What a great first presentation at the research retreat by one of the participants on control theory He ran a quant firm full of mathematicians, so he needed to exactly determine the bonus structure based on profit made by traders It was highly technical so much of it went over my head, but some key points i did get; 1. We should convert global problems (like how much did this person contribute to the company) into local ones (who was responsible for this $100 trade and how much) 2. We separate out estimation or figuring out weights from control or determining payouts based on obtained parameters 3. For control questions, we change from a graph structure into a matrix, making the whole distribution problem more tractable Much of what we discussed was highly relevant to deep funding. My 2 keys takeaways were - If parts of the matrix are unfilled, can we use distilled human judgment to still estimate their answers? - if deep funding is less of a tree structure and more of a directed acyclic graph, then can recommendation algorithms be applied to getting weights between repos?