Tossaporn Saengja back
A quick—incomprehensive—reference for my understanding. I overlooked how academics precisely defined terms. This collection aims to highlight how words capture ideas.
Any algorithm that is used on machine to learn. For example, gradient descent lets a machine learns on a loss function given data.
Score is a gradient of diffusion loss. Distillation implies that it uses a trained model. Sampling indicates that it is a stochastic process.
Flow is a trajectory in a differential equation.
Trial and error and then reinforce positive outcomes.
A model that captures language. It predicts a probability of a sequence of tokens in a language.
A type of programming that is can be in contest for competitive environment. Main difference is the code should be fast and not need maintainance.
A delay in computation until it is needed.
A data structure that doesn’t change original data. Useful for time-travel.
The focus is on the “dynamic” word. The approach computes “dynamically” instead of “statically” for solutions, and while it is computing the original (big) solution. It solves smaller solution and utilizes memory “dynamically” to remember subsolutions that are useful for the original solution.
“Real” means real numbers. There is complex analysis for complex numbers. Real analysis is my first mathematical rigor, as I was exposed to “analysis” the first time.
Is it a repeat of search?