Combining Generative and Discriminative Models for Hybrid Inference

Authors: Victor Garcia Satorras, Zeynep Akata, Max Welling

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

Reproducibility Variable Result LLM Response
Research Type Experimental We apply our ideas to the Kalman filter, a Gaussian hidden Markov model for time sequences, and show, among other things, that our model can estimate the trajectory of a noisy chaotic Lorenz Attractor much more accurately than either the learned or graphical inference run in isolation.
Researcher Affiliation Collaboration Victor Garcia Satorras Uv A-Bosch Delta Lab University of Amsterdam Netherlands v.garciasatorras@uva.nl Zeynep Akata Cluster of Excellence ML University of Tübingen Germany zeynep.akata@uni-tuebingen.de Max Welling Uv A-Bosch Delta Lab University of Amsterdam Netherlands m.welling@uva.nl
Pseudocode No No. The paper describes the model through equations and textual descriptions but does not include a clearly labeled pseudocode or algorithm block.
Open Source Code Yes 2Available at: https://github.com/vgsatorras/hybrid-inference
Open Datasets Yes To demonstrate the generalizability of our Hybrid model to real world datasets, we use the Michigan NCLT [6] dataset which is collected by a segway robot moving around the University of Michigan s North Campus.
Dataset Splits Yes We sample two different motion trajectories from 50 to 100K time steps each, one for validation and the other for training. An additional 10K time steps trajectory is sampled for testing.
Hardware Specification No No. The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments.
Software Dependencies No No. The paper mentions software components like 'Adam optimizer', 'MLPs', and 'GRU', but does not provide specific version numbers for any software dependencies.
Experiment Setup Yes We set γ = 0.005 and use the Adam optimizer with a learning rate 10 3. The number of inference iterations used in the Hybrid model, GNN-messages and GM-messages is N=50. fe and fdec are a 2-layers MLPs with Leaky Relu and Relu activations respectively. The number of features in the hidden layers of the GRU, fe and fdec is nf=48.