Contextually Supervised Source Separation with Application to Energy Disaggregation
Authors: Matt Wytock, J. Kolter
AAAI 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | On smaller datasets which include labels, we demonstrate that contextual supervision improves significantly over a reasonable baseline and existing unsupervised methods for source separation. Finally, we analyze the case of 2 loss theoretically and show that recovery of the signal components depends only on cross-correlation between features for different signals, not on correlations between features for the same signal. In this section we evaluate contextual supervision for energy disaggregation on one synthetic dataset and two real datasets. |
| Researcher Affiliation | Academia | Matt Wytock and J. Zico Kolter Machine Learning Department Carnegie Mellon University Pittsburgh, Pennsylvania 15213 |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide concrete access to source code for the methodology described. |
| Open Datasets | Yes | On real data, we begin with a dataset from Pecan Street, Inc. (http://www.pecanstreet.org/) that is relatively small (less than 100 homes), but comes with labeled data allowing us to validate our unsupervised algorithm quantitatively. Finally, we apply the same methodology to disaggregate large-scale smart meter data from Pacific Gas and Electric (PG&E) consisting of over 4000 homes and compare the results of our contextually supervised model to aggregate statistics from survey data. |
| Dataset Splits | No | The paper mentions segmenting synthetic data ( |
| Hardware Specification | No | The paper does not provide specific hardware details used for running its experiments. |
| Software Dependencies | No | The paper does not provide specific ancillary software details with version numbers. |
| Experiment Setup | Yes | In all experiments, we tune the model using hyperparameters that weigh the terms in the optimization objective; in the case of energy disaggregation, the model including hyperparameters and β is shown in Table 3. We set these hyperparameters using a priori knowledge about the relative frequency of each signal over the entire dataset. ... The specification of our energy disaggregation model is given in Table 3 we capture the non-linear dependence on temperature with radial-basis functions (RBFs), include a Base category which models energy used as a function of time of day, and featureless Appliance category representing large spikes of energy which do not correspond to any available context. For simplicity, we penalize each category s deviations from the model using 1 loss; but for heating and cooling we first multiply by a smoothing matrix Sn (1 s on the diagonal and n super diagonals) capturing the thermal mass inherent in heating and cooling: we expect energy usage to correlate with temperature over a window of time, not immediately. We use gi(yi) and the difference operator to encode our intuition of how energy consumption in each category evolves over time. |