Module neural_physics 

Neural network-based physics acceleration (CPU mock).

Provides a simple feed-forward neural network with multiple activation functions, forward-pass inference, MSE loss, ML force potentials, and collision probability prediction — all on CPU as a GPU mock backend.

Structs

NeuralLayer

A single fully-connected neural network layer.

NeuralNet

A feed-forward neural network composed of stacked NeuralLayers.

Enums

ActivationType

Activation function used in a neural layer.

Functions

activate

Evaluate activation function act at scalar x.

activate_derivative

Evaluate the derivative of activation function act at pre-activation x.

create_network

Create a fully-connected network with the given layer sizes and random weights.

forward_pass

Run a forward pass through net, returning the output vector.

gpu_neural_batch_forward

Run a batched GPU-style forward pass for multiple input vectors.

mse_loss

Mean squared error between predicted and target vectors.

neural_collision_check

Predict collision probability between two spheres using a neural network.

neural_force_prediction

Predict interatomic forces using a neural network potential.