocode Yes Algorithm 1 Monotonic Groves of Neural Additive Model
Open Source Code Yes The code is built and modified based on (Tshitoyan, 2023).
Open Datasets Yes A popularly used dataset is the Kaggle credit score dataset 1. ... A report published by Pro Publica in 2016 provided recidivism data for defendants in Broward County, Florida (Pro, 2016). ... This dataset (Ahmad et al., 2017; Chicco & Jurman, 2020) contains the medical records of 299 patients who had heart failure...
Dataset Splits No For all our experiments, the dataset is randomly partitioned into 75% training and 25% test sets.
Hardware Specification No The paper does not provide specific details about the hardware used for experiments.
Software Dependencies No The paper mentions 'The code is built and modified based on (Tshitoyan, 2023).' but does not specify software versions (e.g., Python, PyTorch, TensorFlow versions).
Experiment Setup Yes For all our experiments, the dataset is randomly partitioned into 75% training and 25% test sets. All neural networks contain 1 hidden layer with 2 units, logistic activation, and no regulation.