Faithful Model Explanations through Energy-Constrained Conformal Counterfactuals
Authors: Patrick Altmeyer, Mojtaba Farmanbar, Arie van Deursen, Cynthia C. S. Liem
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Through extensive empirical studies, we demonstrate that ECCCo reconciles the need for faithfulness and plausibility.Empirical Analysis Our goal in this section is to shed light on the following research questions: |
| Researcher Affiliation | Collaboration | 1Delft University of Technology, Faculty of Electrical Engineering, Mathematics and Computer Science 2ING Bank |
| Pseudocode | No | The paper describes algorithmic procedures through equations and text, but it does not include a clearly labeled pseudocode block or an algorithm block with structured steps. |
| Open Source Code | Yes | All code used for the analysis in this paper can be found here: https://github.com/pat-alt/ECCCo.jl. |
| Open Datasets | Yes | From the credit and finance domain we include three tabular datasets: Give Me Some Credit (GMSC) (Kaggle 2011), German Credit (Hoffman 1994) and California Housing (Pace and Barry 1997). ... we also include two image datasets: MNIST (Le Cun 1998) and Fashion MNIST (Xiao, Rasul, and Vollgraf 2017). |
| Dataset Splits | No | The paper mentions splitting the training data into a proper training dataset and a calibration dataset for conformal prediction, and uses 'calibration data' or a 'holdout set Dcal'. However, it does not provide specific percentages or counts for the overall train/validation/test splits used in their experiments, nor does it cite specific predefined splits. |
| Hardware Specification | Yes | Research reported in this work was partially or completely facilitated by computational resources and support of the Delft Blue (Delft High Performance Computing Centre DHPC) and the Delft AI Cluster (DAIC: https://doc.daic.tudelft.nl/) at TU Delft. Detailed information about the utilized computing resources can be found in the appendix. |
| Software Dependencies | No | The paper mentions that the analysis was performed using a 'Julia package' and refers to 'Julia Con Conferences' in a citation, implying Julia as the programming language. However, it does not specify the version of Julia or any other software dependencies (e.g., libraries, frameworks) with specific version numbers. |
| Experiment Setup | No | The paper states 'To tune these hyperparameters we have relied on grid search' and lists the types of models used (MLP, deep ensembles, JEM, Le Net-5 CNN). However, it defers 'Full details concerning model training' to the appendix and does not provide specific hyperparameter values or detailed training configurations in the main text. |