Libraries for browserbased mathematics, e.g. for browserbased machine learning ranked in descending order of viability:
weblas does GPUaccelerated matehamtics and is sued in hip projects such as kerasjs.

LALOLib (Linear ALgebra Online Library) is a library entirely written in Javascript to enable and ease mathematical computing within web pages. It is developed as a part of the MLweb project and constitutes the core of LALOLab (an online scientific computing environment). It also provides the basic building blocks for ML.js, a javascript library for machine learning.
This looks genuinely amazing in terms of functionality and even includes native support for worker threads and concurrency. However… it is lacking modern web wrappings such as npm packaging etc, so is not conveneient to use from e.g. webpack.

Math.js is an extensive math library for JavaScript and Node.js. It features a flexible expression parser, a large set of builtin functions and constants, and offers an integrated solution to work with different data types like numbers, big numbers, complex numbers, units, and matrices

Efficient, highperformance linear algebra library for node.js and browsers.
This is a lowlevel algebra library which supports basic vector and matrix operations, and has been designed with machine learning algorithms in mind.
Features:
Simple, expressive, chainable API. Array implementation with performance optimizations. Enhanced floating point precision if needed. Comprehensive unit tests. Works in node.js and browsers. Small: ~1 KB minified and gzipped.
sylvester the original, but predates much modern optimisation such as native arrays and asm.js
jStat is a statistical library written in JavaScript that allows you to perform advanced statistical operations without the need of a dedicated statistical language (e.g. MATLAB or R). Includes a tidy linear algebra library, but could be better optimised.

JavaScript (emscripten) port of lmfit library:
“a selfcontained C library for LevenbergMarquardt leastsquares minimization and curve fitting”
Currently only linear curve fitting is implemented.
linalg uses native arrays because of their speed.
I needed a performance focused linear algebra module for visualizing data in 10+ dimensions, and implementing machine learning algorithms. I quickly learned that naive solutions to linear algebra operations can produce numerical errors so significant they are utterly useless for anything other than casual playtime. After that, I prioritized correctness over performance.â€ť
Untouched since released and small community, which is sad because the code looks solid.
glmatrix is WebGL (therefore VERY fast), but 4vector oriented, which is too small for us
numeric looks polished but has been untouched for 2 years
jmat is an actively developed complex matrix library, but we would probably prefer speed to complex number support.
random variables can be simulated easily using the probability distributions library