Smoothed Gradients for Stochastic Variational Inference

Authors: Stephan Mandt, David Blei

NeurIPS 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We test our method on latent Dirichlet allocation with three large corpora. We tested SVI for LDA, using the smoothed stochastic gradients, on three large corpora: 882K scientific abstracts from the Arxiv repository, using a vocabulary of 14K words. 1.7M articles from the New York Times, using a vocabulary of 8K words. 3.6M articles from Wikipedia, using a vocabulary of 7.7K words.
Researcher Affiliation Academia Stephan Mandt Department of Physics Princeton University smandt@princeton.edu David Blei Department of Computer Science Department of Statistics Columbia University david.blei@columbia.edu
Pseudocode Yes Algorithm 1: Smoothed stochastic gradients for Latent Dirichlet Allocation
Open Source Code No The paper does not provide any concrete access information (e.g., specific repository link, explicit code release statement) for the source code of the methodology described.
Open Datasets Yes We tested SVI for LDA, using the smoothed stochastic gradients, on three large corpora: 882K scientific abstracts from the Arxiv repository, using a vocabulary of 14K words. 1.7M articles from the New York Times, using a vocabulary of 8K words. 3.6M articles from Wikipedia, using a vocabulary of 7.7K words.
Dataset Splits No The paper mentions separating a 'test set from the training set' and then splitting the test set for evaluation, but it does not specify a distinct validation set or provide explicit percentages for training, validation, and test splits needed for reproduction.
Hardware Specification No The paper does not provide specific hardware details (e.g., exact GPU/CPU models, processor types) used for running its experiments.
Software Dependencies No The paper does not provide specific ancillary software details, such as library names with version numbers, needed to replicate the experiment.
Experiment Setup Yes We set the minibatch size to B = 300 and furthermore set the number of topics to K = 100, and the hyper-parameters α = η = 0.5. We fixed the learning rate to ρ = 10 3.