Postgraduate Diploma in Lightweight Models and Compression for Edge
Online 800-hour Postgraduate Diploma in Lightweight Models and Compression for Edge. 8 modules covering Regulation (EU) 2024/1689, ISO/IEC 22989 and PyTorch. 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
Analyse the computing, memory and energy constraints of edge devices.
Design efficient neural architectures for embedded inference.
Apply quantisation, pruning and knowledge distillation.
Use compilers and runtimes such as TensorRT, OpenVINO and ONNX Runtime.
Select the appropriate edge hardware for each industrial application.
Validate the accuracy and latency of compressed models.
Document models in line with the requirements of the AI Act.
Who it's for
- Embedded machine learning engineer looking to specialise in Lightweight Models and Compression for Edge.
- Edge computer vision developer looking to specialise in Lightweight Models and Compression for Edge.
- Model optimisation engineer looking to specialise in Lightweight Models and Compression for Edge.
- Embedded systems engineer looking to specialise in Lightweight Models and Compression for Edge.
- TinyML specialist looking to specialise in Lightweight Models and Compression for Edge.
- Researcher in neural network efficiency looking to specialise in Lightweight Models and Compression for Edge.
- Engineering or science graduates looking to enter industrial manufacturing.
What you'll work with
- Regulation (EU) 2024/1689
- ISO/IEC 22989
- ISO/IEC 23053
- ISO/IEC 42001
- ISO/IEC 25059
- IEC 62443-4-2
8 modules · 800 h
1.1Memory, computing and energy constraints1.2MobileNet and EfficientNet architectures1.3Depthwise separable convolutions1.4SqueezeNet and ShuffleNet models1.5Lightweight transformers for vision1.6FLOPs, parameters and latency metrics1.7Profiling models on hardware1.8MLPerf Tiny benchmark comparison1.9Embedded real-time computer vision1.10Practical: profiling a CNN on an embedded board
2.1Post-training quantisation (PTQ)2.2Quantisation-aware training2.3INT8 and INT4 integer quantisation2.4Calibration with representative datasets2.5Per-channel and per-tensor quantisation2.6Mixed-precision inference2.7Accuracy loss and its mitigation2.8Quantisation in TensorFlow Lite2.9Quantisation of object detection models2.10Laboratory: INT8 quantisation of a classifier
3.1Unstructured magnitude pruning3.2Structured pruning of filters and channels3.3The lottery ticket hypothesis3.4Iterative pruning with retraining3.5Structured 2:4 sparsity3.6Pruning tools in PyTorch3.7The real effect of pruning on latency3.8Combining pruning and quantisation3.9Sensitivity-guided layer pruning3.10Workshop: structured pruning of a detection network
4.1Teacher-student knowledge distillation4.2Intermediate feature distillation4.3Low-rank factorisation4.4Neural architecture search (NAS)4.5Hardware-aware NAS4.6Weight sharing and Huffman coding4.7Adaptive-size models4.8Selecting the accuracy-cost trade-off4.9Selecting teacher and student architectures4.10Practical: distilling an inspection model
5.1ONNX format and interoperability5.2TensorFlow Lite and LiteRT5.3ONNX Runtime and execution providers5.4Optimisation with NVIDIA TensorRT5.5Apache TVM and optimised compilation5.6OpenVINO for CPU and VPU5.7Operator and graph fusion5.8Memory management during inference5.9Benchmarking runtimes on the same platform5.10Laboratory: conversion and optimisation with TensorRT
6.1Microcontrollers with TinyML6.2NPUs and dedicated accelerators6.3NVIDIA Jetson embedded GPUs6.4FPGAs for inference6.5Energy consumption and heat dissipation6.6Industrial robustness of hardware6.7Platform selection by application6.8Benchmarking on real hardware6.9Start-up and reliability of field devices6.10Case study: hardware selection for inline vision
7.1Accuracy metrics after compression7.2Robustness to domain shifts7.3Model regression testing7.4Latency percentiles and jitter7.5Numerical verification across platforms7.6Explainability in lightweight models7.7Requirements of the AI Act7.8Model documentation7.9Monitoring deployed models on the factory floor7.10Case study: validating a defect detector at the edge
8.1Definition of the industrial use case8.2Training the baseline model8.3Selecting compression techniques8.4Applying quantisation and pruning8.5Compilation for the target hardware8.6Measuring latency and consumption8.7Accuracy validation8.8Analysis of results8.9Report and presentation8.10Final review with the tutor
How you'll learn
Fully 100% online training on the SEIUM campus, with downloadable material, recorded classes and an assigned technical tutor. No fixed timetable: students progress at their own pace within the intake, making it easy to combine with work.
Each module combines theory, guided reading of the original standard and applied exercises. The working software is PyTorch, TensorFlow Lite, NVIDIA TensorRT, OpenVINO, ONNX Runtime, in free, academic or evaluation versions. Tests are studied using real deployment procedures for defect detectors on NVIDIA Jetson, vibration classification on microcontrollers, and latency benchmarking on accelerators. Assessment is continuous 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.
Analyse the computing, memory and energy constraints of edge devices.
Design efficient neural architectures for embedded inference.
Apply quantisation, pruning and knowledge distillation.
Use compilers and runtimes such as TensorRT, OpenVINO and ONNX Runtime.
Select the appropriate edge hardware for each industrial application.
Where this takes you
Requirements
- Prior technical training or professional experience in the sector. No university degree is required.
- Internet connection and a computer capable of running the programme's software.
- Sufficient 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 totals 800 hours, spread over 6 to 8 months. That is an average of 10 to 15 hours per week, compatible with a full-time job.
You will work with PyTorch, TensorFlow Lite, NVIDIA TensorRT, OpenVINO and ONNX Runtime. Free, academic or evaluation versions are used, and the tutor indicates the open-source alternative where one exists.
SEIUM's own diploma upon passing the continuous assessment and the final integrative project. It is in-house training, not an official qualification from the Ministry.
Yes. Modules passed in a course or postgraduate diploma in the same area can be credited towards the advanced programme. Check the credit-transfer policy when you enrol.
Related programmes
Postgraduate Diploma in Automated Picking and Vision-Guided Robotics
Online 800-hour Postgraduate Diploma in Automated Picking and Vision-Guided Robotics. 8 modules covering ISO 10218-2:2025, EMVA 1288 Release 4.0 and OpenCV. Assigned tutor and open enrolment.
Postgraduate Diploma in Mission Optimisation and Load Balancing
Online 800 h Postgraduate Diploma in Mission Optimisation and Load Balancing. 8 modules with VDA 5050 v2.1, ISO 3691-4:2023 and Google OR-Tools. Dedicated tutor and open…
Diploma in AMR for Last Mile and Micro-Fulfillment
Online postgraduate diploma of 800 hours in AMRs for last-mile delivery and micro-fulfilment: 8 modules covering urban logistics, autonomous mobile robots, localisation and navigation, automated micro-fulfilment,…
More in Robotics, Automation & Control
All Robotics, Automation & Control programmes → · How to apply · Fees & financing · Scholarships