To prove you’re not a robot on a website, you may encounter distorted letters to guess the spellings. This process is known as a CAPTCHA – a test to distinguish between computers and humans or challenge. It’s a security measure that involves data collection on how humans solve problems. Companies use this data to train AI algorithms to read words in various conditions. This approach could also benefit the natural sciences.
Astronomers tried to identify galaxies using volunteers and AI. They gathered data from the Hobby-Eberly Telescope Dark Energy Experiment (HETDEX). HETDEX scientists aim to map distant galaxies and measure the universe’s expansion speed to understand its early history. However, classifying billions of astronomical measurements poses a challenge with trillions of individual pixels.
A program was developed to train volunteers in classifying HETDEX data without complex terminology. This program, called Dark Energy Explorer, engaged 17,000 volunteers worldwide and classified 200,000 galaxy candidates.
Using machine learning, scientists assigned numbers to volunteer classifications and inputted data into the Stochastic Neighborhood Embedding of t-Distribution (t-SNE) algorithm. AI trained on volunteer data achieved 92% accuracy in identifying galaxies with low scores, matching human averages.
While this experiment focused on 1.2 million images, the goal is to apply AI across the larger HETDEX dataset. The AI flagged 5% of images for removal, although there’s a systematic bias due to data exclusion closer to Earth. Despite this, the collaboration among scientists, volunteers, and AI can enhance astronomical data processing.
Post views: 544
Source: sciworthy.com
