Last week, Science Buddies joined with Symantec, a sponsor of the 2010 Intel International Science and Engineering Fair (ISEF), to evaluate projects in the area of Computer Science and to name winners of the 2010 Symantec Science Buddies Special Award in Computer Science.
This special award reflects the high-quality computer science projects that are being conducted by top students around the world.
"Symantec was honored to be a Special Awards Organization recognizing outstanding computer science research at this year's Intel ISEF competition," says Zulfikar Ramzan, Technical Director at Symantec Corporation. "The quality of this year's projects in the area of Computer Science was exceptional, making it challenging to identify the winners. As one of Symantec's Special Award judges, I would like to congratulate the top five winners for their dedication, hard work and talent."
The following projects were selected as winners of the 2010 Symantec Science Buddies Special Award in Computer Science:
First Award, $1,500
Automatic Parallelization through Dynamic Analysis
Kevin Michael Ellis
The Catlin Gabel School
Abstract: Parallel programming is necessary both for high-performance computing and for continued performance increases in consumer hardware. To advance parallel programming, we develop an automatically parallelizing tool called Dyn. Dyn uses novel dynamic analyses to perform data dependency analysis, data privatization, control flow analysis, and profile the program. The data from these analyses is used to generate parallel C code from a serial C program and is capable of targeting a variety of implementations of parallelism, currently multicore via OpenMP, GPU via CUDA, and clusters thereof. Dyn also uses its own new software distributed memory system which, in concert with profiling data, will automatically tune itself to the cluster in question. We develop a set of benchmarks which would be difficult to automatically parallelize using conventional static analysis, yet we show to be easily automatically parallelizable using our dynamic analysis. We also test Dyn against scientific computing libraries and applications, achieving speedups comparable to, and occasionally exceeding, those obtained by manual parallelization. We also develop a formal system for describing dynamic analysis and parallel computing known as the Calculus of Parallel States. We prove semantics preservation with respect to parallelization of terms without data dependencies. Our final result is a dynamic-analysis based method of automatic parallelization and a rigorous mathematical theory to support it.
Second Award $1,000
Novel Computer Controlling Wireless Device for Handicapped People
Ananda College, Colombo - 10
Western, Sri Lanka
Abstract: Physically disabled people lose the ability of experiencing benefits of the modern technology. Preliminary research was carried out, in order to analyze characteristics such as simplicity, reliability, customizability and affordability of ICT-based products available in the market, specifically designed for handicaps. As a result, I figured out that their qualities are not adequate enough to satisfy requirements of such users. Therefore, invention of a computer controlling tool with all the above qualities was considered as a necessity. The developed product is an interplay of hardware and software, which controls an entire computer system, depending only on 4 input commands. Its driver software contains all the basic features a user expects from a PC. The method of providing user inputs is totally adjustable depending on the user's requirements. Also, it is extremely simple, customizable and affordable, so that any kind of a handicap can use and afford one. This product is also responsible from the environmental point-of-view. Combination of a number of hardware and software based special features enables the invention to stand as an environmentally friendly "green product." In conclusion, the developed product is outstanding under a number of sectors such as functionality, economy, Eco-friendliness and simplicity. Therefore, it is ideal to be used not only by handicaps, but also by ordinary PC users; although its design is particularly focused on the former party.
The Classification and Recognition of Emotions in Prerecorded Speech
Akash Krishnan and Matthew Fernandez
Oregon Episcopal School
Using Matlab and a German emotional speech database with 534 files and seven emotions (anxiety, disgust, happiness, boredom, neutral, sadness, and anger), we developed, trained, and tested a classification engine to determine emotions from an input signal. Emotion recognition has applications in security, gaming, user-computer interactions, lie-detection, and enhancing synthesized speech. After our speech isolation algorithm and normalization was applied, 57 features were extracted, consisting of the minimum, mean, and maximum values of fundamental frequency, first three formant frequencies, log energy, average magnitude difference, 13 Mel-frequency cepstral coefficients (MFCC), and its first and second derivatives. The MFCC
data, resorted from minimum to maximum, resembled a tangent function, so we developed a program to determine the optimal values of a and b in the tangent equation: f(x)=a*tan((pi/b)(x-500)). Clusters of the first 18 features were grouped and, in conjunction with a weighting system, were used to train and classify features of every emotion. In addition, an MFCC input feature matrix was compared against each emotion's MFCC feature matrix with another weighting system that gives importance to dissimilarity among emotions. Overall, our program was 77% accurate, only 3% worse than an average person who identifies emotions with 80% accuracy. Anxiety was 99% accurate, sadness had zero correlation with anger, and with neutral removed from the results our accuracy increased to 84%, implying that neutral is in the middle of emotional spectrum. Future work will involve comparing the results of human subjects to our program's results, and training our program with new speech databases.
