transfer_learning/
transfer_learning.rs

1//! Quantum Transfer Learning Example
2//!
3//! This example demonstrates how to use pre-trained quantum models
4//! and fine-tune them for new tasks with transfer learning.
5
6use scirs2_core::ndarray::{Array1, Array2};
7use quantrs2_ml::autodiff::optimizers::Adam;
8use quantrs2_ml::prelude::*;
9use quantrs2_ml::qnn::QNNLayerType;
10
11fn main() -> Result<()> {
12    println!("=== Quantum Transfer Learning Demo ===\n");
13
14    // Step 1: Load a pre-trained model from the model zoo
15    println!("1. Loading pre-trained image classifier...");
16    let pretrained = QuantumModelZoo::get_image_classifier()?;
17
18    println!("   Pre-trained model info:");
19    println!("   - Task: {}", pretrained.task_description);
20    println!(
21        "   - Original accuracy: {:.2}%",
22        pretrained
23            .performance_metrics
24            .get("accuracy")
25            .unwrap_or(&0.0)
26            * 100.0
27    );
28    println!("   - Number of qubits: {}", pretrained.qnn.num_qubits);
29
30    // Step 2: Create new layers for the target task
31    println!("\n2. Creating new layers for text classification task...");
32    let new_layers = vec![
33        QNNLayerType::VariationalLayer { num_params: 6 },
34        QNNLayerType::MeasurementLayer {
35            measurement_basis: "Pauli-Z".to_string(),
36        },
37    ];
38
39    // Step 3: Initialize transfer learning with different strategies
40    println!("\n3. Testing different transfer learning strategies:");
41
42    // Strategy 1: Fine-tuning
43    println!("\n   a) Fine-tuning strategy (train last 2 layers only)");
44    let mut transfer_finetune = QuantumTransferLearning::new(
45        pretrained.clone(),
46        new_layers.clone(),
47        TransferStrategy::FineTuning {
48            num_trainable_layers: 2,
49        },
50    )?;
51
52    // Strategy 2: Feature extraction
53    println!("   b) Feature extraction strategy (freeze all pre-trained layers)");
54    let transfer_feature = QuantumTransferLearning::new(
55        pretrained.clone(),
56        new_layers.clone(),
57        TransferStrategy::FeatureExtraction,
58    )?;
59
60    // Strategy 3: Progressive unfreezing
61    println!("   c) Progressive unfreezing (unfreeze one layer every 5 epochs)");
62    let transfer_progressive = QuantumTransferLearning::new(
63        pretrained.clone(),
64        new_layers.clone(),
65        TransferStrategy::ProgressiveUnfreezing { unfreeze_rate: 5 },
66    )?;
67
68    // Step 4: Generate synthetic training data for the new task
69    println!("\n4. Generating synthetic training data...");
70    let num_samples = 50;
71    let num_features = 4;
72    let training_data = Array2::from_shape_fn((num_samples, num_features), |(i, j)| {
73        (i as f64 * 0.1 + j as f64 * 0.2).sin()
74    });
75    let labels = Array1::from_shape_fn(num_samples, |i| if i % 2 == 0 { 0.0 } else { 1.0 });
76
77    // Step 5: Train with fine-tuning strategy
78    println!("\n5. Training with fine-tuning strategy...");
79    let mut optimizer = Adam::new(0.01);
80
81    let result = transfer_finetune.train(
82        &training_data,
83        &labels,
84        &mut optimizer,
85        20, // epochs
86        10, // batch_size
87    )?;
88
89    println!("   Training complete!");
90    println!("   - Final loss: {:.4}", result.final_loss);
91    println!("   - Accuracy: {:.2}%", result.accuracy * 100.0);
92
93    // Step 6: Extract features using pre-trained layers
94    println!("\n6. Extracting features from pre-trained layers...");
95    let features = transfer_feature.extract_features(&training_data)?;
96    println!("   Extracted feature dimensions: {:?}", features.dim());
97
98    // Step 7: Demonstrate model zoo
99    println!("\n7. Available pre-trained models in the zoo:");
100    println!("   - Image classifier (4 qubits, MNIST subset)");
101    println!("   - Chemistry model (6 qubits, molecular energy)");
102
103    // Load chemistry model
104    let chemistry_model = QuantumModelZoo::get_chemistry_model()?;
105    println!("\n   Chemistry model info:");
106    println!("   - Task: {}", chemistry_model.task_description);
107    println!(
108        "   - MAE: {:.4}",
109        chemistry_model
110            .performance_metrics
111            .get("mae")
112            .unwrap_or(&0.0)
113    );
114    println!(
115        "   - R² score: {:.4}",
116        chemistry_model
117            .performance_metrics
118            .get("r2_score")
119            .unwrap_or(&0.0)
120    );
121
122    println!("\n=== Transfer Learning Demo Complete ===");
123
124    Ok(())
125}
126
127// Helper function to visualize layer configurations
128fn print_layer_configs(configs: &[LayerConfig]) {
129    for (i, config) in configs.iter().enumerate() {
130        println!(
131            "   Layer {}: frozen={}, lr_multiplier={:.2}",
132            i, config.frozen, config.learning_rate_multiplier
133        );
134    }
135}
transfer_learning/transfer_learning.rs

transfer_learning/
transfer_learning.rs
