there is literally an entire paper published by Deepmind (the first pioneers in Reinforcement learning algorithms) which basically tries to prove that Graph Neural Networks are basically dynamic programmers.
I am myself working on RLHF at the moment and I can argue that dp is involved everywhere in RL due to policy optimization, monte carlo methods, etc. These types of methods are used in AI technologies as RLHF (Reinforcement learning through human feedback) and a lot of dp is involved to optimize the policies due to high amount of trajectories (Data essentially). This is why I argued that dp is better for AI compared to other types of problems.
there is literally an entire paper published by Deepmind (the first pioneers in Reinforcement learning algorithms) which basically tries to prove that Graph Neural Networks are basically dynamic programmers.
Here is the source — https://arxiv.org/pdf/2203.15544
I am myself working on RLHF at the moment and I can argue that dp is involved everywhere in RL due to policy optimization, monte carlo methods, etc. These types of methods are used in AI technologies as RLHF (Reinforcement learning through human feedback) and a lot of dp is involved to optimize the policies due to high amount of trajectories (Data essentially). This is why I argued that dp is better for AI compared to other types of problems.