Density Ratio Estimation with Doubly Strong Robustness

Authors: Ryosuke Nagumo, Hironori Fujisawa

ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Numerical experiments show that our proposals are more robust than the previous methods. In Section 4, numerical experiments illustrate that the proposed methods are more robust than the past ones.
Researcher Affiliation Collaboration 1The Graduate University for Advanced Studies (SOKENDAI), Tokyo, Japan 2Panasonic Holdings Corporation, Osaka, Japan 3Institute of Statistical Mathematics, Tokyo, Japan.
Pseudocode No The paper describes its methods and optimization procedures using mathematical formulations and descriptive text, but it does not include structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide a concrete access link or explicit statement for the open-sourcing of the code for its proposed methodology. While it references a GitHub link for a comparison method (Ru LSIF), it does not do so for its own contributions.
Open Datasets Yes We used a human activity dataset provided by the Human Activity Sensing Consortium (HASC) Challenge 2011, which was used in the experiment of change detection with DRE (Liu et al., 2013).
Dataset Splits Yes The dataset sizes were set to np = nq = 100. The dataset sizes of the reference and target datasets were set to np = nq = 100.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments.
Software Dependencies No The paper mentions using a Python code for a comparison method but does not provide specific software dependencies with version numbers for its own implementation or experimental setup.
Experiment Setup Yes The weight function was set to w(x) = exp x 4 4/50 . The trimming quantile ν in Trimmed DRE was set to the true contamination ratio. The parameter γ in γ-DRE was set to 0.01. No regularization term was added to the objective function. We added the elastic net regularization term λ1 θ 1 + λ2 θ 2 2 with λ1 = λ2 = 0.5 to the objective function.