Fisher Efficient Inference of Intractable Models

Authors: Song Liu, Takafumi Kanamori, Wittawat Jitkrittum, Yu Chen

NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Numerical studies validate our asymptotic theorems and we give an example where DLE successfully estimates an intractable model constructed using a pre-trained deep neural network.
Researcher Affiliation Academia Song Liu University of Bristol The Alan Turing Institute, UK song.liu@bristol.ac.uk Takafumi Kanamori Tokyo Institute of Technology, RIKEN, Japan kanamori@c.titech.ac.jp Wittawat Jitkrittum Max Planck Institute for Intelligent Systems, Germany wittawat@tuebingen.mpg.de Yu Chen University of Bristol, UK yc14600@bristol.ac.uk
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code Yes See https://github.com/lamfeeling/ Stein-Density-Ratio-Estimation for code demos on SDRE and model inference.
Open Datasets Yes Figure 2: MNIST images with the highest (upper red box) and the lowest unnormalized density (lower green box) estimated on each digit by DLE and NCE.
Dataset Splits No The paper mentions drawing samples (e.g., "nq = 500 samples" or "nq = 100 randomly selected images") but does not specify explicit training, validation, or test splits by percentages, counts, or pre-defined methodologies.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running the experiments.
Software Dependencies No The paper does not specify particular software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions).
Experiment Setup Yes We run DLE 6000 times with new batch of Xq each time and obtain an empirical distribution of nqˆθ. ... For DLE, we set f(x) := [x, x2] and for KSD, we use a polynomial kernel with degree 2. ... The dataset Xqi contains nq = 100 randomly selected images from a single digit i and we use DLE and NCE to estimate ˆθi for each digit i. For DLE, we set f(x) = ψ(x).