Structured Inference Networks for Nonlinear State Space Models

Authors: Rahul Krishnan, Uri Shalit, David Sontag

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

Reproducibility Variable Result LLM Response
Research Type Experimental We apply the learning algorithm to both synthetic and real-world datasets, demonstrating its scalability and versatility. We find that using the structured approximation to the posterior results in models with significantly higher held-out likelihood.
Researcher Affiliation Academia Rahul G. Krishnan, Uri Shalit, David Sontag Courant Institute of Mathematical Sciences, New York University {rahul, shalit, dsontag}@cs.nyu.edu
Pseudocode Yes Algorithm 1 depicts an overview of the learning algorithm.
Open Source Code Yes Code for learning DMMs and reproducing our results may be found at: github.com/clinicalml/structuredinference
Open Datasets Yes Polyphonic Music: We train DMMs on polyphonic music data (Boulanger-lewandowski, Bengio, and Vincent 2012).
Dataset Splits Yes We learn for 2000 epochs and report results based on early stopping using the validation set.
Hardware Specification Yes The Tesla K40s used for this research were donated by the NVIDIA Corporation.
Software Dependencies Yes Our models and learning algorithm are implemented in Theano (Theano Development Team 2016). We use Adam (Kingma and Ba 2015) with a learning rate of 0.0008 to train the DMM.
Experiment Setup Yes We use Adam (Kingma and Ba 2015) with a learning rate of 0.0008 to train the DMM.