In our modern world, a very popular topic seems to be AI. And there is a reason for it! Just imagine automating any existing process or making tasks that take weeks to be completed in just seconds! AI is a very exciting and interesting concept to understand, and the only problem with it is the need for tons of computational power and training time (for example, Llama 2 with 7 billion parameters was trained for 184320 GPU hours!).

During my study of Calculus, I accidentally stumbled on a new idea for AI model training. And after embarking on a journey to create and develop this algorithm, it yielded some interesting results - the simple training tasks were handled around 20% faster than the existing algorithms. The new idea was published online, called OPE Prop, or Once per Epoch backpropagation, related to its unique idea of updating parameters only once per epoch.

I believe, OPE Prop will develop and grow even further in the future. Collaborating with other students and professors, I plan to enrich OPE Prop with new activation and error functions and improve the program's efficiency. OPE Prop was designed with accessibility in mind, making AI more available to everyone. The potential use cases are infinite - OPE Prop can come into play in any AI-related task!