Regression with Sensor Data Containing Incomplete Observations

Authors: Takayuki Katsuki, Takayuki Osogami

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

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the advantages of our algorithm through numerical experiments. and Extensive experiments on synthetic and six real-world regression tasks including a real use case for healthcare demonstrate the effectiveness of the proposed method.
Researcher Affiliation Industry 1IBM Research Tokyo, Tokyo, Japan. Correspondence to: Takayuki Katsuki <kats@jp.ibm.com>.
Pseudocode Yes Algorithm 1 U2 regression based on stochastic gradient method.
Open Source Code No The paper does not provide a specific link or explicit statement about the availability of its own source code for the methodology described.
Open Datasets Yes We use three synthetic tasks, Low Noise, High Noise, and Breathing, collected from the Kaggle dataset (Sen, 2016). and We next apply the proposed method and baselines to five different real-world healthcare tasks from the UCI Machine Learning Repository (Velloso, 2013; Velloso et al., 2013)
Dataset Splits Yes We conducted 5-fold cross-validation, each with a different randomly sampled training-testing split. For evaluation purposes, we do not include incomplete observations in these test sets. For each fold of the cross-validation, we use a randomly sampled 20% of the training set as a validation set to choose the best hyperparameters for each algorithm.
Hardware Specification Yes All of the experiments were carried out with a Python and Tensor Flow implementation on workstations having 80 GB of memory, a 4.0 GHz CPU, and an Nvidia Titan X GPU.
Software Dependencies No The paper mentions using 'Python and Tensor Flow implementation' but does not specify version numbers for these software components or any other libraries.
Experiment Setup Yes We set the candidates of the hyperparameters, ρ and λ, to {10−3, 10−2, 10−1, 100}. and We used Adam with the hyperparameters recommended in (Kingma & Ba, 2015), and the number of samples in the mini-batches was set to 32. and We used a 6-layer multilayer perceptron with Re LU (Nair & Hinton, 2010) (more specifically, D-100-100-100-1) as f(x)