- S
- J
Your contributions will go to assist us (The Artificial Intelligence based Mathematical Analysis for Theorem Evaluation and Construction Team): Kyle Nagasawa, Zahir Ahmed, and Satyam Sharma, three 10th graders attending Del Norte High School in San Diego, California.We are participating in the upcoming Google Science Fair, where teams of ambitious high school students compete to innovate new technology of the future. Winners are decided by a panel of judges, evaluating on both creativity and impact on the world. For instance, Olivia Hallisey won the 2015 Google Science Fair. She managed to design and create an efficient, yet cheap assay for ebola virus.
Our goal is to design a machine with the ability to automatically produce theorems following a rigorous set of axioms and inference rules. Furthermore, we are designing a neural network to act as a trash collector: it will learn which theorems are useful. Finally, we must design a clean UI to clearly present each theorem through abbreviations of strings of base-level symbols. Each condition provides a host of problems: from prime number generation and storage of theorems to variable name generation and training data. Currently, we have a working system able to successfully generate true theorems, but each generated theorem takes too much time and memory to be truly effective. Furthermore, every added layer to this project introduces a higher workload to the computer running it, and in total the programmed components would not even run on our current computers..
Thus, in order to successfully manage this load of information and processing, we require a strong, efficient computer. This means that the device will be costly. After picking parts that matched our specifications, we estimated the bill to be approximately $3000. A simplified cost plan can be seen below:
{https://docs.google.com/document/d/1Xs-3C0DsWfQ_Ji0ZjI7pTHR3jsw2RusC9a1xFLuaHQg/edit?usp=sharing}
As shown above, our most costly component is the Nvidia Titan Xp. While it serves as a graphics processing unit, it also functions for parallel processing. In short, this allows our program to run on numerous cores (“miniature computers”), increasing the efficiency tenfold. In addition, our second most expensive items is AMD Ryzen 7 2700x and x470 motherboard. The Ryzen 7 2700x gives us the most cores for the cheapest price, which can further expedite the process. Meanwhile, this motherboard is compatible with all of our required parts, which is necessary for our computer to function.
Any excess funds will be used to increase the memory of the computer as well as go towards purchasing a server, further speeding up the theorem generation and machine learning. Our eventual goal is to broadcast the programs to other willing computers in order to have them do chunks of work, which would optimize the system to the greatest extent possible.
Thank you for aiding us in our reimagining of mathematical investigation!
Our goal is to design a machine with the ability to automatically produce theorems following a rigorous set of axioms and inference rules. Furthermore, we are designing a neural network to act as a trash collector: it will learn which theorems are useful. Finally, we must design a clean UI to clearly present each theorem through abbreviations of strings of base-level symbols. Each condition provides a host of problems: from prime number generation and storage of theorems to variable name generation and training data. Currently, we have a working system able to successfully generate true theorems, but each generated theorem takes too much time and memory to be truly effective. Furthermore, every added layer to this project introduces a higher workload to the computer running it, and in total the programmed components would not even run on our current computers..
Thus, in order to successfully manage this load of information and processing, we require a strong, efficient computer. This means that the device will be costly. After picking parts that matched our specifications, we estimated the bill to be approximately $3000. A simplified cost plan can be seen below:
{https://docs.google.com/document/d/1Xs-3C0DsWfQ_Ji0ZjI7pTHR3jsw2RusC9a1xFLuaHQg/edit?usp=sharing}
As shown above, our most costly component is the Nvidia Titan Xp. While it serves as a graphics processing unit, it also functions for parallel processing. In short, this allows our program to run on numerous cores (“miniature computers”), increasing the efficiency tenfold. In addition, our second most expensive items is AMD Ryzen 7 2700x and x470 motherboard. The Ryzen 7 2700x gives us the most cores for the cheapest price, which can further expedite the process. Meanwhile, this motherboard is compatible with all of our required parts, which is necessary for our computer to function.
Any excess funds will be used to increase the memory of the computer as well as go towards purchasing a server, further speeding up the theorem generation and machine learning. Our eventual goal is to broadcast the programs to other willing computers in order to have them do chunks of work, which would optimize the system to the greatest extent possible.
Thank you for aiding us in our reimagining of mathematical investigation!