Gradientscape
A 3D interactive environment for exploring gradient descent — built for anyone trying to understand how machine learning actually learns.
“When I first truly understood gradient descent, it felt like watching the universe explain itself. I had to build something to give other people that same feeling.”
Gradient descent was the concept that made machine learning click for me—not because of the formula, but because of the idea behind it. The notion that a model could learn by simply asking 'which direction is down?' and taking a small step in that direction, over and over, until it found the minimum—it felt like magic. I devoted a serious amount of time to building Gradientscape because I wanted to give other people that same moment of clarity. It's a fully interactive 3D environment where you drop a sphere onto a loss landscape and watch gradient descent work in real time. You can choose from different surface types, adjust the learning rate, swap between optimizers like SGD and Adam, tune the data noise, and modify the loss function—and everything updates live in three dimensions, rendered with Three.js. For those who want to understand not just the visuals but the theory, Gradientscape includes 'The Complete Beginner's Guide to Gradient Descent'—an extensive walkthrough that takes someone from curious beginner to genuine understanding of the algorithm, the math, and the intuition behind it.
3D Landscape Explorer
Choose a surface type, drop a sphere anywhere on it, and watch gradient descent play out in three dimensions—rendered in real time with Three.js.
Full Parameter Control
Adjust the learning rate, optimizer, loss function, and data noise in real time. Every change is reflected immediately in the 3D visualization.
Multiple Optimizers
Compare SGD, Adam, RMSProp, and others to understand exactly how different optimization strategies behave on the same landscape.
The Complete Beginner's Guide
An extensive built-in guide that takes anyone from zero to a solid understanding of gradient descent, the underlying math, and the intuition behind why it works.