I work in Erez Lieberman Aiden's group on a number of different projects, including 3D reconstruction of the genome from massive amounts of data; statistical analysis of texts to identify linguistic and cultural trends; and software engineering support of an econometrics game.
From March 2009 to September 2010, I held a postdoc at WILLOW, an INRIA/ENS project, working under the supervision of Josef Sivic, Ivan Laptev, and Andrew Zisserman. My research project involved automatically recognizing attributes in unconstrained video.
I earned the Ph.D. from the University of Washington in March 2009. I worked in a number of different areas, including accessibility, human-computer interaction, multimedia, auction theory, and data compression. My dissertation research was on activity analysis of sign language video and was a part of the ongoing Mobile ASL project, in which we compress sign language video to enable mobile phone communication for deaf people. My advisors were Richard Ladner and Eve Riskin.
We are building an object recognition system for aerial video that includes refinement of neuromorphic features, dictionary learning and feature selection, classification, and tracking. We are also developing a general framework for object recognition that is highly configurable, in order to test different kinds of features and different classification algorithms. We plan to make our architecture available to other researchers so that they may easily build and test their own systems.
The display of human actions in mass media and its implications for our society is intensively studied in sociology, marketing and health care. For example, researchers have looked at the relationship between the incidence of characters who smoke in movies and adolescent smoking; the occurrence of drinking acts in movies and the consumption of alcohol; and the impact over time of the evolution of women activities depicted by TV shows. Video analysis for these purposes currently requires hours of tedious manual labeling, rendering large-scale experiments infeasible. Automating the detection and classification of human traits and actions in video will potentially increase the quantity and diversity of experimental data.
In this project, we tackle the problem of automatically finding facial attributes in video. We focus on static attributes that do not change over the course of a track. We want to minimize supervision, and we do so by using unlabeled video to improve generalization. Please see the project page at INRIA and our Parts and Attributes workshop paper in ECCV 2010.
The goal of the Mobile ASL project is to compress sign language video so that deaf users can communicate via cell phone. While the deaf community has welcomed new technologies such as the Blackberry and other PDAs, it is cumbersome to use text messaging when compared to signing, since the speed of sign language is equivalent to speech. However, with current compression technology and low mobile phone bit rates, real-time video transmission is not possible. Our goal is to make cell phones accessible by supporting real-time compression and transmission of sign language video.
Through user studies, we have discovered that deaf users can understand signs at very low frame rates. However, this is not true of finger spelling, which consists of small, specific motion. My goal is to distinguish between signing and finger spelling so we can adjust frame and bit rate to maximize efficiency while still remaining intelligible. We have verified the intelligibility of varying the frame rate and quantified the power savings on the phone. Our machine learning techniques have been successful at distinguishing, in real-time, signing frames from listening frames.
Video-on-demand is a service in which a user interactively selects and downloads movies and other content. There are two main ways to implement video-on-demand from the server's perspective. The server could wait for requests and then react to them, perhaps by bundling requests together. However, this method requires a two-way communication channel between the server and client. Some servers, such as satellite TV, don't have two-way capabilities. The other choice is to proactively broadcast movies that are likely to be on demand. Since requests for movies follow a Zipf distribution, broadcast methods could serve the vast majority of on-demand needs. In Data Compression Conference 2006, I detailed our method for periodic broadcast of compressed variable bit rate movies.
Auction theory is the study of problems at the intersection of game theory, economics, and computer science. Christos Papadimitriou essentially introduced the area with a paper describing how the internet has spawned new problems that may be answered using techniques from economics.
In joint work with Atri Rudra, Ning Chen, Nikhil Bansal, Baruch Schieber, and Maxim Sviridenko, we looked at the problem of selling items to impatient bidders. Suppose an online music store wishes to sell unlimited copies of a song to a set of customers, and suppose each customer has a maximum price, an arrival time, and a departure time. If the store sets a price below a customer's maximum price while they are in the store, the customer will immediately buy the song. How can we set prices so as to maximize the store's revenue? We prove tight bounds on the competitiveness of deterministic online algorithms for this model.
For my qualifying project, I explored data compression of DNA. Sequitur is an elegant, linear-time online compression algorithm that also finds structure. I experimented with using Sequitur to compress DNA sequences. This work was presented at the DIMACS Working Group on the Burrows-Wheeler Transform and is published as UW CSE Technical Report 2007-05-02: Grammar-based compression of DNA Sequences (paper) and talk.
Also in the field of data compression, I extended our group's work on MultiStage, a rate control algorithm, to H.264.
Copyright notice: This material is presented to ensure timely dissemination of scholarly and technical work. Copyright and all rights therein are retained by authors or by other copyright holders. All persons copying this information are expected to adhere to the terms and constraints invoked by each author's copyright. In most cases, these works may not be reposted without the explicit permission of the copyright holder.
N. Cherniavsky, J. Chon, J. O. Wobbrock, R. E. Ladner, and E. A. Riskin. Activity Analysis Enabling Real-time Video Communication on Mobile Phones for Deaf Users. In UIST '09: Proceedings of the ACM Symposium on User Interface Software and Technology, Oct 2009.
