A personalized copilot for streamlining PINN analysis and improvement
Some of the fascinating releases within the current OpenAI’s DevDay is the GPTs. Basically, GPTs are customized variations of ChatGPT that anybody can create for particular functions. The method of configuring a workable GPT entails no coding however purely by way of chatting. Consequently, for the reason that launch, a various of GPTs have been created by the group to assist customers be extra productive and create extra enjoyable in life.
As a practitioner within the area of physics-informed neural networks (PINN), I take advantage of ChatGPT (GPT-4) lots to assist me perceive advanced technical ideas, debug points encountered when implementing the mannequin, and recommend novel analysis concepts or engineering options. Regardless of being fairly helpful, I typically discover ChatGPT struggles to offer me tailor-made solutions past its normal data of PINN. Though I can tweak my prompts to include extra contextual data, it’s a slightly time-consuming apply, and might shortly deplete my endurance generally.
Now with the potential of simply customizing ChatGPT, a thought occurred to me: why not develop a personalized GPT that acts as a PINN skilled 🦸♀️, attracts data from my curated sources, and strives to reply my queries about PINN in a tailor-made manner?
So, on this weblog put up, let’s see make it a actuality! We are going to begin with introducing the method of constructing our goal GPT, offering particulars on the instruction design and provided data base. Then, we’ll undergo some demos to see greatest work together with the newly created GPT. Lastly, we’ll contact upon alternatives for future improvement.
Does this concept resonate with you? let’s get began🗺️📍🚶♀️
That is one other weblog on my sequence of physics-informed machine studying. The others embody:
Unraveling the Design Pattern of Physics-Informed Neural Networks
Discovering Differential Equations with PINN and Symbolic Regression
