Differentiable programming

Differentiable programming is a computer programming approach used to make systems that can learn from data. Differentiable programming allows us to create programs that can adjust themselves to get better at a task, such as recognizing patterns in images or understanding language. These adjustments happen through a process called back-propagation, where the program looks at how it did on a task (its “performance”) and then tries to adjust the code to do better next time. This process is “differentiable” because it uses math to figure out what changes to make--math that is easy for a computer to understand.