Boosted Density Estimation Remastered
Authors: Zac Cranko, Richard Nock
ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Let t 2 {0, . . . , T}. Our experiments mostly take place in X def= R2, where we use a simple neural network classifier ct 2 C(X), which we train using cross entropy error by post composing it with the logistic sigmoid: σ ct. After training ct we transform it into to a density ratio using an exponential function: dt def= exp ct (cf. 2) which we use to update Qt 1. |
| Researcher Affiliation | Collaboration | Zac Cranko 1 2 Richard Nock 2 1 3 1The Australian National University 2Data61 3The University of Sydney. |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | The boosted density implementation is available at https://github.com/Zac Cranko/Boosted Densities.jl. |
| Open Datasets | Yes | To compare the performance of our model (here called DISCRIM) with ADAGAN we replicate their Gaussian mixture toy experiment,5 fitting a randomly located eight component isotropic Gaussian mixture where each component has constant variance. These are sampled using the code provided by Tolstikhin et al. (2017).5This is the experiment gaussian gmm.py at github.com/tolstikhin/adagan |
| Dataset Splits | No | The paper describes metrics like 'test accuracy' and mentions 'mean and 95% confidence intervals', but does not provide specific details on training/validation/test dataset splits (e.g., percentages or sample counts), nor does it explicitly mention a separate validation set. |
| Hardware Specification | No | The paper does not provide specific hardware details (like GPU/CPU models, memory amounts, or detailed computer specifications) used for running its experiments. |
| Software Dependencies | No | The paper mentions 'Julia' and references 'Flux' (an ML library for Julia) in the bibliography, but it does not provide specific version numbers for these or any other software dependencies used in the experiments. |
| Experiment Setup | Yes | Complete details about the experimental procedures including target data and network architectures are deferred to the supplementary material ( E). ... For this experiment the target distribution P is a mixture of 8 Gaussians that are randomly positioned at the beginning of each run of training. Let m n denote a fully connected neural network ct with m hidden layers and n nodes per layer. After each hidden layer we apply the SELU activation function. |