Postgraduate Diploma in Safe Retraining and Model Production Deployment
Online 800-hour Postgraduate Diploma in Safe Retraining and Model Production Deployment. 8 modules covering ISO/PAS 8800:2024, UNECE R156 and MLflow. Assigned tutor and open enrolment.
Take the controls.
Run a real engineering trade-off from this field. Move the sliders, watch the model react in real time and beat the challenge.
What you will be able to do
Assess the risks of retraining models in safety-critical systems.
Select retraining strategies that prevent regressions.
Design non-regression suites by ODD segment.
Validate models for release in the vehicle.
Plan progressive deployments with safe rollback.
Monitor models after they go into production.
Govern the change cycle for models and data.
Who it's for
- MLOps engineer who wants to specialise in Safe Retraining and Model Production Deployment.
- Model validation engineer who wants to specialise in Safe Retraining and Model Production Deployment.
- Software release manager who wants to specialise in Safe Retraining and Model Production Deployment.
- AI safety engineer who wants to specialise in Safe Retraining and Model Production Deployment.
- Perception engineer who wants to specialise in Safe Retraining and Model Production Deployment.
- Fleet deployment manager who wants to specialise in Safe Retraining and Model Production Deployment.
- Graduates in engineering or science who want to join autonomous driving companies.
What you'll work with
- ISO/PAS 8800:2024
- UNECE R156
- ISO 24089:2023
- ISO 26262-8:2018
- ISO 21448:2022
- Regulation (EU) 2024/1689
- ISO/IEC 42001:2023
- ISO/IEC 5338:2023
8 modules · 800 h
1.1Reasons to retrain: drift, new ODDs and defects1.2Catastrophic forgetting and hidden regressions1.3Unintended behaviour changes1.4Impact on the existing safety case1.5ISO/PAS 8800 change requirements1.6Change impact analysis under ISO 26262-81.7Data poisoning risks1.8Lessons from field software recalls1.9Responsibilities in the release decision1.10Case study: impact analysis of a detector retraining
2.1Full retraining versus fine-tuning2.2Continual and incremental learning2.3Experience replay and regularisation2.4Knowledge distillation between versions2.5Selection of new and old data2.6Layer freezing and adapters2.7Training reproducibility2.8Computational cost by strategy2.9Selection criteria2.10Lab: fine-tuning with catastrophic forgetting control
3.1Frozen evaluation sets3.2Testing by ODD segment and by class3.3Instance-level regression metrics3.4Bias comparison between versions3.5Repeated robustness tests3.6Closed-loop evaluation in simulation3.7Replaying field scenarios3.8Acceptance thresholds and blocking criteria3.9Version comparison report3.10Practice: non-regression suite for a segmentation model
4.1Model quality gates4.2Verification on target hardware and quantisation4.3Latency and memory usage testing4.4Integration with the driving stack4.5HIL and track testing of the new model4.6Updating safety arguments4.7Approval by the release committee4.8Required technical documentation4.9Signature and traceability of the artefact4.10Workshop: release checklist for a perception model
5.1Shadow mode and in-vehicle comparison5.2Canary deployments by fleet segment5.3A/B testing with safety metrics5.4Feature flags for models5.5Criteria for progression and rollback5.6UNECE R156 and ISO 24089 requirements5.7Communication with users and operators5.8Management of hardware variants5.9Planning update windows5.10Case study: canary deployment plan for a planner
6.1Field performance indicators6.2Detection of behavioural anomalies6.3Telemetry of disengagements and warnings6.4Comparison with the previous version6.5Drift detection after deployment6.6Feedback loop to data6.7Incident reporting to authorities6.8Post-market monitoring under the AI Act6.9Safety dashboards6.10Practice: dashboard for a version in the field
7.1Configuration management of models and data7.2Model registry and approvals7.3Change management under ISO/IEC 420017.4Relationship with the software update management system7.5Auditing release decisions7.6Security of the training chain7.7Roles and responsibility matrix7.8Process quality indicators7.9Continuous improvement7.10Workshop: model change management procedure
8.1Model selection and reason for change8.2Impact analysis8.3Retraining strategy8.4Non-regression suite8.5Validation for release8.6Progressive deployment plan8.7Production monitoring plan8.8Update of the safety case8.9Report and presentation8.10Final review with the tutor
How you'll learn
100 % online training on the SEIUM Institute campus, with downloadable material, recorded classes and an assigned technical tutor. There are no fixed timetables: students progress at their own pace within the enrolment period, which makes it easy to combine with work.
Each module combines theory, guided reading of the original standard and applied exercises. The working software is MLflow, PyTorch, NVIDIA TensorRT, Weights & Biases, Kubeflow, in free, academic or evaluation versions. Tests are studied using real model non-regression testing procedures, shadow-mode fleet execution and canary deployments with rollback. Assessment is continuous, module by module, plus the final integrative project.
Build it end to end
Your final project follows the real engineering workflow, step by step, reviewed by your tutor.
Assess the risks of retraining models in safety-critical systems.
Select retraining strategies that prevent regressions.
Design non-regression suites by ODD segment.
Validate models for release in the vehicle.
Plan progressive deployments with safe rollback.
Where this takes you
Requirements
- Prior technical training or professional experience in the sector. A university degree is not required.
- Internet connection and a computer capable of running the programme's software.
- Sufficient level of Spanish to follow technical regulatory texts.
Ready to start?
Send your application today. Our admissions team replies within 72 hours with your study plan and funding options.
Frequently asked questions
The programme consists of 800 hours in total, spread over 6 to 8 months. That amounts to between 10 and 15 hours per week on average, compatible with a full-time job.
You will work with MLflow, PyTorch, NVIDIA TensorRT, Weights & Biases, Kubeflow. Free, academic or evaluation versions are used, and the tutor will indicate a free alternative where one exists.
A SEIUM Institute diploma upon passing the continuous assessment and the integrative project. It is an in-house programme, not an official Ministry qualification.
Yes. Modules passed in a course or postgraduate diploma in the same area can be credited towards the advanced programme. Check the recognition policy when you enrol.
Related programmes
Postgraduate Diploma in Dense LiDAR and 3D SLAM for Automotive
Online 800-hour Postgraduate Diploma in Dense LiDAR and 3D SLAM for Automotive. 8 modules covering IEC 60825-1:2014, ISO 26262:2018 and Open3D. Assigned tutor and open enrolment.
Postgraduate Diploma in Runtime Monitoring and Safety Kernels
Online 800-hour Postgraduate Diploma in Runtime Monitoring and Safety Kernels. 8 modules covering ISO 26262:2018, UL 4600 and RTAMT. Assigned tutor and open enrolment.
Postgraduate Diploma in Semantic Cartography and Risk-Prioritised Maps
Online Postgraduate Diploma of 800 hours in Semantic Cartography and Risk-Prioritised Maps. 8 modules covering Directive (EU) 2019/1936, Directive 2008/96/CE and QGIS. Dedicated tutor and enrolment…
More in Autonomous, ADAS & Software-Defined Vehicle
All Autonomous, ADAS & Software-Defined Vehicle programmes → · How to apply · Fees & financing · Scholarships