/explore
|
You are here 
|
www.jeremykun.com | ||
| | | | |
francisbach.com
|
|
| | | | | [AI summary] This text discusses the scaling laws of optimization in machine learning, focusing on asymptotic expansions for both strongly convex and non-strongly convex cases. It covers the derivation of performance bounds using techniques like Laplace's method and the behavior of random minimizers. The text also explains the 'weird' behavior observed in certain plots, where non-strongly convex bounds become tight under specific conditions. The analysis connects theoretical results to practical considerations in optimization algorithms. | |
| | | | |
sebastianraschka.com
|
|
| | | | | I'm Sebastian: a machine learning & AI researcher, programmer, and author. As Staff Research Engineer Lightning AI, I focus on the intersection of AI research, software development, and large language models (LLMs). | |
| | | | |
fa.bianp.net
|
|
| | | | | The Langevin algorithm is a simple and powerful method to sample from a probability distribution. It's a key ingredient of some machine learning methods such as diffusion models and differentially private learning. In this post, I'll derive a simple convergence analysis of this method in the special case when the ... | |
| | | | |
peterbloem.nl
|
|
| | | [AI summary] The pseudo-inverse is a powerful tool for solving matrix equations, especially when the inverse does not exist. It provides exact solutions when they exist and least squares solutions otherwise. If multiple solutions exist, it selects the one with the smallest norm. The pseudo-inverse can be computed using the singular value decomposition (SVD), which is numerically stable and handles cases where the matrix does not have full column rank. The SVD approach involves computing the SVD of the matrix, inverting the non-zero singular values, and then reconstructing the pseudo-inverse using the modified SVD components. This method is preferred due to its stability and ability to handle noisy data effectively. | ||
