Constraining the Dynamics of Deep Probabilistic Models
Authors: Marco Lorenzi, Maurizio Filippone
ICML 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The proposed approach is extensively tested in several experimental settings, leading to highly competitive results in challenging modeling applications, while offering high expressiveness, flexibility and scalability. |
| Researcher Affiliation | Academia | 1University of Cote d Azur, INRIA Sophia Antipolis, EPIONE research group, France 2EURECOM, Department of Data Science, Sophia Antipolis, France. |
| Pseudocode | No | The paper describes the proposed methods and inference schemes mathematically and in prose, but it does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not include an explicit statement about releasing its source code or a direct link to a code repository for the methodology described. |
| Open Datasets | No | The paper mentions using "Lotka-Volterra", "Fitz Hugh-Nagumo", "protein biopathways from Vyshemirsky & Girolami (2007)", "Lorenz96 system" (for which data was generated: "generated 32 equally spaced observations"), and "mortality dataset from (Broffitt, 1988)". None of these provide a specific link, DOI, or repository name for a publicly available dataset, nor are they explicitly referred to as standard, publicly accessible datasets with direct access information. |
| Dataset Splits | No | The paper specifies total sample sizes like 'n < 50' or '80 and 1000 points' and refers to '5 different realizations of the noise' for experimental results, but it does not provide specific train/validation/test dataset split percentages, absolute sample counts for splits, or clear citations to predefined splits. |
| Hardware Specification | Yes | All the experiments were performed on a 1.3GHz Intel Core i5 Mac Book. |
| Software Dependencies | No | The paper mentions software like the 'GPstuff toolbox,' 'R package KGode,' 'R package de Grad Infer,' and the 'Adam' optimizer, but it does not provide specific version numbers for any of these software dependencies. |
| Experiment Setup | Yes | The length-scale of the RBF covariances was initialized to λ0 = log(tmax tmin), while the marginal standard deviation to α0 = log(ymax ymin); the initial likelihood noise was set to σ2 0 = α0/105. Finally, the initial ODE parameters were set to the value of 0.1. The optimization was carried out through stochastic gradient descent with Adaptive moment Estimation (Adam) (Kingma & Ba, 2017), through the alternate optimization of i) the approximated posterior over W and likelihood/covariance parameters (q(W) and ψ), and ii) likelihood parameters of ODE constraints and the approximate posterior over ODE parameters (ψD and q(θ)). |