The current algorithm used for generative AI is often very computationally expensive. Processes such as simulated annealing and gradient descent, while mathematically very sound often stress the hardware to it's breaking point due to how many calculations are needed to operate it. This current method of generative AI is often software heavy, leading to this incredibly computationally heavy algorithm.
AI currently relies on an iterative process, as brute forcing the minima calculations for 100s of thousands of inputs would be incredibly computationally expensive. However, the current process also takes quite a lot of time, Thus, why not make it rely on hardware instead? A small unit, one that can graph out any function required, and scan through it to find the local minima. Have 10s of thousands of those, and then have them all run simultaneously to run the calculations for the cost minima of the function.
Although music is my number one passion when it comes to what I will pursue in college, I also hope to major in electrical engineering alongside this. Because of that, I thought this project idea, a hardware component responsible in aiding the calcuations in generative AI would be a good idea to endeavour in. My strongest skillset is math, and therefore if we decide to go with this project, my primary focus will be to aid in the calculations.