SymbolicPhysicsLearner<T>
Models & Types · AiDotNet.PhysicsInformed.ScientificML
Implements Symbolic Physics Learning for discovering interpretable equations from data.
Inheritance: ModelBase<T, Matrix<T>, Vector<T>> → SymbolicPhysicsLearner<T>
How It Works
For Beginners: Symbolic Physics Learner discovers equations in symbolic form (like f = ma, E = mc^2).
Traditional ML:
- Neural networks are “black boxes”
- Learn complex functions but hard to interpret
- Can’t extract simple equations
Symbolic Regression:
- Discovers actual mathematical equations
- Interpretable results (can publish in papers!)
- Can rediscover known physics laws
- Can discover NEW laws
Example: Input: Data of planetary positions vs. time Output: F = Gm1m2/r^2 (Newton’s law of gravitation)
How It Works:
- Search space: Library of operators (+, -, *, /, sin, exp, etc.)
- Search algorithm: Genetic programming, reinforcement learning, etc.
- Fitness: Balance between accuracy and simplicity (Occam’s razor)
- Output: Symbolic expression
Applications:
- Discovering physical laws from experiments
- Automating scientific discovery
- Interpretable AI for science
- Finding conservation laws
Famous Success:
- Rediscovered Kepler’s laws from planetary data
- Found new equations in materials science
- Discovered patterns in quantum mechanics
Example
var inputData = new double[,] { { 1.0 }, { 2.0 }, { 3.0 }, { 4.0 } };
var outputData = new double[] { 1.0, 4.0, 9.0, 16.0 };
var learner = new SymbolicPhysicsLearner<double>();
var equation = learner.DiscoverEquation(inputData, outputData, maxComplexity: 10);
string latex = learner.ToLatex(equation);
Constructors
| Constructor | Summary |
|---|---|
SymbolicPhysicsLearner() |
Properties
| Property | Summary |
|---|---|
ILossFunction<T> DefaultLossFunction { get; } |
Methods
| Method | Summary |
|---|---|
IEnumerable<int> GetActiveFeatureIndices() |
|
IFullModel<T, Matrix<T>, Vector<T>> WithParameters(Vector<T> parameters) |
|
SymbolicExpression<T> DiscoverEquation(T[] inputs, T[] outputs, int maxComplexity, int numGenerations) |
The search itself, on inputs [samples, features] and outputs [samples] already validated. |
SymbolicExpression<T> Simplify(SymbolicExpression<T> expression) |
Simplifies an expression using symbolic algebra rules. |
Vector<T> Predict(Matrix<T> input) |
Predicts outputs using the discovered symbolic equation. |
string ToLatex(SymbolicExpression<T> expression) |
Converts expression to human-readable string. |
void Train(Matrix<T> input, Vector<T> expectedOutput) |
Trains the symbolic physics learner by discovering an equation from the data. |