For the first question, the first step in the UC Santa Cruz prediction strategy, after they gathered the data, was to perform delay-coordinate embedding of the data It should be noted that they also used several other prediction strategies But, the one were referring to here: use delay-coordinate embedding as the first preprocessing step of the data For Question 2, a neural net-based method worked better than a nonlinear dynamics-based method on SFI dataset A in the long term And this is false While Wans method, which was the neural net method, worked great in the short term, Tim Sauers nonlinear dynamics-based method did far better in the long term So this question is false The third question is simply, How does Lorenzs method of analogues work? And this is near-neighbor prediction on the trajectory of a dynamical system So given a known trajectory of a dynamical system, and a point in that state space, it looks for the points nearest neighbor and the known trajectory, then uses the next point in the trajectory as the forecast of where that point will go next So this first answer is correct The third answer is a simple modification of Lorenzs method of analogues using k nearest neighbors However, the original Lorenzs method of analogues used one neighbor So the first answer is correct