Discriminative Recurring Signal Detection and Localization Zeyu You, Raviv Raich*, Xiaoli Z. Fern, and Jinsub Kim School of EECS, Oregon State University,

Slides:



Advertisements
Similar presentations
ETHEM ALPAYDIN © The MIT Press, Lecture Slides for.
Advertisements

Advanced topics.
Optimization Tutorial
Adaption Adjusting Model’s parameters for a new speaker. Adjusting all parameters need a huge amount of data (impractical). The solution is to cluster.
Supervised Learning Recap
Complex Feature Recognition: A Bayesian Approach for Learning to Recognize Objects by Paul A. Viola Presented By: Emrah Ceyhan Divin Proothi Sherwin Shaidee.
On-Line Probabilistic Classification with Particle Filters Pedro Højen-Sørensen, Nando de Freitas, and Torgen Fog, Proceedings of the IEEE International.
Computer vision: models, learning and inference
Rob Fergus Courant Institute of Mathematical Sciences New York University A Variational Approach to Blind Image Deconvolution.
Hidden Markov Model based 2D Shape Classification Ninad Thakoor 1 and Jean Gao 2 1 Electrical Engineering, University of Texas at Arlington, TX-76013,
Lecture 17: Supervised Learning Recap Machine Learning April 6, 2010.
Lecture 5: Learning models using EM
Effective Gaussian mixture learning for video background subtraction Dar-Shyang Lee, Member, IEEE.
Selective Sampling on Probabilistic Labels Peng Peng, Raymond Chi-Wing Wong CSE, HKUST 1.
Extracting Places and Activities from GPS Traces Using Hierarchical Conditional Random Fields Yong-Joong Kim Dept. of Computer Science Yonsei.
Overview G. Jogesh Babu. Probability theory Probability is all about flip of a coin Conditional probability & Bayes theorem (Bayesian analysis) Expectation,
Tracking Pedestrians Using Local Spatio- Temporal Motion Patterns in Extremely Crowded Scenes Louis Kratz and Ko Nishino IEEE TRANSACTIONS ON PATTERN ANALYSIS.
Alignment and classification of time series gene expression in clinical studies Tien-ho Lin, Naftali Kaminski and Ziv Bar-Joseph.
SPECTRO-TEMPORAL POST-SMOOTHING IN NMF BASED SINGLE-CHANNEL SOURCE SEPARATION Emad M. Grais and Hakan Erdogan Sabanci University, Istanbul, Turkey  Single-channel.
A General Framework for Tracking Multiple People from a Moving Camera
Crowdsourcing with Multi- Dimensional Trust Xiangyang Liu 1, He He 2, and John S. Baras 1 1 Institute for Systems Research and Department of Electrical.
TEMPLATE DESIGN © Zhiyao Duan 1,2, Lie Lu 1, and Changshui Zhang 2 1. Microsoft Research Asia (MSRA), Beijing, China.2.
Structure Discovery of Pop Music Using HHMM E6820 Project Jessie Hsu 03/09/05.
CS 782 – Machine Learning Lecture 4 Linear Models for Classification  Probabilistic generative models  Probabilistic discriminative models.
Empirical Research Methods in Computer Science Lecture 7 November 30, 2005 Noah Smith.
CS Statistical Machine learning Lecture 10 Yuan (Alan) Qi Purdue CS Sept
Jun-Won Suh Intelligent Electronic Systems Human and Systems Engineering Department of Electrical and Computer Engineering Speaker Verification System.
The Group Lasso for Logistic Regression Lukas Meier, Sara van de Geer and Peter Bühlmann Presenter: Lu Ren ECE Dept., Duke University Sept. 19, 2008.
