Scientific ML Studio
Home

SciML Studio

Bring Scientific knowledge and Machine Learning together.

Scientific machine learning (SciML) brings scientific computing and machine learning together to model physical systems, infer unknown parameters, and support engineering analysis. Its practical value extends from combining equations with observations to developing models for repeated predictions.

Start reading Open the Lab
2 Open textbooks

How SciMLStudio works

Define the physics. Connect the blocks. Get the code.

Scientific machine learning lets a neural network learn from governing equations, measurements and simulations at the same time, and then checks what it predicts against a reference. The physics is the hard, interesting part. SciMLStudio makes the formulation visible and writes the implementation for you.

Open the Lab Read the textbook

Version 0 supports PINNs. Code runs in your own Python environment.

Equations Measurements Simulations Prediction check against a reference
Equations, measurements and simulations all inform the model, and its predictions are checked against a reference.

From problem to script

Five steps, and you stay in charge of the physics.

Domain PDE Conditions Loss Network ✓ checks passed
train.py ready
# 1 · domain: x, t in [0, 1] x = torch.rand(N, 1, requires_grad=True) t = torch.rand(N, 1, requires_grad=True)
# 2 · heat equation: u_t = alpha * u_xx def residual(net, x, t): u = net(torch.cat([x, t], 1)) u_t, u_x = grad(u, t, x) (u_xx,) = grad(u_x, x) return u_t - alpha * u_xx
# 3 · conditions def conditions(net): ic = net(pts_t0) - sin(pi * x0) bc = net(pts_edges) return ic, bc
# 4 · network, loss, optimiser net = MLP([2, 64, 64, 64, 1], tanh) opt = Adam(net.parameters(), lr=1e-3) def loss(): ic, bc = conditions(net) pde = residual(net, x, t) return mse(pde) + mse(ic) + mse(bc)
# 5 · train, then run it anywhere for step in range(20000): opt.zero_grad() loss().backward() opt.step()

Simplified excerpt, for illustration.

  • Visual blocks

    Domain, PDE, conditions, network, loss, optimizer.

  • Compatibility checks

    Warnings for incompatible blocks or settings.

  • Code generation

    Download a Python script.

  • Starter examples

    Heat, wave, Burgers’, Poisson, Laplace.

Learn

The textbook. Chapters on the left, maths set properly, your place kept in this browser.

Free to read Lab

Wire a problem together on a canvas; take away the PyTorch that solves it.

Runs in your browser

From the textbook

Physics-Informed Neural Networks, from scratch

From the first derivative to a solver you can defend.

5 sections · free to read

Studio Tour

Step-by-step problems to wire up in the Lab.

2 sections · free to read

Read, watch, build

SciMLStudio makes this implementation more accessible. Define your problem using configurable blocks, connect them into a workflow, and generate Python code. Compatibility checks and automatic detection of supported settings guide the process, while you retain control over the physical formulation and modeling choices. The textbook stays free. The video guides show each step being done, and the Lab turns what you have read into a PyTorch script you can run and defend.

Open the Lab What this is