Allen Y. Yang

Research Engineer, EECS, Berkeley.

PhD, 2006. ECE, University of Illinois at Urbana-Champaign.

A wonderful life is to comfort, but is not comfortable.

Address:	Contact:
Rm 307, Cory Hall UC Berkeley Berkeley, CA 94720	Email: yang AT eecs Office: 510-643-5798 Fax: 510-643-2356

What is New!
CVPR 08 Tutorial: Computational Symmetry. [presentation]
New paper: Distributed Segmentation and Classification of Human Actions Using a Wearable Motion Sensor Network. Workshop on Human Communicative Behavior Analysis, CVPR 08. [website] [presentation]
In the news: Wired.com -- Engineers Test Highly Accurate Face Recognition. [website]
In the blog: Monday Morning Algorithm -- Compressed Sensing Meets Machine Learning. [website]
To appear in PAMI: Robust Face Recognition via Sparse Representation. [website]

Research Overview:

I am currently a postdoctoral researcher in Professor Shankar Sastry's Heterogeneous Sensor Network (HSN) group. In 2006, I finished my PhD in ECE at the University of Illinois. My primary research has been estimation and segmentation of multivariate mixture models such as subspaces and manifolds in very high-dimensional data space, and applications in computer vision and sensor networks. My early PhD research developed an algebro-geometric framework for Generalized Principal Component Analysis (GPCA), and several robust implementations of the algorithm in the presence of large data noise and outliers.

My most recent research is focused on advancing new design of next generation high-bandwidth heterogeneous sensors and distributed pattern recognition algorithms to provide on-demand surveillance functionalities. Sensor networks traditionally involve low-bandwidth sensors and dedicated/centralized classification algorithms. With the arrival of next generation high-bandwidth wireless camera motes here at Berkeley, we are interested in designing more general-purpose recognition systems that are distributed, multitasking, and adaptive to changing environments. In the long term, we intend to introduce mobility into the sensor network via ground vehicles and unmanned air vehicles (UAVs), and provide the ability for the network to interact with its environment.

Publications

Presentations

Software

Research Interests:

Pattern Analysis: Mixture-model estimation; Distributed recognition; Compressed sensing.

Computer Vision: Motion segmentation and multiple-view geometry.

Systems: On-demand surveillance via heterogeneous sensor networks.

Current Projects:

	d-Oracle: Distributed Object Recognition via a Camera Wireless Net Harnessing the multiple-view information from a wireless camera sensor network to improve the recognition of objects or actions.
	d-WAR: Distributed Wearable Action Recognition We propose a distributed recognition method to classify human actions using a low-bandwidth wearable motion sensor network. Given a set of pre-segmented motion sequences as training examples, the algorithm simultaneously segments and classifies human actions, and it also rejects outlying actions that are not in the training set. The classification is distributedly operated on individual sensor nodes and a base station computer. Using up to eight body sensors, the algorithm achieves state-of-the-art 98.8% accuracy on a set of 12 action categories. We further demonstrate that the recognition precision only decreases gracefully using smaller subsets of sensors, which validates the robustness of the distributed framework.
	Feature Selection in Face Recognition: A Sparse Representation Perspective Formulating the problem of face recognition under the emerging theory of compressed sensing, we examine the role of feature selection/dimensionality reduction from the perspective of sparse representation. Our experiments show that if sparsity in the recognition problem is properly harnessed, the choice of features is no longer critical. What is critical is whether the number of features is sufficient and whether the sparse representation is correctly found.

Past Projects:

	Robust Face Recognition with Facial Occlusion
	Image Analysis and Segmentation via Lossy Data Compression
	Generalized Principal Component Analysis (GPCA) An algebraic framework for modeling and segmenting mixed data using a union of subspaces, a.k.a. subspace arrangements. Yet the statistical implementation of the framework is robust to data noise and outliers.
	RoboTalk A unified robot motion interface and tele-communication protocols for controlling arms, bases, and androids. Copyright (c) Honda Research, Mountain View, CA.

Teaching:

Spring, 2008: Compressed Sensing Meets Machine Learning -- A Mini Course. [lecture 1, lecture 2, lecture 3]
Spring, 2007: Generalized Principal Component Analysis -- A Mini Course. [lecture 1, lecture 2, lecture 3]
Fall, 2006: CS294-6: Theoretical, Conceptual and Experimental Vision.

Life outside Cory

Frankly, Cory Hall feels like a bunker inside out. My phone calls are constantly being cut-off. That is the Berkeleyway saying: Hey, no sweat in an earthquake! :)
I enjoy listening to musical CDs in my apartment from a pair of Paradigm Reference Studio 20 speakers biwired to an NAD C350 amp.
Sometimes in a weekend, I will drive to St. Helena and Rutherford in Napa for a wine tasting.

Links:

		NEC Research Index			SIAM Journals Online

(Last Modified on March. 10, 2008)