14 October, 2010LRI Seminar 2010 (Univ. Paris-Sud)1 Statistical performance analysis by loopy belief propagation in probabilistic image processing Kazuyuki.
A Passive Approach to Sensor Network Localization Rahul Biswas and Sebastian Thrun International Conference on Intelligent Robots and Systems 2004 Presented.
Sparse Bayesian Learning for Efficient Visual Tracking O. Williams, A. Blake & R. Cipolloa PAMI, Aug Presented by Yuting Qi Machine Learning Reading.
Sequential Monte-Carlo Method -Introduction, implementation and application Fan, Xin
6. Population Codes Presented by Rhee, Je-Keun © 2008, SNU Biointelligence Lab,
Javad Azimi, Ali Jalali, Xiaoli Fern Oregon State University University of Texas at Austin In NIPS 2011, Workshop in Bayesian optimization, experimental.
Discriminative Training and Machine Learning Approaches Machine Learning Lab, Dept. of CSIE, NCKU Chih-Pin Liao.
Introduction to Gaussian Process CS 478 – INTRODUCTION 1 CS 778 Chris Tensmeyer.
SUPERVISED AND UNSUPERVISED LEARNING Presentation by Ege Saygıner CENG 784.
An Automatic Method for Selecting the Parameter of the RBF Kernel Function to Support Vector Machines Cheng-Hsuan Li 1,2 Chin-Teng.
A Study on Speaker Adaptation of Continuous Density HMM Parameters By Chin-Hui Lee, Chih-Heng Lin, and Biing-Hwang Juang Presented by: 陳亮宇 1990 ICASSP/IEEE.
ETHEM ALPAYDIN © The MIT Press, Lecture Slides for.
Overview G. Jogesh Babu. R Programming environment Introduction to R programming language R is an integrated suite of software facilities for data manipulation,
Naive Bayes (Generative Classifier) vs. Logistic Regression (Discriminative Classifier) Minkyoung Kim.
Introduction to Machine Learning Nir Ailon Lecture 11: Probabilistic Models.
A NONPARAMETRIC BAYESIAN APPROACH FOR
Neural networks and support vector machines
Who am I? Work in Probabilistic Machine Learning Like to teach 
Learning Deep Generative Models by Ruslan Salakhutdinov
DEEP LEARNING BOOK CHAPTER to CHAPTER 6
Introduction to Audio Watermarking Schemes N. Lazic and P
Zeyu You, Raviv Raich, Yonghong Huang (presenter)
LABEL CORRECTION AND EVENT DETECTION FOR ELECTRICITY DISAGGREGATION
Classification: Logistic Regression
Outlier Processing via L1-Principal Subspaces
Chapter 2 Simple Comparative Experiments
Cold-Start Heterogeneous-Device Wireless Localization
Intelligent Information System Lab
Machine Learning Basics
Probabilistic Models for Linear Regression
Combining Species Occupancy Models and Boosted Regression Trees
Akio Utsugi National Institute of Bioscience and Human-technology,
Ying shen Sse, tongji university Sep. 2016
Biointelligence Laboratory, Seoul National University
CSCE833 Machine Learning Lecture 9 Linear Discriminant Analysis
Finding Periodic Discrete Events in Noisy Streams
Recap: Naïve Bayes classifier
Linear Discrimination
Deep neural networks for spike sorting: exploring options
Speech Enhancement Based on Nonparametric Factor Analysis
Maximum Likelihood Estimation (MLE)
Outlines Introduction & Objectives Methodology & Workflow
Presentation transcript:

