|
9:20 | Applications of Optical Flow to Image Stylization by Mihai Parparita, senior [Joint work with Szymon Rusinkiewicz]
A system for the generation of stylized video from real-world captured
2D material will be presented. A set of passes is used to extract
interesting features such as edges and large color areas. To resolve
previously encountered issues with frame-to-frame coherence, an
approach using optical flow is chosen. The end result is a stylized
drawing of configurable style that does not exhibit the traditional
"jitter" seen in systems that are not aware of image sequences.
|
9:40 | Planet Lab by Akihiro Nakao, graduate student |
10:05 | Capturing Pen Drawings by Frank Battaglia, Wilkie Kiefer and Jason Yau, seniors |
10:30 | Break |
10:45 | Recognizing Handwritten Chinese by Jing Ge, senior
This paper describes the development and implementation of a new
pen-based approach to Chinese character input on a personal
computer. As opposed to the standard method of keyboard input, an
electronic tablet and pen is used to draw the Chinese character on a
digital canvas. We explain how we process the collection of strokes
generated by the pen and tablet, and how we use this data to predict
the character represented by the strokes. Using 10-fold cross
validation to analyze the character recognition engine, we achieve an
accuracy rate of 74.9% on the top match, 86.4% on the top 2 matches
and 90.1% on the top 3 matches.
|
11:05 | Exploration of Connectivity and Its Mechanism in Transcriptional Regulatory Networks by Jeffrey Lange, senior
Through computational clustering methods, genes in various organisms
have traditionally been organized into groups exhibiting similar
activity under similar conditions. By probing the biological basis
for such gene activity and by exploiting the architecture of gene
networks through the use of network search algorithms, a more
biologically realistic picture of gene networks begins to emerge.
|
11:25 | The Smallest Grammar- A Logarithmic Approximation Factor
by Manoj M. Prabhakaran, graduate student [Joint work with Moses Charikar, Ding Liu and Amit Sahai]
It is often useful to represent a string by a context-free grammar which
produces the singleton language containing just that string. We
investigate the problem of finding the smallest such grammar for a given
string.
This is a problem of significant theoretical and practical interest.
Recently much work has been done on using the grammar for string
compression, and competent schemes have been developed. An algorithm
which produces a small grammar from a string has direct application
to such schemes. The size of the smallest grammar for a string is of
theoretical interest as an efficiently computable complexity measure
of the string, in lieu of Kolmogorov complexity. The grammar
representation has also been suggested as a heuristic for uncovering
hierarchical structures in the string.
We give the first efficient algorithm with a logarithmic approximation
factor. Our algorithm is based on the Lempel-Ziv compression algorithm
LZ77, and a novel balancing procedure for grammars.
The smallest grammar is known to be hard to approximate within a certain
constant factor, and an approximation factor of $o(\log(n)/\log\log n)$
would imply progress on an outstanding problem in a well-studied
area. Previously, the best proven approximation ratio was
$O(n^{1/2})$. Our main result is an exponential improvement of this
ratio; we give an $O(\log \frac{n}{g^*})$ approximation algorithm,
where $g^*$ is the size of the smallest grammar.
Other results we have show that the complexity of a string defined under
certain other natural models can also be well-approximated: we give an
$O(\log^2 n)$ approximation for the smallest non-deterministic finite
automaton (Mealy machine) with advice producing the string; we show the
equivalence of the grammar model to the ``advice-grammar'' model which is
a natural and well-motivated model of defining the complexity of a
string; also we show how our algorithm for approximating the smallest
grammar can be used to approximate the smallest ``edit-grammar'', which
is a more powerful model.
|
11:50 | Liberty Simulation Environment by Manish Vachharajani, graduate student
Since high-level design is critical for system performance, designers
use high-level simulation models to guide their design decisions.
Unfortunately, modeling times permit only a few alternatives to be
explored before committing to a high-level design. However, a system
that allows models to be created from easy to use and easy to build
reusable components can dramatically increase the number of designs
explored.
This talk will describe the Liberty Simulation Environment (LSE). LSE
is a high-level hardware modeling tool designed to allow and encourage
extensive component reuse among hardware models. Experience with LSE
has shown great promise for the tool. A model of an out-of-order IA64
processor was single handedly constructed in a couple months.
Furthermore, with only modest optimization techniques, LSE yields
performance similar to models built with non-reusable components in
other structural modeling systems.
|
|
2:15 | Assigned Rooms for Affiliate/Student Office Sessions after the lunch/talk session
Please look for orange balloons for the affiliate office meeting locations to personally meet our affiliates for Q&A + submitting resumes
- Bloomberg, Room 205 - Drs. Jim Driscoll and Marcos Caro
- Hewlett Packard, Room 324 - Dr. Bill Horne
- Lucent, Room 401 - Drs. Howard Trickey and Peter Patel-Schneider
- Microsoft, Tea Room, after the lunch/talk session
- Telcordia, Room 301 until 2:45 - Dr. Josephine Micallef
Following affiliates will be here also
- Dr. Hal Stern from Sun Microsystems,
- Dr. Robert Fish from Panasonic
|