Verifica dell'IA
Utilizzare tecniche di verifica basate sull'IA per identificare e mitigare i rischi, verificando che i modelli di IA e i sistemi basati sull'IA siano conformi agli standard e alle normative dell'industria. AI Verification Library for Deep Learning Toolbox fornisce gli strumenti per la valutazione e la verifica delle proprietà delle reti neurali profonde. Ad esempio, è possibile verificare le proprietà di solidità di una rete, calcolare i limiti di output della rete, trovare esempi avversari, rilevare i dati fuori distribuzione e verificare la conformità agli standard industriali. Inoltre, il pacchetto di supporto Deep Learning Toolbox Interface for alpha-beta-CROWN Verifier consente la verifica formale delle reti PyTorch® e ONNX™, ad esempio per dimostrare le proprietà di robustezza.
Funzioni
Argomenti
Algoritmi
- Verification of Neural Networks
Learn about verification of neural networks using AI Verification Library for Deep Learning Toolbox™. - Verify Robustness of Deep Learning Neural Network
This example shows how to verify the adversarial robustness of a deep learning neural network. - Verify Robustness of Imported ONNX Network
This example shows how to verify the adversarial robustness of an imported ONNX™ deep neural network. (Da R2024a) - Out-of-Distribution Detection for Deep Neural Networks
This example shows how to detect out-of-distribution (OOD) data in deep neural networks. - Train Robust Deep Learning Network with Jacobian Regularization
Train a neural network that is robust to adversarial examples using a Jacobian regularization scheme. - Reproduce Network Training on a GPU
This example shows how to train a network several times on a GPU and get identical results. (Da R2024b) - Uncertainty Estimation for Regression (Statistics and Machine Learning Toolbox)
Learn about estimating the uncertainty of the true response for a regression problem. - Train Custom Quantile Neural Network
This example shows how to customize and train a neural network that makes quantile predictions. (Da R2026a) - Quantify Uncertainty in Object Detection Using Split Conformal Prediction
This example shows how to apply split conformal prediction (SCP) to an object detection model to quantify uncertainty in the predicted labels and bounding boxes. (Da R2026a)
Serie temporali
- Battery State of Charge Estimation Using Deep Learning
Define requirements, prepare data, train deep learning networks, verify robustness, integrate networks into Simulink, and deploy models. (Da R2024b)
- PASSAGGIO 1: Define Requirements for Battery State of Charge Estimation
- PASSAGGIO 2: Prepare Data for Battery State of Charge Estimation Using Deep Learning
- PASSAGGIO 3: Train Deep Learning Network for Battery State of Charge Estimation
- PASSAGGIO 4: Compress Deep Learning Network for Battery State of Charge Estimation
- PASSAGGIO 5: Test and Verify Deep Learning Network for Battery State of Charge Estimation
- PASSAGGIO 6: Integrate AI Model into Simulink for Battery State of Charge Estimation
- PASSAGGIO 7: Generate Code for Battery State of Charge Estimation Using Deep Learning
- ECG Signal Classification Using Deep Learning
This example shows how to develop and verify a deep learning model that classifies electrocardiogram (ECG) signals to detect atrial fibrillation (AFib). (Da R2026a)
- PASSAGGIO 1: Define Requirements for ECG Signal Classification Using Deep Learning
- PASSAGGIO 2: Prepare Data for ECG Signal Classification
- PASSAGGIO 3: Train Deep Learning Network for ECG Signal Classification
- PASSAGGIO 4: Improve Adversarial Robustness of Deep Learning Network for ECG Signal Classification
- PASSAGGIO 5: Test Deep Learning Network for ECG Signal Classification
- PASSAGGIO 6: Out-of-Distribution Detection for ECG Signal Classification
- PASSAGGIO 7: Uncertainty Quantification for ECG Signal Classification
- PASSAGGIO 8: Investigate ECG Signal Classifications Using Grad-CAM
- PASSAGGIO 9: Build Deep Learning ECG Signal Classification App Using App Designer
Dati tabellari
- Verify and Deploy Airborne Collision Avoidance System (ACAS) Xu Neural Networks
