Thank you for sending your enquiry! One of our team members will contact you shortly.
Thank you for sending your booking! One of our team members will contact you shortly.
Course Outline
MATLAB Deep Learning Environment & GPU Validation
- Overview of the Deep Learning Toolbox architecture and operational workflow
- Confirming GPU availability, CUDA/cuDNN compatibility, and driver settings
- Setting up parallel workers, managing memory, and acquiring fundamental knowledge of
gpuArray - Lab 1: Validating the environment and executing an initial GPU-accelerated deep learning script
Core Deep Learning Constructs in MATLAB
- Neural network layers: conv, pooling, batch norm, dropout, residual, and dense layers
- Essentials of
dlarray,dlnetwork, and implementing custom training loops - Loss functions, optimizers (Adam, SGD, RMSProp), and strategies for learning rate scheduling
- Visualizing network structures, weight distributions, and gradient flow for effective debugging
- Lab 2: Constructing a custom
dlnetworkfrom the ground up and troubleshooting layer interactions
Designing CNNs for Image Recognition
- CNN design principles: feature extraction, spatial hierarchies, and receptive fields
- Transfer learning: utilizing pre-trained networks such as ResNet, EfficientNet, and MobileNet
- Data augmentation workflows using
imageDatastore,augmentedImageDatastore, and custom transformations - Lab 3: Training a CNN from scratch on a bespoke image classification dataset with augmentation
Automated Data Labeling & Reproducible Pipelines
- Utilizing MATLAB’s active learning and semi-supervised labeling capabilities
- Importing and exporting annotation formats (COCO, Pascal VOC, YOLO, CSV)
- Creating version-controlled, parameterized data preparation scripts
- Lab 4: Automating the labeling workflow and integrating it into a training script
Scalable Training: Multi-GPU, Cloud & Clusters
- Multi-GPU training tactics: batch size optimization, gradient accumulation, and data parallelism
- Distributed training utilizing MATLAB Parallel Server and on-premises clusters
- Cloud training workflows (AWS, Azure, GCP) via MATLAB cloud compute profiles
- Monitoring training, checkpointing, and hyperparameter optimization methods
- Lab 5: Scaling a model to a multi-GPU/cloud configuration and analyzing training throughput
Cross-Framework Interoperability & Model Exchange
- Importing pre-trained Caffe and TensorFlow/Keras models into MATLAB
- Verifying accuracy parity and adapting architectures for MATLAB workflows
- Exporting models to ONNX, TensorFlow, or Core ML for cross-platform deployment
- Lab 6: Importing a TF-Keras model, fine-tuning it within MATLAB, and exporting to ONNX
Capstone Project & Production Readiness
- End-to-end pipeline: data ingestion, training, validation, optimization, and deployment
- Model compression: pruning, quantization, and code generation using GPU Coder
- Best practices for reproducibility: logging, seeding, and sharing MATLAB deep learning apps
- Capstone: Design, train, optimize, and export a complete image recognition system tailored to your specific domain
To request a customized course outline for this training, please contact us.
Requirements
- Competency in MATLAB, including syntax, programming workflows, and toolbox usage
- No previous background in data science or deep learning is necessary
- Availability of a local GPU-enabled workstation (CUDA-compatible) or an approved cloud cluster for practical sessions
Target Audience
- Developers & Software Engineers
- Research Engineers & Domain Specialists
- Teams shifting from conventional signal/image processing to AI-driven workflows
14 Hours
Testimonials (2)
The training was organized and well-planned out, and I come out of it with systematized knowledge and a good look at topics we looked at
Magdalena - Samsung Electronics Polska Sp. z o.o.
Course - Deep Learning with TensorFlow 2
I really liked the end where we took the time to play around with CHAT GPT. The room was not set up the best for this- instead of one large table a couple of small ones so we could get into small groups and brainstorm would have helped
