Knockout: A simple way to handle missing inputs
Authors: Minh Nguyen, Batuhan K. Karaman, Heejong Kim, Alan Q. Wang, Fengbei Liu, Mert R. Sabuncu
TMLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate Knockout across a wide range of simulations and real-world datasets and show that it offers strong empirical performance. |
| Researcher Affiliation | Academia | Minh Nguyen EMAIL Cornell University Batuhan K. Karaman EMAIL Cornell University Heejong Kim EMAIL Weill Cornell Medicine Alan Q. Wang EMAIL Cornell University Fengbei Liu EMAIL Cornell University Mert R. Sabuncu EMAIL Cornell University and Weill Cornell Medicine for the Alzheimer s Disease Neuroimaging Initiative |
| Pseudocode | No | The paper describes the Knockout method and its theoretical justification using mathematical equations and descriptive text, but it does not include any explicitly labeled pseudocode or algorithm blocks with structured steps formatted like code. |
| Open Source Code | No | The paper does not contain any explicit statements about releasing source code for the described methodology, nor does it provide links to any code repositories. |
| Open Datasets | Yes | Data used in preparation of this article were obtained from the Alzheimer s Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu). We use data from the Alzheimer s Disease Neuroimaging Initiative (ADNI) database (Mueller et al., 2005) and adopt the state-of-the-art model from (Karaman et al., 2022). We evaluate model performance on CIFAR-10H (Peterson et al., 2019) and CIFAR-10/100N (Wei et al., 2021). In particular, we experiment on a multi-modal tumor segmentation task (Baid et al., 2021). The Auto Arborist dataset (Beery et al., 2022), a multi-view (street and aerial) image dataset, is used for this purpose. The dataset consists of T2-weighted (T2w), diffusion-weighted (DWI) and apparent diffusion coefficient (ADC) MR images per subject (Saha et al., 2022). To further evaluate the generalizability of our approach, we extend our experiments to the UPMC Food-101 dataset (Wang et al., 2015) |
| Dataset Splits | Yes | In each simulation, we sample 30k data points in total and use 10% for training. These results are averaged over 10 random 80-20 train-test splits. We follow the standard train-test split, reserving 7,000 samples from the training set for validation. |
| Hardware Specification | No | The paper does not specify any particular GPU models, CPU models, or other hardware components used for running the experiments. |
| Software Dependencies | No | The paper mentions 'Adam (Kingma & Ba, 2014)' as an optimizer and 'Wide-Res Net-10-28 (Zagoruyko & Komodakis, 2016)', 'Res Net-50 (He et al., 2016)', 'Vi T-B-16 (Dosovitskiy et al., 2021)', '3D UNet (Ronneberger et al., 2015)', 'Res Net50 (vision)', and 'Ro BERTa-base (text)' as models, but it does not specify any software libraries or packages with their version numbers required for replication. |
| Experiment Setup | Yes | All methods use the same neural network architecture composed of a 3-layer multi-layer perceptron (MLP) with hidden layers 100 and Re LU activations. Training is done using Adam (Kingma & Ba, 2014) with learning rate 3e-3 for 5k steps. We minimize a sum of cross-entropy loss and Dice loss with equal weighting and use Adam optimizer with a learning rate of 1e-3. Models are trained using Adam W with a learning rate of 3 10 5, for 8 epochs and a batch size of 72. |