Written by You ll find the shortcuts for Microsoft office 2003 and 2007 shortcuts in the following link.
Posts in category Uncategorized
Current wr0k, Spring Break, exams and etc.
Lately, I have been enhancing the web service that I have been working for a while. Developing web services using C#.Net is really easy and fast. I have built documentation using Doxygen for web service invokers.
Yet another project I m working on is populating data of an application and saving the results in a data warehouse. So that Pentaho will analyze the data and generate reports.
I m almost done with another project for Huge Inc, which is a Flex FTP Client.
School is in Spring break. Last Thursday I had two midterms on the same day which was a disaster. from 9.30 am to 2 pm. Certainly, the PhD program could schedule better exam dates. :-/
Writing a paper on data center network design. Interesting topic. Not easy.
A Security note on Cloud Computing
Here goes a hacker and cloud computing :
Hacker sparks an cloud instance and installs bunches of software and programs, such as email servers, dns servers, builds a system, then start spamming and attacking users. After an amount of time, hacker kills the instance, leaving almost no fingerprint behind.
I imagine Amazon and Microsoft would be keeping record of instances, IP addresses and logs but it s also difficult to do this.
Therefore, Cloud Computing need strong security policies and implementations.
Similarity measures for categorical and numerical patterns
Today, Professor Wei talked about similarity measures between matching and recognition purposes in information systems. Professor started introducing algorithmic approaches such as dynamic programming, linear programming and edit distance.
Professor Wei talked about Edit Distance algorithm, how it works and what it does and introduced Markov Edit Distance algorithm which is another approach of edit distance algorithm to process the similarities of different objects. Edit distance is an important algorithm that is being used by many researches for string and other objects searching similarities, mismatches and differences.
Kolmogorov complexity was also mentioned in the talk which is related to information theory, compression and presentation of data. For example: printing out 1 million 0’s can be done by writing them in a loop which the loop invariant would run 1 million times and print 0. The amount of information is same but it s presented with less information.
Professor Wei, also talked about use of histograms which are not any longer applicable therefore there s been a new development of local sum algorithm that is used to for swapping, splitting, merging and differing quantization schemes in case there is some noise. Professor showed us image examples which were processed and different results were produced using local sum algorithm.
Classes, WrOk and the rest in March 2010
Remaining of March 2010 surely will be pretty busy.
I have 2 midterms, Logic, Distributed Systems. I need to write a paper for the networking class, need to study for Qualifiers exam and try to make living.
I m enjoying the logic class even though it s somewhat hard to grasp for someone who never took a logic class before. Professor Fitting is great, very helpful. The class is about different types of logic, so far we ve seen, classical logic, tableaus, intiutionistic logic, Kripke Model, First Order Logic automated theorem proving, Hintikka’s lemma, Konig’s lemma and etc..
Networking class is a fun class, we are studying some distributed algorithms, some networking principles and paradigms. Professor Griffeth is great, she is very clear and helpful. So far we have studied OSI layers, TCP/IP layers, Wireless nets, routing etc. I m supposed to write a paper on scalable data center network design. Interesting topic, there are many data centers build everyday to serve business for colocating and hosting their services, servers and applications and there is an increasing demand for data centers due to the economies and advantages data centers offer. As the number of clients increases, number of servers, hardware and networking devices increases as well. To support growing of network, data centers need scalable network design and infrastructure. That s the topic of my paper. Pretty interesting and new topic.
There is also this class, research seminar which every week, there comes a professor and gives a talk about his/her research area and we write reports about them etc. One thing I realized is that almost everywhere Algorithms is used.
The rest of the time, trying to make living, how ? programming.
Developing new Metrics for computer based multitasking behavior
Firat Atagun
Research Seminar
Talk 5
Developing new Metrics for computer based multitasking behavior
In this talk Professor Raquel Benbunan-Fich, presented new metrics and approaches to investigate computer based multitasking behavior. Multitasking is a fundamental behavior in our everyday life. Understanding multitasking and efficiency is the challenge and this talked presented some approaches taken against this problem. This research is interested in finding metrics of multitasking unrelated computer tasks.
The metrics are calculated using self reported user logs by about 50 users and these logs are being parsed and analyzed to metrics are created. A program has been developed by a PhD student in Perl language and upon installing in the user’s computer program traces the windows and activities of users and records them as log files. These log files later are being parsed and processed and metrics are being generated and analyzed.
This talk was lays the foundation of human computer interaction based on 3 elements: user, task and technology. Moreover, overlapping of the tasks is also being analyzed and processed.
One of the interesting point I realized it that multitasking and task switching resembles context switching in processors. One of the important aspect of context switching is that it s an expensive operation in the sense that there needs to be several operations done in the operating system to switch between processes, in this model tasks. This brings up the question of efficiency in task switching and multitasking.
My interest in this research area would be to analyze the connection between the human – computer interaction and greedy algorithms. While using a computer we intent to have a greedy criterion that we derive more benefit then others, I m curious the relation between greedy algorithms and human-computer interaction. Moreover, if the user behavior is learned by computers we might develop intelligent agents and operating systems to serve the users better.