To explore science and engineering projects in this area, check these Science Buddies project ideas:
Third Award $750
A Parallel Computational Framework for Solving Quadratic Assignment Problems Exactly
Michael Christopher Yurko
Detroit Catholic Central High School
Abstract: The Quadratic Assignment Problem (QAP) is a combinatorial optimization problem used to model a number of different engineering applications. Originally it was the problem of optimally placing electronic components to minimize wire length. However, essentially the same problem occurs in backboard and circuit wiring and testing, facility layout, urban planning, ergonomics, scheduling, and generally in location problems. Additionally, it is one of the most computationally difficult combinatorial problems known. For example, a recent solution of a problem of size thirty using a state-of-the-art solver took the equivalent of 7 single-CPU years. The goal of this project was to create an open and easily extendable parallel framework for solving the QAP exactly. This framework has shown good scalability to many cores. It experimentally has over 95% efficiency when run on a system with 24 cores. This framework is designed to be modular to allow for the addition of different initial heuristics and lower bounds. The framework was tested with many heuristics including a new gradient projection heuristic and a simulated annealing procedure. The framework was also tested with different lower bounds including the Gilmore-Lawler bound (GLB). The GLB was computed using a custom implementation of the Kuhn-Munkres algorithm to solve the associated linear assignment problem (LAP). The core backtracking solver uses the unique approach of only considering partial solutions rather than recursively solving sub-problems. This allows for more efficient parallelization as inter-process communication is kept to a minimum.
To explore science and engineering projects in this area, check these Science Buddies project ideas:
Does Practice Make Perfect? The Role of Training Neural Networks
Brittany Michelle Wenger
The Out-Of-Door Academy
Abstract: Does practice really make perfect when applied to neural networks? Neural Networks operate by selecting the most successful option based on prior experiences in a certain situation. This project explores the difference in learning levels between a soccer neural network trained in games versus a neural network that was trained via scenarios, which emulate a practice type atmosphere, to determine which training mechanism is most beneficial.
This project was developed from the existing soccer neural network. The program was enhanced to allow for the implementation of scenario based training. Ten scenarios were defined to optimize the training experience. Twenty trials of scenario trained teams were compared to twenty trials of game trained teams. To assure the results were statistically significant; a t-Test was conducted comparing both winning percentage and goal differential.
Out of forty trials, eleven trials achieved nearly optimal learning capacity - eight trained via scenarios and three trained through games. The average goal differential and winning percentage is better for the scenario trained teams and the results proved to be statistically significant at a 95% confidence level. Scenario based training is more effective than game or simulation based training.
The results confirm that the hypothesis was correct and that convention wisdom is effective. Especially for those creating a medical neural network, I would recommend following the idiom "Practice Makes Perfect" when running simulations of the neural model because you can never be too careful.
Project Ideas and the Advanced Guide
While we have noted a few science project ideas that would allow students to explore topics in the general area of some of these award-winning projects, these projects are not intended to offer Intel ISEF-level research and exploration. These projects can, however, offer an introduction to a new area of research for a student and may offer building blocks upon which advanced projects can be envisioned and conceived.
Students working on the kinds of advanced and highly specialized projects that appear at the Intel ISEF benefit from resources available in the Science Buddies Advanced Project Guide. For example, for students already thinking about next year's top science competitions, reviewing the roundtable discussion Finding an Idea for an Advanced Science Fair Project can help point students in the right direction.
Congratulations to these special award winners and to all students who competed in this year's Intel ISEF!
UPDATE: We've published overviews of some of the winning projects on the Science Buddies website.
Symantec is the sponsor of the Computer Science interest area in the project directory of over 1,100 science project ideas on the Science Buddies website.