Langevin dynamics does: adds noise to gradient descent to sample from a distribution

How can we simulate randomness in physics?

Imagine you're trying to predict the path of a leaf falling from a tree. The wind's unpredictable gusts make it hard to predict exactly where it will land.

Langevin dynamics adds random 'wind gusts' to the leaf's fall, helping us understand its unpredictable path. This randomness helps us sample from a distribution of possible outcomes.

Example

If the leaf's path is influenced by wind gusts, adding random gusts can show us many possible landing spots, not just one predicted path.

Remember this

Langevin dynamics helps us understand complex systems by adding randomness to predict outcomes.

Text adapted from Wikipedia, licensed under CC BY-SA 4.0.

Related concepts

Diffusion model

q(x_t|x_{t-1}) adds Gaussian noise at each step

Lyapunov exponents measure: rate of divergence of nearby trajectories in a dynamical system

How do tiny differences grow over time in chaotic systems?

Markov chain Monte Carlo

MCMC samples from complex posterior distributions

the reverse process learns: p_θ(x_{t-1}|x_t)

Can we trace back the roots of life?

Stable Diffusion

Stable Diffusion generates images from text descriptions

List of unsolved problems in mathematics

Why do random points in high dimensions seem to be evenly spaced?

Swipe through 100 ML concepts daily

Open Pocket Polymath