Verify a set of neural networks trained for airborne collision avoidance integrated into a Simulink model using formal methods and scenario-based closed-loop testing. (Da R2026a)
- PASSAGGIO 1: Explore ACAS Xu Neural Networks
- PASSAGGIO 2: Verify Local Robustness of ACAS Xu Neural Networks
- PASSAGGIO 3: Verify Global Stability of ACAS Xu Neural Networks
- PASSAGGIO 4: Verify Global Stability of ACAS Xu Neural Network Using Adaptive Mesh
- PASSAGGIO 5: Verify VNN-COMP Benchmark for ACAS Xu Neural Networks
- PASSAGGIO 6: Verify VNN-COMP Benchmark for ACAS Xu Neural Networks Using α,β-CROWN
- PASSAGGIO 7: Define and Verify AI Constituent Requirements for ACAS Xu Neural Networks
- PASSAGGIO 8: Integrate ACAS Xu Neural Networks into Simulink
- PASSAGGIO 9: Define and Verify AI System Requirements for ACAS Xu Neural Networks Integrated Into Simulink
Visione
- Generate Untargeted and Targeted Adversarial Examples for Image Classification
This example shows how to use the fast gradient sign method (FGSM) and the basic iterative method (BIM) to generate adversarial examples for a pretrained neural network. - Train Image Classification Network Robust to Adversarial Examples
This example shows how to train a neural network that is robust to adversarial examples using fast gradient sign method (FGSM) adversarial training. - Generate Adversarial Examples for Semantic Segmentation
This example shows how to generate adversarial examples for a semantic segmentation network using the basic iterative method (BIM). - Out-of-Distribution Data Discriminator for YOLO v4 Object Detector
This example shows how to detect out-of-distribution (OOD) data in a YOLO v4 object detector. - Verify an Airborne Deep Learning System
This example shows how to verify a deep learning system for airborne applications and is based on the work in [5,6,7], which includes the development and verification activities required by DO-178C [1], ARP4754A [2], and prospective EASA and FAA guidelines [3,4]. (Da R2023b)
Testo
- Out-of-Distribution Detection for BERT Document Classifier
This example shows how to detect out-of-distribution data for a BERT document classifier. (Da R2024b) - Out-of-Distribution Detection for LSTM Document Classifier
This example shows how to detect out-of-distribution (OOD) data in an LSTM document classifier. (Da R2024a)
Workflow di certificazione
- Runway Sign Classifier: Certify an Airborne Deep Learning System (DO Qualification Kit)
Demonstrates the certification of airborne deep learning system.
Esempi in primo piano
Verify Robustness of Deep Learning Neural Network
Verify the adversarial robustness of a deep learning neural network.
Out-of-Distribution Detection for Deep Neural Networks
Detect out-of-distribution (OOD) data in deep neural networks.
Quantify Uncertainty in Object Detection Using Split Conformal Prediction
Apply split conformal prediction (SCP) to an object detection model to quantify uncertainty in the predicted labels and bounding boxes.
Verify an Airborne Deep Learning System
Verify a deep learning system for airborne applications and is based on the work in [5,6,7], which includes the development and verification activities required by DO-178C [1], ARP4754A [2], and prospective EASA and FAA guidelines [3,4]. To verify that the system complies with aviation industry standards and prospective guidelines, including activities such as requirements tracing, testing and reporting, see Runway Sign Classifier: Certify an Airborne Deep Learning System (DO Qualification Kit).
ECG Signal Classification Using Deep Learning
Develop and verify a deep learning model that classifies electrocardiogram (ECG) signals to detect atrial fibrillation (AFib). The example guides you through a complete workflow for building an AI system, with a focus on robustness, explainability, and verification against defined requirements.
Verify and Deploy Airborne Collision Avoidance System (ACAS) Xu Neural Networks
Verify a set of neural networks trained for airborne collision avoidance integrated into a Simulink model using formal methods and scenario-based closed-loop testing.
