For example i have a set of 9 images from which i am supposed to select the ones containing bridges or traffic lights. If i dont select them all, it wont let me pass. How does it know if we’re training them by completing?
In: 2
By receiving results from thousands of users, statistical models are used to separate real from fake users by looking at the grouping of data. Fun fact: data collected from certain captcha were actually used to train algorithms to read blurry text. By giving two tasks to the user, the first verified the user to be human and the second was used to train computer to be better at reading.
One way is to apply a controlled experiment method by including a mix of known (control) and unknown (experimental) images. For example, six of the nine images might already be known to contain a bridge.
The system uses your picks on those to decide the probability that you are both [A] a human and [B] OK at picking bridges. It does that based not only on the images you picked, but also other factors such as your IP address, pick speeds, mouse movements, and whatnot.
If those conditions are met, it can assume you more or less picked correctly on the unknown three images. After that they just need to lather, rinse, and repeat the process.
Latest Answers