Approximate Inference in Additive Factorial HMMs with Application to Energy Disaggregation

By J. Kolter, Tommi S. Jaakkola. International Conference on Artificial Intelligence and Statistics, 2012

678 citations recorded by the source corpus.

View Paper: https://www.semanticscholar.org/paper/3692d46d3d73666bff2701270fa2bbb21439dbc3

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Is disaggregation the holy grail of energy efficiency? The case of electricity (2013, background): “One of these studies (Kolter and Jaakkola, 2012) was able to identify a number of appliance types from hourly data by utilizing behavioral patterns, such as time of day and duration of use.”
Big IoT data mining for real-time energy disaggregation in buildings (2016, methodology): “More recent researches have explored a wide range of different machine learnings methods, using both supervised and unsupervised learning, such as sparse coding [14], clustering [19], [20] or different graphical models (e.g. Factorial Hidden Markov models (FHMM) [13], Factorial Hidden Semi-Markov Mo”
The ECO data set and the performance of non-intrusive load monitoring algorithms (2014, methodology): “To this end, they developed AFAMAP, an approximate inference technique for factorial HMMs [22].”
Tackling Climate Change with Machine Learning (2019, methodology): “[278] J Zico Kolter and Tommi Jaakkola.”
A Systematic Review of Hidden Markov Models and Their Applications (2020, background): “Kolter and Jaakkola [102] worked on approximate inference problem in additive FHMM. Table 4 represents the major research findings of FHMM and its applications.”
AMPds: A public dataset for load disaggregation and eco-feedback research (2013, background): “Gaussian) [5], [6], [8] and generally ignores the idea of discrete probability.”
Non-Intrusive Load Monitoring Using Prior Models of General Appliance Types (2012, methodology): “Second, we evaluate the joint probability of all sequences in our model x, y and z using the product of Equations 1, 2, 4 and 5: p(x,y, z|θ) = p(z1|π) T∏ t=2 p(zt|zt−1,A) T∏ t=1 P (wzt ≤ xt|zt,φ) ∏ t∈S p(yt|zt, zt−1,φ) (8) This is similar to the Viterbi algorithm’s joint probability evaluation in a ”
Non-Intrusive Load Monitoring Approaches for Disaggregated Energy Sensing: A Survey (2012, methodology): “The non-power features such as duration and time of appliance usage along with their real power consumption are used to model device specific HMMs.”
BLUED : A Fully Labeled Public Dataset for Event-Based Non-Intrusive Load Monitoring Research (2012, methodology): “The non event-based approach attempts to directly separate the sources that compose the overall signal (e.g., total power consumption) by using techniques such as latent variable models [7], blind source separation [5] or time-series motif mining [10].”
Inference Networks for Sequential Monte Carlo in Graphical Models (2016, methodology): “The additive model we consider is inspired by the model studied in Kolter and Jaakkola (2012) for disaggregation of household energy usage; effective inference in this model is a subject of continued research.”
Sequence-to-point learning with neural networks for nonintrusive load monitoring (2016, methodology): “…in the literature these signatures were the key constraints added to the models, but it had to resort to hand-crafted feature selection procedures [ Kolter and Jaakkola, 2012; Parson et al. , 2012; Pattem, 2012; Zhao et al. , 2015; Zhong et al. , 2014; Zhong et al. , 2015; Batra et al. , 2016;…”
NILMTK: an open source toolkit for non-intrusive load monitoring (2014, background): “Kolter and Jaakkola later proposed an extension to the benchmark algorithm [18], however the extension was only evaluated using features extracted from 15 kHz data from a single house from the data set, and therefore the performance results are not directly comparable.”
Exploiting HMM Sparsity to Perform Online Real-Time Nonintrusive Load Monitoring (2016, methodology): “Kolter et al. [10] used a combination of two FHMM variants (additive FHMM and difference FHMM) using high-frequency sampling.”