Cross-Document Cross-lingual Cross-Media Information Extraction
This talk given by Professor Heng Ji was quite interesting. I have been interested
in web searching for a while and what Professor Ji talked about was related to web
searching. Analyzing documents, media and data is a very important topic these days, due
to the growth of data on the internet and media. Since data growth outruns the ability to
manage it, we need scalable and reliable algorithms. There are several difficulties to
analyze the data due to the complexity of language and the environment the data resides.
Due to the complexity of parsing and analyzing the data because of the language
and grammar varieties, there have been studies on how to approach these problems.
Professor Ji talked about machine learning while analyzing the data so that pattern
recognition and other techniques would help analyze the data better.
Moreover, natural language processing is also related to data mining in the
context of looking for valuable information in the data and find relevancy and knowledge
discovery. Professor Ji, mentioned about use of n-gram, developed by google, which is
“An n-gram is a subsequence of n items from a given sequence. The items in question
can be phonemes, syllables, letters, words or base pairs according to the application.†[1]
I have great interest in this particular topic and uses of this topic in web searching
and semantic web. There are several open problems and there is need for developing new
algorithms. One of the problems to investigate is finding relevant data. Today there are
several search engines which are trying to serve users to find what they are looking for
but accuracy of search results tend to fail usually, because it s hard to understand, analyze
and return the relevant search results from the data on the web. There have been several
strings algorithms developed such as Edit Distance [2], hamming distance [3] which
are long running algorithms. We need efficient algorithms to process the large amount of
data and to reason it and find relevancy between documents.
References:
PhotoSketch: A Photo-Centric 3D Modeling Tool
PhotoSketch is a Google sketchup Plug-in for Photo-Centric 3D Modeling presented by Prof. George Wolberg. Difference between the approach by Professor Wolberg and Professor Stamos is that Professor Stamos uses very expensive, high-end digital equipment such as laser. Professor Stamos takes several very high resolution images with laser equipment and process the images using a program which takes about 18 hours to generate an image. On the other hand Professor Wolberg uses basic digital cameras that can be purchased from any electronic store, and using the PhotoSketch plug-in, 3D modeling can be done.
Another difference is that the images that are used for PhotoSketch has average resolution. Therefore 3D model that comes out doesn’t contain too much information. On the other hand, Professor Stamos equipment takes very high resolution images therefore 3D models that came out contains more information.
The applications PhotoSketch is being used is Urban/Military planning, virtual tours, realistic      sets such as movies and video games, 3D auto navigation systems, architecture and maps. Some of the features of PhotoSketch are Camera calibration, easy to use sketching interface, texture mapping, and photos to guide drawing and so on.
PhotoSketch, after taking several images of an object combines the images using algorithms that overlap the images. Finally, a 3D model of the object is being created. PhotoSketch also uses push and pull technologies to add textures to the 3D images. PhotoSketch is designed to make it easy for sketchup users to build photo textured 3D models of buildings directly from photographs. PhotoSketch simplifies the 3D modeling workflow by leveraging the sketchup push-pull interface.
My interest in this area would be to process large scale of images and create 3D models of a large scale environment. This would require distributed computing and processing the data on clusters.
3D Modeling of Large Scale Scenes
The talk that took place on 02/09/2010 by Professor Ioannis Stamos was about photorealistic 3D modeling of large scenes. The goal is to produce highly accurate photorealistic descriptions of the 3D world with a minimum of human interaction and increased computational efficiency. One of the two main devices that are used to collect data is a spot laser scanner with a maximum range of 100 meters, a scanning time of 16 minutes, and an accuracy of about 5 mm per range point. The other device is a Leica scanstation that has a spherical ï¬eld-of-view and maximum range of 300 meters.  A major bottleneck in the process of 3D scene acquisition is the automated registration of a large number of geometrically complex 3D range scans and high-resolution 2D images in a common frame of reference. The methods utilize range segmentation and feature extraction algorithms. We have also developed a context-sensitive user interface to overcome problems emerging from scene symmetry. Moreover, what I understand from this process is taking 2D images and merging them in a way that, at the end the image would be identical as the object whose picture was taken, moreover, the several images captured from different angles are being merged to give a 3D representation of the object. I believe this is being done, be merging, overlapping the images at some degree.
Finally, he presented some results of a 3D photorealistic model of CCNY Shepard Hall and Thomas Hunter building. The results look spectacular, although not 100% perfect because the range data has missing data such as sky and windows. Windows was black in the images because the laser that was used for this was just going through the window, it wasn’t reflecting back.
My interest in this area is processing the information that is captured by devices with more efficient algorithms and on distributed systems. As Professor Stamos mentioned processing an image takes about 18 hours and there are lots of work to be done to improve the running time of the program. Moreover, the program can be run on a distributed environment with efficient divide and conquer algorithms. Currently there are several distributed frameworks to process large amount of data, and perhaps this project can be adopted to run with these frameworks. Also, in terms of processing the images, the APIs being used to process the images can be researched on.
Recent Comments