Maximum Entropy Flow Networks

Authors: Gabriel Loaiza-Ganem *, Yuanjun Gao *, John P. Cunningham

ICLR 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Simulation results demonstrate the effectiveness of our method, and applications to finance and computer vision show the flexibility and accuracy of using maximum entropy flow networks.
Researcher Affiliation Academia Gabriel Loaiza-Ganem , Yuanjun Gao & John P. Cunningham Department of Statistics Columbia University New York, NY 10027, USA {gl2480,yg2312,jpc2181}@columbia.edu
Pseudocode Yes Algorithm 1 Training the MEFN
Open Source Code No The paper adapts existing codebases (e.g., 'https://github.com/Proof By Construction/texture-networks' and 'https://github.com/taesung89/real-nvp') but does not explicitly state that the authors' own implementation code for MEFN is open-source or provide a link to it.
Open Datasets No The paper uses real-world data like 'European call options on Nov. 1 2016 for the stock AAPL (Apple inc.)' and 'RGB representation of the 224 224 pixel images' for texture synthesis, and also constructs a synthetic problem with a known solution (Dirichlet), but it does not provide concrete access information, citations to established benchmark datasets, or links for any of the data used.
Dataset Splits No The paper mentions using 'training data' and 'testing data' for the option pricing application, and specifies sample sizes for SGD steps (n, n), but it does not provide explicit train/validation/test dataset splits, percentages, or absolute counts, nor does it refer to predefined standard splits for any dataset.
Hardware Specification No The paper does not provide specific hardware details such as GPU or CPU models, or cloud computing specifications used for running its experiments.
Software Dependencies No The paper mentions software like 'ADADELTA', 'ADAM', and 'Tensor Flow' but does not specify their version numbers, which is necessary for reproducible software dependencies.
Experiment Setup Yes For 4.1 and B, We use 10 layers of planar flow... For 4.3 we use real NVP structure and use ADAM (Kingma & Ba, 2014) with learning rate = 0.001. For all our experiments, we use imax = 3000, β = 4, γ = 0.25. For 4.1 and B we use n = 300, n = 1000, kmax = 10; For 4.3 we use n = n = 2, kmax = 8.