Infinite Factorial Dynamical Model

Authors: Isabel Valera, Francisco Ruiz, Lennart Svensson, Fernando Perez-Cruz

NeurIPS 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate the performance of our i FDM on four well-known applications: multitarget tracking, cocktail party, power disaggregation, and multiuser detection. Our experimental results show that our approach for source separation does not only outperform previous approaches, but it can also handle problems that were computationally intractable for existing approaches.
Researcher Affiliation Collaboration Isabel Valera Max Planck Institute for Software Systems ivalera@mpi-sws.org Francisco J. R. Ruiz Department of Computer Science Columbia University f.ruiz@columbia.edu Lennart Svensson Department of Signals and Systems Chalmers University of Technology lennart.svensson@chalmers.se Fernando Perez-Cruz Universidad Carlos III de Madrid, and Bell Labs, Alcatel-Lucent fernandop@ieee.org
Pseudocode Yes Figure 2b: PGAS algorithm.
Open Source Code Yes Code for these applications can be found at https://github.com/franrruiz/i FDM
Open Datasets Yes We validate the performance of the i FDM on two different real databases: the Reference Energy Disaggregation Data Set (REDD) [11], and the Almanac of Minutely Power Dataset (AMP) [15].
Dataset Splits No The paper describes data collection and mixing for the applications but does not specify explicit train/validation/test splits (e.g., percentages, counts, or references to predefined splits).
Hardware Specification No The paper does not provide any specific hardware details (e.g., CPU, GPU models, memory, or cloud instance types) used for running the experiments.
Software Dependencies No The paper does not list specific software dependencies with version numbers.
Experiment Setup Yes For the PGAS kernel, we use P = 3, 000 particles in all our experiments. For Multitarget Tracking: Ts = 0.5 is the sampling period, and the initial velocity is Gaussian distributed with zero mean and covariance 0.01I. For Cocktail Party: Gaussian noise with standard deviation 0.3, and the weighting vector wm N(0, I). For Power Disaggregation: Q = 4 different states, symmetric Dirichlet prior over the transition probability vectors, additive Gaussian noise nt N(0, 0.5), and prior power consumption P m q N(15, 10). For Multiuser Detection: T = 2, 000, and L from 1 to 5.