Contextual Feature Selection with Conditional Stochastic Gates

Authors: Ram Dyuthi Sristi, Ofir Lindenbaum, Shira Lifshitz, Maria Lavzin, Jackie Schiller, Gal Mishne, Hadas Benisty

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

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we conduct an extensive benchmark using simulated and real-world datasets across multiple domains demonstrating that c-STG can lead to improved feature selection capabilities while enhancing prediction accuracy and interpretability. and We conduct comprehensive empirical evaluations on simulated and real-world datasets across healthcare, housing, and neuroscience, demonstrating the effectiveness and adaptability of our proposed methods compared to existing techniques.
Researcher Affiliation Academia 1University of California San Diego, La Jolla, California, USA 2Bar-Ilan University , Ramat Gan, Israel 3Technion Israel Institute of Technology, Haifa, Israel
Pseudocode Yes Algorithm 1 Weighted c-STG
Open Source Code Yes Code for c-STG is available in https://github.com/ Mishne-Lab/Conditional-STG
Open Datasets Yes Heart disease dataset: We now focus on medical data, specifically, the heart disease dataset from UCI ML repository (Janosi et al., 1988). and The Housing dataset (Lianjia, 2017).
Dataset Splits Yes The selection of model parameters/hyperparameters was based on preventing issues like underfitting and overfitting and ensuring optimal 5-fold cross-validated performance. and Table 1 shows the 5-fold cross-validation accuracy where we surpass other methods.
Hardware Specification Yes We trained all networks using CUDA-accelerated Py Torch implementations on a NVIDIA Quadro RTX8000 GPU.
Software Dependencies No It mentions "CUDA-accelerated Py Torch implementations" but does not provide version numbers for PyTorch or CUDA.
Experiment Setup Yes To determine the best hyperparameters, namely the learning rate (η) and regularization coefficient (λ), we performed a grid search over the following values: η {1e 1, 5e 2, 1e 2, 5e 3, 1e 3, 5e 4, 1e 4} and λ {1, 5e 1, 1e 1, 5e 2, 1e 2, 5e 3, 1e 3}. The same set of values was used for the grid search across all the datasets.