Curve Your Enthusiasm: Concurvity Regularization in Differentiable Generalized Additive Models
Authors: Julien Siems, Konstantin Ditschuneit, Winfried Ripken, Alma Lindborg, Maximilian Schambach, Johannes Otterbach, Martin Genzel
NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We validate the effectiveness of our regularizer in experiments on synthetic as well as real-world datasets for time series and tabular data. |
| Researcher Affiliation | Collaboration | Julien Siems University of Freiburg siemsj@cs.uni-freiburg.de Konstantin Ditschuneit* Scenarium AI ko.ditschuneit@gmail.com Winfried Ripken* Merantix Momentum winfried.ripken@merantix.com Alma Lindborg* Merantix Momentum alma.lindborg@merantix.com Maximilian Schambach Merantix Momentum maximilian.schambach@merantix.com Johannes S. Otterbach nyonic johannes@nyonic.ai Martin Genzel Merantix Momentum martin.genzel@merantix.com |
| Pseudocode | No | The paper does not include any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | 1Code: https://github.com/merantix-momentum/concurvity-regularization |
| Open Datasets | Yes | Boston Housing [23], California Housing [37], Adult [18], MIMIC-II [29], MIMIC-III [27] and Support2 [14]. |
| Dataset Splits | Yes | We sample 10,000 datapoints from the model and use 7,000, 2,000, 1,000 for training, validation and testing respectively. We use the validation split to find adequate hyperparameters via a small manual search. |
| Hardware Specification | Yes | The results are shown in Figure 10, all obtained with an M1 Mac Book Pro. |
| Software Dependencies | No | The paper mentions several software components and libraries like "Adam W optimizer", "Cosine Annealing", "py GAM", "Optuna", "TensorFlow", "JAX", and "PyTorch", but it does not specify their version numbers. |
| Experiment Setup | Yes | The hyperparameter space and default parameters are shown in Table 1 and the hyperparameters per dataset are shown in Figure 7. (Table 1 includes Learning Rate [1e-4, 1e-1], Weight Decay [1e-6, 1], Activation [ELU, GELU, Re LU], # of neurons per layer [2, 256], # of hidden layers [1, 6], Num. Epochs [10, 500]) |