Alumni

JiHwan Min

Mathematica track

Class of 2016

Bio

JiHwan Min is a rising junior at Catalina Foothills High School. He enjoys studying math, physics and other STEM classes. He has been a member of his school's math and science team for competitions such as Science Olympiad and AMC 12. In his free time, he plays video games and browses the web. With his extensive knowledge of English grammar, he loves to correct others' English and argues that English is one of the most complicated languages to learn. He looks forward to furthering his knowledge of mathematics and science at the Wolfram High School Summer Camp.​​​​

Project: Create and Break CAPTCHA

No text CAPTCHAs are effective against machine learning solvers. This machine learning solver had a 70% success rate per letter, which means 0.70^(# of letters). There is a 2.8% chance of solving for 10 letters and a .67% chance for 14. Though the solver might seem ineffective, the machine learning program only had 5,000 words for data with which to train, and it was able to bring the accuracy of guessing a letter correctly from 0% to 70%. If there was more data, it might have been possible to bring the accuracy to 90% per letter, which is 0.90^10≈35% for 10 characters and 0.90^14=23% for 14 characters. The ImageIdentify function in Mathematica uses innumerable pictures from the user input on the website imageidentify.com and search engines to come up with its results, and the function does its job very well. If a machine learning program can reach a 10–20% success rate, then it can try a few times and access the website.