Master’s in Multimodality Engineering
Online master's programme of 1,900 hours in multimodal AI engineering. 12 modules covering modality fusion and alignment, per-modality representations, transformer architectures, contrastive learning, multimodal language models, multimodal generation, sensor fusion, multimodal RAG, agents and efficient deployment under the AI Act, the GDPR and ISO/IEC 42001, with an assigned technical tutor.
Master’s in Multimodality Engineering — studying from Canada
Canada is a recognised centre for artificial intelligence research, with hubs in Toronto, Montréal and Edmonton supported by institutes such as the Vector Institute, Mila and Amii. Engineering teams across manufacturing, energy, finance and transport apply simulation, machine learning and data analytics to real operational problems. The federal government's Pan-Canadian AI Strategy and privacy regulation shape how organisations handle data. Typical employers include technology firms, banks, aerospace and automotive companies, and public research bodies.
Fees for Canada: C$13,900 (Canadian dollar), with instalments and employer sponsorship available.
Relevant bodies in Canada
Typical roles
Questions from Canada
Do Canadian bodies recognise data or AI qualifications from private institutes?
Data and AI roles in Canada are generally not licensed in the way engineering is, so recognition depends on employers and professional communities rather than a regulator. Engineering licensing through provincial associations applies only to engineering titles. SEIUM's qualifications are private, not official state degrees and not accredited by a national body, so employers decide how much weight to give them.
Do Canadian employers hire for data and AI skills, and would this help?
Canada has a large and active data and AI sector, with notable hubs in Toronto, Montreal and Vancouver, and employers hire widely for these skills. They usually look first at portfolios, technical interviews and proven projects. A SEIUM final project can demonstrate your ability, but a private qualification's value is decided by each employer, so present it alongside concrete work.
Is this manageable with a job, and what level of English is needed?
Because the programmes are fully online, you can study from home around your job, and no student visa is needed. Specialist courses are 300 hours, postgraduate diplomas 800 and master's 1,900, so plan your weekly time. Teaching is in English, so strong written and reading English is important. Instalment payments and possible employer sponsorship can help with funding.
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
Represent data from different modalities in shared spaces.
Design transformer-based multimodal fusion architectures.
Train models with contrastive learning.
Adapt multimodal language models to specific tasks.
Build multimodal retrieval and RAG systems.
Fuse heterogeneous sensors in engineering applications.
Assess the risks and regulatory compliance of multimodal systems.
Who it's for
- Multimodal AI engineers building systems that combine text, image, audio and video.
- Machine learning engineers expanding beyond single-modality models.
- Vision-language engineers adapting foundation models to industrial tasks.
- Sensor fusion engineers working on perception and diagnostics.
- Generative AI researchers interested in multimodal generation.
- RAG systems engineers working with complex technical documentation.
- Engineering or science graduates looking to enter the technology sector.
What you'll work with
- Regulation (EU) 2024/1689 (AI Act)
- Regulation (EU) 2016/679 (GDPR)
- ISO/IEC 42001 (AI management systems)
- ISO/IEC 22989 (AI concepts and terminology)
- ISO/IEC 23053 (framework for AI systems using machine learning)
- Directive (EU) 2019/790 (copyright in the Digital Single Market)
- ISO/IEC 23894 (AI risk management)
- PyTorch
- Hugging Face Transformers
- OpenCLIP
- LlamaIndex
- Weaviate
12 modules · 1900 h
1.1Modalities: text, image, audio, video and sensors1.2Heterogeneity, alignment and complementarity1.3Joint and coordinated representations1.4Early, late and intermediate fusion1.5Translation between modalities1.6Co-learning and transfer1.7Taxonomy of multimodal tasks1.8Historical evolution of the models1.9Open research challenges1.10Case study: analysis of an industrial multimodal system
2.1Text embeddings and tokenisation2.2Image representations with CNNs and ViTs2.3Audio representations and spectrograms2.4Speech models such as wav2vec and Whisper2.5Video representations2.6Sensor time series2.7Graphs and structured data2.8Point clouds and 3D2.9Normalisation across modalities2.10Lab: extracting embeddings from several modalities
3.1Attention and self-attention3.2Cross-attention between modalities3.3Encoders and decoders3.4Visual tokens and patches3.5Position and time in multimodal sequences3.6Mixture of experts3.7Model scaling3.8Attention efficiency3.9Interpreting attention maps3.10Practical: image-text cross-attention transformer
4.1Contrastive learning with CLIP4.2The InfoNCE loss function4.3Shared embedding spaces4.4Zero-shot classification4.5Cross-modal image-text retrieval4.6Audio-text alignment4.7ImageBind and multiple modalities4.8Noisy web data4.9Biases in pre-training data4.10Lab: semantic image search engine with CLIP
5.1Vision-language model architectures5.2Visual adapters and projectors5.3Multimodal instruction tuning5.4Visual question answering5.5Document understanding5.6Video understanding5.7Conversational speech models5.8Hallucinations in multimodal models5.9Evaluation with benchmarks5.10Workshop: efficient fine-tuning of a vision-language model
