Andrew Drakeford discusses techniques for optimizing data-oriented design, drawing from mathematician George Polya's problem-solving approach. He emphasizes understanding the problem domain and using heuristics like decomposition and drawing pictures to find solutions. Drakeford presents two real-world examples: optimizing a troublesome function in a moment matching pricing algorithm and accelerating leave-one-out regression in a machine learning quant fund. He highlights the importance of expanding the problem's context, creating auxiliary problems, and avoiding premature optimization of inner loops. The talk concludes with a discussion on improving summation accuracy and a call to action for listeners to apply these techniques and share their results. A brief Q&A follows the presentation.
Sign in to continue reading, translating and more.
Continue
