Learning Equations for Extrapolation and Control

Authors: Subham Sahoo, Christoph Lampert, Georg Martius

ICML 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 4. Experimental evaluation, In Fig. 5 the numerical results and also an illustrative output of EQL and the baselines are presented.
Researcher Affiliation Academia 1Indian Institute of Technology, Kharagpur, India 2IST Austria, Klosterneuburg, Austria 3Max Planck Institute for Intelligent Systems, Tübingen, Germany.
Pseudocode No The paper describes the method using diagrams and text, but does not contain structured pseudocode or algorithm blocks.
Open Source Code Yes The code and some data is available at https://github.com/martius-lab/EQL.
Open Datasets No The paper describes generating its own training data by sampling points and adding noise, but does not provide concrete access information for a publicly available or open dataset.
Dataset Splits Yes For all experiments, we have training data in a restricted domain, usually [ 1, 1]d corrupted with noise which is split into training and validation with 90% 10% split.
Hardware Specification No The paper does not provide specific hardware details (exact GPU/CPU models, processor types, or memory amounts) used for running its experiments.
Software Dependencies No The paper states it was implemented in 'python based on the theano framework', but does not provide specific version numbers for these software components.
Experiment Setup Yes The following hyper-parameters were fixed: learningrate (Adam) α = 0.001, regularization (Adam) of ϵ = 0.0001, minibatch size of 20, number of units u = v = 10, i. e. 10 units per type in each layer. We use t1 = 1 4T and t2 = 19 20T, where T is the total number of epochs, large enough to ensure convergence, i. e. T = (L 1) 10000. Note, that early stopping will be disadvantageous.