6.1Text-to-image diffusion models6.2Controlling generation6.3Video generation6.4Speech synthesis6.5Audio and music generation6.6Text-to-3D generation6.7Guided image editing6.8Generative quality metrics6.9Marking synthetic content6.10Practical: controlled generation with diffusion
7.1Vision and LiDAR in perception7.2Radar and camera7.3Audio and vibration in diagnostics7.4Thermal and visible imaging7.5Fusion with Bayesian filters7.6Fusion with deep networks7.7Synchronisation and calibration7.8Robustness to sensor failures7.9Industrial applications of sensor fusion7.10Lab: camera-LiDAR fusion for 3D detection
8.1Multimodal vector indexes8.2Retrieving documents with images8.3RAG over technical manuals8.4Video search8.5Re-ranking multimodal results8.6Evaluating retrieval8.7Chunking complex documents8.8Vector databases8.9Latency and cost8.10Workshop: RAG assistant over drawings and manuals
9.1Agents that see and act9.2Tool use from models9.3Voice interfaces9.4Agents for graphical interfaces9.5Robotics with vision-language-action models9.6Orchestrating specialised models9.7Memory and long context9.8Agent security9.9Agent evaluation9.10Practical: multimodal agent for incident inspection
10.1Fine-tuning with LoRA10.2Quantisation of multimodal models10.3Distillation of large models10.4Edge inference10.5Serving large models10.6Optimising training data10.7Curating multimodal datasets10.8Computational cost of training10.9MLOps practices for multimodal systems10.10Lab: quantising and deploying a vision-language model
11.1Multimodal evaluation benchmarks11.2Human evaluation protocols11.3Multimodal bias and fairness11.4Safety and harmful content11.5Multimodal adversarial attacks11.6Privacy in images and voice11.7The AI Act and general-purpose models11.8Copyright in training data11.9Accessibility through multimodal descriptions11.10Case study: risk assessment of a multimodal assistant
12.1Defining the multimodal task12.2Data and curation12.3Architecture design12.4Training or adaptation12.5Quantitative evaluation of the system12.6Error and bias analysis12.7Prototype deployment12.8Conclusions and future work12.9Master's Thesis report12.10Master's Thesis: defence before a panel
How you'll learn
100% online training on the SEIUM campus, with downloadable material, recorded classes and an assigned technical tutor. There are no fixed timetables: you progress at your own pace within the intake, so you can combine the programme with your job.
Each module combines theory, guided reading of the original standard and applied exercises. The working software is PyTorch, Hugging Face Transformers, OpenCLIP, LlamaIndex and Weaviate, in free, academic or evaluation versions. Real procedures are studied: a semantic image and text search engine, a multimodal RAG assistant over technical documentation, and camera-LiDAR fusion for perception. Assessment is continuous by module, plus the final Master's Thesis.
Standards covered: Regulation (EU) 2024/1689 (AI Act), Regulation (EU) 2016/679 (GDPR), ISO/IEC 42001, ISO/IEC 22989, ISO/IEC 23053, Directive (EU) 2019/790 (copyright) and ISO/IEC 23894.
Build it end to end
Your final project follows the real engineering workflow, step by step, reviewed by your tutor.
Where this takes you
Requirements
- University degree in engineering, science or architecture, or equivalent demonstrable professional experience.
- Internet connection and a computer capable of running the programme software.
- English sufficient to read technical standards.
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 totals 1,900 hours over 12–14 months. You set your own pace within each intake, and your tutor helps you plan the workload around your job.
You work with PyTorch, Hugging Face Transformers, OpenCLIP, LlamaIndex and Weaviate, using free, academic or evaluation versions. Your tutor indicates the free alternative wherever one exists.
A SEIUM Institute diploma, awarded on passing the continuous assessment and the Master's Thesis. It is SEIUM's own qualification, not an official state degree; its recognition depends on the programme completed and the regulations of each country.
Yes. Modules passed in a course or postgraduate diploma in the same area are recognised in the higher programme. Ask about the recognition policy when you enrol.
A university degree in engineering, science or architecture (or equivalent professional experience), a computer able to run the programme software, and English sufficient to read technical standards.
Start dates are set by SEIUM's faculty for each intake. Our admissions team will confirm the next available date when you apply.
Related programmes
Master’s in Reinforcement & Imitation Learning for Control Engineering
Online master's programme of 1,900 hours in reinforcement and imitation learning for control. 12 modules covering RL algorithms for robots, imitation and inverse reinforcement learning, demonstration…
Master’s in Traffic and Mobility Twins
Online master's programme of 1,900 hours in traffic and mobility twins: micro/macro simulation, signal control and demand. 12 modules covering traffic engineering, mobility data, demand modelling,…
Master’s in Technical NLP and Engineering Assistants (RAG)
Online master's programme of 1,900 hours in technical NLP and engineering assistants (RAG) for documentation, logs and expert support. 12 modules covering natural language in engineering,…
More in Data, AI, Simulation & Analytics
All Data, AI, Simulation & Analytics programmes → · How to apply · Fees & financing · Scholarships