Course Objective & Scope
Imperceptible mathematically crafted perturbations in images or subtle feature shifts in tabular records can cause computer vision, fraud detection and malware classification models to fail with high confidence. This course dives deep into the optimisation foundations of inference-time evasion attacks and robust model engineering.
Using PyTorch and the IBM Adversarial Robustness Toolbox (ART), participants execute white-box (FGSM, PGD, C&W, AutoAttack) and black-box (HopSkipJump, Square Attack, transferability) attacks, then harden models and benchmark robustness using adversarial training, input transformations and randomized smoothing.
Who Should Attend?
- Data scientists building computer vision, fraud detection and cybersecurity ML models
- Deep learning and machine learning engineers
- AI security researchers and model validation teams
- Prerequisite: hands-on PyTorch or TensorFlow experience and familiarity with gradient-based optimisation
Key Takeaways
- Generate FGSM, PGD and Carlini & Wagner adversarial examples under L2 and L-infinity norm constraints
- Mount query-based and transfer-based black-box evasion attacks against gradient-free APIs
- Craft realistic tabular adversarial examples that respect domain and feature-mutability constraints
- Apply Madry PGD-AT and TRADES adversarial training while managing the clean-accuracy vs robustness trade-off
- Benchmark and report a model's adversarial robustness curve using AutoAttack and IBM ART
Curriculum
Geometry and Optimisation Formulation of Adversarial Examples
- Decision boundaries, Lp norm constraints (L0, L2, L-infinity) and targeted vs untargeted objectives
- Differences between continuous image/audio inputs and constrained tabular (finance/security) features
White-Box Gradient-Based Evasion Attacks
- Implementing FGSM, BIM, Projected Gradient Descent (PGD) and Carlini & Wagner (C&W)
- Detecting obfuscated gradients and running reliable evaluations with AutoAttack
Black-Box, Transferability and Physical-World Attacks
- Surrogate-model transferability, score-based Square Attack and decision-based HopSkipJump
- Adversarial patches resilient to camera angles, lighting and physical printing
Defenses: Adversarial Training and Certified Robustness
- Min-max saddle-point optimisation with PGD Adversarial Training and TRADES
- Certified robustness radii via randomized smoothing plus input sanitisation and out-of-distribution detectors
Enterprise Robustness Benchmarking with IBM ART
- Auditing an enterprise model end-to-end with the Adversarial Robustness Toolbox (ART)
- Reporting clean accuracy versus robust accuracy across perturbation budgets
Other Courses in This Category
-
Teaches how to detect and prevent label flipping, clean-label poisoning and hidden backdoor/trojan attacks in training and fine-tuning datasets using spectral signatures, activation clustering and data sanitisation.
- Live Course
- 5 modules
-
In-depth training on protecting vision, audio, biometric and multimodal deep neural networks against weight and architecture theft, feature reconstruction, sponge latency attacks and inference-layer compromises.
- Live Course
- 5 modules
-
Teaches how to secure the end-to-end MLOps lifecycle—across data pipelines, Jupyter workspaces, MLflow, Kubeflow, model registries, CI/CD and KServe/Kubernetes serving—using Sigstore/Cosign, ModelScan and Zero Trust controls.
- Live Course
- 5 modules
-
Covers the full OWASP Top 10 for LLM Applications—including direct and indirect prompt injection, improper output handling, unbounded consumption and sensitive data disclosure—from a secure software development (SSDLC) perspective.
- Live Course
- 5 modules
-
Hands-on training on conducting end-to-end penetration tests against classical ML, deep learning and LLM APIs using MITRE ATLAS, Counterfit, ART, garak and PyRIT, and producing actionable pentest reports.
- Live Course
- 5 modules
-
Hands-on training on applying machine learning and LLMs across cybersecurity—from phishing and malware detection to vulnerability prioritisation, cyber threat intelligence (CTI) analysis and SecOps assistants.
- Live Course
- 5 modules
Get in touch