N. Cherniavsky, R. E. Ladner, and E. A. Riskin. Activity Detection in Conversational Sign Language Video for Mobile Telecommunication . In IEEE Int'l Conference on Automatic Face and Gesture Recognition, Sept 2008.
N. Cherniavsky, A. C. Cavender, R. E. Ladner, and E. A. Riskin. Variable Frame Rate for Low Power Mobile Sign Language Communication. In ASSETS '07: Proceedings of the Ninth International ACM SIGACCESS Conference on Computers and Accessibility, Oct 2007.
N. Cherniavsky and R. E. Ladner. Practical Low Delay Broadcast of Compressed Variable Bit Rate Movies. In Data Compression Conference (DCC), Mar 2006.
N. Bansal, N. Chen, N. Cherniavsky, A. Rudra, B. Scheiber and M. Sviridenko. Dynamic Pricing for Impatient Bidders. In Symposium on Discrete Algorithms (SODA), Jan 2007.
N. Bansal, N. Chen, N. Cherniavsky, A. Rudra, B. Scheiber and M. Sviridenko. Dynamic Pricing for Impatient Bidders. ACM Transactions on Algorithms, 6 (2), March 2010.
N. Cherniavsky, G. Shavit, M. F. Ringenburg, R. E. Ladner, and E. A. Riskin. MultiStage: A MINMAX Bit Allocation Algorithm for Video Coders. IEEE Transactions on Circuits and Systems for Video Technology, 17 (1), 59-67, January 2007.
Workshop papers, invited talks, technical reports
N. Cherniavsky, I. Laptev, J. Sivic, and A. Zisserman. Semi-supervised learning of facial attributes in video. In First International Workshop on Parts and Attributes, in conjunction with ECCV 2010, September 2010.
A. Cavender, N. Cherniavsky, J. Chon, R. Ladner, E. Riskin, R. Vanam, and J. Wobbrock. MobileASL: Overcoming the technical challenges of mobile video conversation in sign language. In LREC 2010 4th Workshop on the Representation and Processing of Sign Languages, May 2010.
J. Chon, N. Cherniavsky, E. A. Riskin, and R. E. Ladner. Enabling Access through Real-Time Sign Language Communication over Cell Phones. In Asilomar Conference on Signals, Systems, and Computers 2009, Nov 2009. (Invited talk)
N. Cherniavsky. Activity Analysis of Sign Language Video for Mobile Telecommunication. PhD thesis. University of Washington, March 2009.
N. Cherniavsky and R. E. Ladner. Grammar-based compression of DNA Sequences. UW CSE Technical Report (TR2007-05-02), presented at the DIMACS Working Group on the Burrows-Wheeler Transform, August 2004.
I have taught one class as the sole instructor and served as a teaching assistant for five others. Additionally, I volunteered to tutor undergraduates throughout graduate school. I also volunteered for Making Connections, a program that aims to help disadvantaged high school women by matching them with mentors.
Data Structures (CSE 326)
I served as the sole instructor for a 28 student class consisting of computer science majors, supervising one teaching assistant. (See course website above.) I taught three weekly lectures, combining previous material with original work; held office hours; prepared a new project, including a new code base with visualization; wrote and graded exams; answered student questions via a newsgroup and private email; and maintained the course website with up-to-date lecture materials and assignments.
Introduction to Computer Science (CSE 142)
I served as one of eleven teaching assistants for the intro computer science course. The size of the class was about 300 and included about half non-majors. I taught two different hour-long weekly sessions, each containing about 30 students. The lecture material was provided by the instructor, but I had some leeway in how to teach it. I chose to use examples, asking the students to help me solve the problem. I also had them work in small groups on pieces of the problem. My duties also included holding office hours for individual help and grading assignments and exams.
Discrete Structures (CSE 321)
I served as one of two teaching assistants for Discrete Structures. I taught a weekly hour-long session attended by around 30 students. I had great leeway to teach what I thought was interesting and relevant. In the student comments, I got high marks for coordinating lecture and section and for the usefulness of section content. My duties also included holding office hours for individual help and grading assignments and exams.
Data Compression (CSE 490g)
I served as the sole teaching assistant for a special topics course on data compression. My duties included website maintenance, providing extra help in office hours, grading assignments and exams, writing solutions to assignments, and setting up and maintaining project code.
Graduate Algorithms (CSE 521 and CSE P521)
I served as one of two teaching assistants for the graduate level algorithms class. The class is quite challenging and one of my hardest jobs was to write good solutions for the homework problems. My duties also included website maintenance, providing extra help in office hours, and grading assignments and exams.
I also served as the sole teaching assistant for the professional master's program (PMP) algorithms class. The needs of the PMP students are different from full-time students, since they are only on campus for the class. I communicated a lot with the students over email, answered questions from both the class mailing list and the message board, wrote solutions to homework problems, and held office hours before class each week. My duties also included website maintenance and grading assignments and exams.