Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Deep Legendre Transform
Authors: Aleksey Minabutdinov, Patrick Cheridito
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Numerical experiments demonstrate our method s ability to deliver accurate results across different high-dimensional examples. Moreover, by employing symbolic regression with Kolmogorov Arnold networks, it is able to obtain the exact convex conjugates of specific convex functions. |
| Researcher Affiliation | Academia | Aleksey Minabutdinov Center of Economic Research and Risk Lab ETH Zurich, Switzerland EMAIL. Patrick Cheridito Department of Mathematics and Risk Lab ETH Zurich, Switzerland EMAIL. |
| Pseudocode | No | The paper describes the Deep Legendre Transform (DLT) method and related procedures in prose and mathematical formulations (e.g., equations 1.5, 2.2, 4.2a, 4.2b), but it does not include a clearly labeled pseudocode block, algorithm box, or structured steps formatted like code. |
| Open Source Code | Yes | Code is available at https://github.com/lexmar07/Deep-Legendre-Transform |
| Open Datasets | No | The paper describes the mathematical functions (e.g., Quadratic, Neg-Log, Neg-Entropy functions from Table 1, and quadratic-over-linear function) for which convex conjugates are computed. It also details the distributions from which samples (e.g., from a standard half-normal distribution, uniform distribution, multivariate normal distribution) are generated for training and testing, rather than using specific named publicly available datasets. |
| Dataset Splits | Yes | Let Xtest be a test set of random points x C independently sampled from a probability distribution ยต over C. ... Xtrain = C h หฯ(Ytrain) and Xtest = C h หฯ(Ytest) can be used for training and testing gฮธ. ... Table 2: Errors were computed with Monte Carlo on test set of 4096 i.i.d. random points sampled from Unif ( f) 1. |
| Hardware Specification | Yes | All computations were performed on NVIDIA T4 GPU and Intel Core Xeon CPUs running Python 3.11.12. |
| Software Dependencies | No | The paper mentions 'Python 3.11.12' and 'JAX' for efficient computation, but JAX is not accompanied by a specific version number. It also mentions 'Adam optimizer' without specifying a software library version. The rule requires multiple key software components with versions or a self-contained solver with a version, which is not met here. |
| Experiment Setup | Yes | We use MLPs with two hidden layers containing 128 units and GELU activation. Our Res Nets have two residual blocks, each containing two dense layers, resulting in a 4-layer architecture with skip connections and GELU activation. ... Parameters were initialized according to a Gaussian distribution... For training, we used the Adam optimizer. ... Adam optimizer with learning rate 10^-3 and batch size 128d. ... 120,000 iterations with a learning rate of 10^-3 and Huber loss (ฮด = 1.0). |