Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The Shift Toward Open and Reproducible AI Research," Under Review, 2026. doi: 10.48550/arXiv.2606.16974..

Boosted Generative Models

Authors: Aditya Grover, Stefano Ermon

AAAI 2018 | Venue PDF | LLM Run Details | Input Tokens: 16,468 Total number of tokens sent to the LLM as input for this paper's analysis. | Output Tokens: 6,119 Total number of tokens produced by the LLM (including reasoning/thinking tokens) for this paper's analysis.

Reproducibility Variable Result LLM Response
Research Type Experimental Our experiments are designed to demonstrate the superiority of the proposed boosting meta-algorithms on a wide variety of generative models and tasks.
Researcher Affiliation Academia Aditya Grover, Stefano Ermon Department of Computer Science Stanford University EMAIL
Pseudocode Yes Algorithm 1 Gen BGM(X = {xi}m i=1, T rounds) and Algorithm 2 Disc BGM(X = {xi}m i=1, T rounds, f-div)
Open Source Code Yes A reference implementation of the boosting meta-algorithms is available at https://github.com/ermongroup/bgm.
Open Datasets Yes binarized MNIST dataset of handwritten digits (Le Cun, Cortes, and Burges 2010). http://yann. lecun. com/exdb/mnist. benchmark datasets (Van Haaren and Davis 2012).
Dataset Splits Yes We observe 1, 000 training samples drawn independently from the data distribution... The test set contains 1, 000 samples from the same distribution. Model weights are chosen based on cross-validation.
Hardware Specification No No specific hardware details (e.g., CPU, GPU models, or memory) used for running experiments were provided in the paper.
Software Dependencies No The paper mentions using the 'Adam optimizer' and 'variational autoencoders (VAE)' and 'convolutional neural network (CNN)' but does not provide specific version numbers for any software dependencies or libraries.
Experiment Setup Yes The classi๏ฌers for Disc BGM are multi-layer perceptrons with two hidden layers of 100 units each and Re LU activations, trained to maximize f-divergences corresponding to the negative cross-entropy (NCE) and Hellinger distance (HD) using the Adam optimizer (Kingma and Welling 2014). We set T = 2 rounds for additive boosting and Gen BGM. VAE hidden layer architecture given in parenthesis (e.g., 200-100, 200-100-100, 300-100, 100-50).