transfer_learning/
transfer_learning.rs

1#![allow(clippy::pedantic, clippy::unnecessary_wraps)]
2//! Quantum Transfer Learning Example
3//!
4//! This example demonstrates how to use pre-trained quantum models
5//! and fine-tune them for new tasks with transfer learning.
6
7use quantrs2_ml::autodiff::optimizers::Adam;
8use quantrs2_ml::prelude::*;
9use quantrs2_ml::qnn::QNNLayerType;
10use scirs2_core::ndarray::{Array1, Array2};
11
12fn main() -> Result<()> {
13    println!("=== Quantum Transfer Learning Demo ===\n");
14
15    // Step 1: Load a pre-trained model from the model zoo
16    println!("1. Loading pre-trained image classifier...");
17    let pretrained = QuantumModelZoo::get_image_classifier()?;
18
19    println!("   Pre-trained model info:");
20    println!("   - Task: {}", pretrained.task_description);
21    println!(
22        "   - Original accuracy: {:.2}%",
23        pretrained
24            .performance_metrics
25            .get("accuracy")
26            .unwrap_or(&0.0)
27            * 100.0
28    );
29    println!("   - Number of qubits: {}", pretrained.qnn.num_qubits);
30
31    // Step 2: Create new layers for the target task
32    println!("\n2. Creating new layers for text classification task...");
33    let new_layers = vec![
34        QNNLayerType::VariationalLayer { num_params: 6 },
35        QNNLayerType::MeasurementLayer {
36            measurement_basis: "Pauli-Z".to_string(),
37        },
38    ];
39
40    // Step 3: Initialize transfer learning with different strategies
41    println!("\n3. Testing different transfer learning strategies:");
42
43    // Strategy 1: Fine-tuning
44    println!("\n   a) Fine-tuning strategy (train last 2 layers only)");
45    let mut transfer_finetune = QuantumTransferLearning::new(
46        pretrained.clone(),
47        new_layers.clone(),
48        TransferStrategy::FineTuning {
49            num_trainable_layers: 2,
50        },
51    )?;
52
53    // Strategy 2: Feature extraction
54    println!("   b) Feature extraction strategy (freeze all pre-trained layers)");
55    let transfer_feature = QuantumTransferLearning::new(
56        pretrained.clone(),
57        new_layers.clone(),
58        TransferStrategy::FeatureExtraction,
59    )?;
60
61    // Strategy 3: Progressive unfreezing
62    println!("   c) Progressive unfreezing (unfreeze one layer every 5 epochs)");
63    let transfer_progressive = QuantumTransferLearning::new(
64        pretrained,
65        new_layers,
66        TransferStrategy::ProgressiveUnfreezing { unfreeze_rate: 5 },
67    )?;
68
69    // Step 4: Generate synthetic training data for the new task
70    println!("\n4. Generating synthetic training data...");
71    let num_samples = 50;
72    let num_features = 4;
73    let training_data = Array2::from_shape_fn((num_samples, num_features), |(i, j)| {
74        (i as f64).mul_add(0.1, j as f64 * 0.2).sin()
75    });
76    let labels = Array1::from_shape_fn(num_samples, |i| if i % 2 == 0 { 0.0 } else { 1.0 });
77
78    // Step 5: Train with fine-tuning strategy
79    println!("\n5. Training with fine-tuning strategy...");
80    let mut optimizer = Adam::new(0.01);
81
82    let result = transfer_finetune.train(
83        &training_data,
84        &labels,
85        &mut optimizer,
86        20, // epochs
87        10, // batch_size
88    )?;
89
90    println!("   Training complete!");
91    println!("   - Final loss: {:.4}", result.final_loss);
92    println!("   - Accuracy: {:.2}%", result.accuracy * 100.0);
93
94    // Step 6: Extract features using pre-trained layers
95    println!("\n6. Extracting features from pre-trained layers...");
96    let features = transfer_feature.extract_features(&training_data)?;
97    println!("   Extracted feature dimensions: {:?}", features.dim());
98
99    // Step 7: Demonstrate model zoo
100    println!("\n7. Available pre-trained models in the zoo:");
101    println!("   - Image classifier (4 qubits, MNIST subset)");
102    println!("   - Chemistry model (6 qubits, molecular energy)");
103
104    // Load chemistry model
105    let chemistry_model = QuantumModelZoo::get_chemistry_model()?;
106    println!("\n   Chemistry model info:");
107    println!("   - Task: {}", chemistry_model.task_description);
108    println!(
109        "   - MAE: {:.4}",
110        chemistry_model
111            .performance_metrics
112            .get("mae")
113            .unwrap_or(&0.0)
114    );
115    println!(
116        "   - R² score: {:.4}",
117        chemistry_model
118            .performance_metrics
119            .get("r2_score")
120            .unwrap_or(&0.0)
121    );
122
123    println!("\n=== Transfer Learning Demo Complete ===");
124
125    Ok(())
126}
127
128// Helper function to visualize layer configurations
129fn print_layer_configs(configs: &[LayerConfig]) {
130    for (i, config) in configs.iter().enumerate() {
131        println!(
132            "   Layer {}: frozen={}, lr_multiplier={:.2}",
133            i, config.frozen, config.learning_rate_multiplier
134        );
135    }
136}
transfer_learning/transfer_learning.rs

transfer_learning/
transfer_learning.rs