Discriminative Recurring Signal Detection and Localization Zeyu You, Raviv Raich*, Xiaoli Z. Fern, and Jinsub Kim School of EECS, Oregon State University, Corvallis, OR 97331-5501

Organization Introduction Related work and our contribution Motivation Problem formulation Maximum likelihood estimation(MLE) Synthetic data experimental Results Real-world data experimental Results

Introduction What is a recurring pattern? Pattern characteristics: DNA motifs Music motifs Home appliance activations Pattern characteristics: Sharing same structure Recurring in nature

Applications Motifs Air-conditioning activation signals D'haeseleer, Patrik. "How does DNA sequence motif discovery work?." Nature biotechnology 24.8 (2006): 959-961. From Fee Lab Research in http://web.mit.edu/feelab/research.html Air-conditioning activation signals From Pecan Street dataset (Source: Pecan Street Research Institute)

Related work and our contribution Previous works [1-5] focus on: discover recurrent patterns from data finding the fundamental characteristics of the signal pattern Our contribution: a novel formulation of auto-detecting recurring signal patterns a maximum likelihood estimation (MLE) solution for the problem an increased detection performance on a real-world data

Generative vs. Discriminative

Motivation for our approach Flavor in discriminative for two reasons Robust to variations with in the pattern Robust to low signal to noise ratio

Problem formulation System diagram: System description: System target: Observed data: a collection of M signals Hidden data: System target: To learn a convolutional kernel w. y(t) w(t)* x(t) x(t) Signal Labeler w LR Y

The graphical model Instance labeler Signal labeler xmt ymt Ym T M w

The probabilistic model Instance labeler (logistic regression): Signal labeler: Condition model:

The data likelihood Data likelihood: Maximum likelihood estimation (MLE): Minimizing the negative log likelihood: Independence Data distribution Difference of convex

Convex-concave procedure (CCCP) Solution with CCCP [6]: Upper bound function (linearization): Gradient descent: Prior Posterior

Synthetic data experiment Setup: Train on M=160; Test on 40; Setting kernel size to be F=10, T0=7; 10 MC runs of different initialization. Data generation: Generate a rectangular pattern; Create an empty spectrogram with F=10, T=50; Random placing the pattern with varying magnitude into one time index out of 50; Add gaussian noise.

Synthetic results Discriminative vs. generative approach: True pattern Learned kernel Data Generative Discriminative localization localization ROC

Real world experiment Setup: Data generation: Four home ps-025,029,046,051, 25 days of disaggregated, time-sampled electricity usage data from the Pecan Street dataset ({Source: Pecan Street Research Institute}) Training period 11/17/2012-11/25/2012 meter reading; Test period 11/26/2012-12/11/2012; Validating kernel size and compared with general approach with window size set to be T0=700; Data generation: Extract activations based on power ground truth; Extract negative data by random selecting the time where the power has no significant increase; Remove DC offset and Despike large spike noise by median filter;

Experiment results Tuning T0 Fridge activation and data sample Generative detection Discriminative detection

Detection accuracy Performance: Discriminative is better at localization Discriminative is more Invariant to the slight variations of activation signals Discriminative has higher AUC than generative in general AUC table for both generative and discriminative

Discussion Can we extent our model to multi-class to give more discrimination between different activation patterns? Can we speedup the algorithm by converging quicker? Can we find more applicable real-world application areas for the algorithm?

References [1]  Zeyu You, Raviv Raich, and Yonghong Huang, “An inference framework for detection of home appliance activation from voltage measurements,” in 2014 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2014, pp. 6033–6037. [2]  Alex S Park and James R Glass, “Unsupervised pat- tern discovery in speech,” IEEE Transactions on Audio, Speech, and Language Processing, vol. 16, no. 1, pp. 186–197, 2008. [3]  Aline Cabasson and Olivier Meste, “Time delay estimation: a new insight into the woody’s method,” IEEE signal processing letters, vol. 15, pp. 573–576, 2008. [4]  Yoshiki Tanaka, Kazuhisa Iwamoto, and Kuniaki Uehara, “Discovery of time-series motif from multi-dimensional data based on mdl principle,” Machine Learning, vol. 58, no. 2-3, pp. 269–300, 2005. [5]  Jessica Lin, Eamonn Keogh, Stefano Lonardi, and Pranav Patel, “Finding motifs in time series,” in Proc. of the 2nd Workshop on Temporal Data Mining, 2002, pp. 53–68. [6]  Alan L Yuille and Anand Rangarajan, “The concave- convex procedure (cccp),” Advances in neural information processing systems, vol. 2, pp. 1033–1040, 2002.

Questions? Thank You!