About This Course
Artificial intelligence and deep learning are reshaping every sector of the global economy — from healthcare diagnostics and financial fraud detection to autonomous vehicles and natural language processing. Organisations that can harness these technologies gain a decisive competitive edge, yet the skills gap in applied deep learning remains one of the most significant barriers to adoption. TrainingComplex designed this programme to close that gap by placing participants directly at the intersection of theory and practice.
Deep Learning with TensorFlow and Keras Training is a rigorous, ten-day immersion into the world's most widely adopted open-source deep learning ecosystem. TensorFlow, developed by Google Brain, powers production AI systems at global scale, while its high-level Keras API enables rapid prototyping without sacrificing flexibility. Together they offer a complete toolkit for building, optimising, and deploying models that solve complex, high-dimensional problems.
The curriculum is structured to build competence progressively. Early modules lay a solid mathematical and programmatic foundation, ensuring that participants understand not just how to call an API, but why each architectural decision matters. Subsequent modules advance into convolutional neural networks for image intelligence, recurrent and sequence models for time-series and text, transfer learning strategies, and modern transformer-based architectures. The programme concludes with model optimisation, interpretability, and deployment pipelines — skills that bridge the gap between a working notebook and a production system.
All learning is grounded in hands-on labs using real datasets drawn from business contexts relevant to participants across Africa and the Middle East. Whether you are a data scientist seeking to formalise your deep learning knowledge, a software engineer transitioning into AI, or a technical lead responsible for scoping AI initiatives, this course provides the structured expertise and portfolio of completed projects you need to move forward with confidence.
Who Should Attend
- Data scientists and machine learning practitioners seeking to deepen their TensorFlow and Keras expertise
- Software engineers and developers transitioning into artificial intelligence and deep learning roles
- Research analysts working with large, complex datasets who need scalable modelling capabilities
- IT professionals and solutions architects responsible for designing or evaluating AI-powered systems
- Business intelligence and analytics professionals looking to extend beyond traditional statistical models
- Academics, postgraduate researchers, and technical educators building applied AI curricula
- Technical project managers and AI team leads who need hands-on understanding to oversee deep learning projects
Learning Objectives
- Explain the mathematical foundations of neural networks including forward propagation, loss functions, and backpropagation
- Build, compile, and train deep neural networks using the TensorFlow 2.x and Keras APIs
- Design convolutional neural networks for image classification, object detection, and feature extraction tasks
- Apply recurrent architectures including LSTMs and GRUs to sequential data, time-series forecasting, and text modelling
- Implement transfer learning and fine-tuning strategies using pre-trained models from TensorFlow Hub and Keras Applications
- Construct and train transformer-based models for natural language processing tasks including classification and generation
- Apply regularisation, hyperparameter tuning, and model interpretability techniques to improve generalisation
- Build end-to-end data pipelines using the TensorFlow Data API and TFRecords for efficient model training
- Deploy trained models using TensorFlow Serving, TensorFlow Lite, and REST API wrappers for production environments
- Evaluate model performance using appropriate metrics, confusion matrices, ROC curves, and cross-validation strategies
- Use TensorBoard and Keras callbacks for experiment tracking, early stopping, and learning rate scheduling
- Apply deep learning techniques to domain-specific problems in computer vision, NLP, and tabular data
- Develop awareness of AI ethics, model fairness, and responsible deployment practices
Course Modules
Module 1: Foundations of Deep Learning and the TensorFlow Ecosystem
- The AI landscape: machine learning, deep learning, and neural networks in context
- Overview of TensorFlow 2.x architecture: eager execution, graphs, and the Keras integration
- Setting up the development environment: Anaconda, Google Colab, and GPU configuration
- Tensors, tensor operations, and automatic differentiation with tf.GradientTape
- Review of linear algebra and calculus concepts underpinning neural network training
- Hands-on: building and manipulating tensors, verifying GPU availability, and running your first TensorFlow computation
Module 2: Building Neural Networks with Keras
- The Keras Sequential API: stacking layers to build feedforward networks
- Activation functions: ReLU, sigmoid, softmax, and when to use each
- Loss functions and optimisers: MSE, cross-entropy, SGD, Adam, and RMSProp
- Compiling and training a model: the fit, evaluate, and predict workflow
- Validation splits, overfitting detection, and the bias-variance trade-off
- Hands-on: training a multi-layer perceptron on the UCI Adult Income dataset
Module 3: The Keras Functional and Subclassing APIs
- Limitations of Sequential models and when to use the Functional API
- Building multi-input and multi-output models with the Functional API
- Custom layers and models using the Keras subclassing approach
- Callback functions: ModelCheckpoint, EarlyStopping, ReduceLROnPlateau
- Saving, loading, and versioning models in SavedModel and HDF5 formats
- Hands-on: constructing a dual-input model combining structured and text features
Module 4: Data Pipelines with the TensorFlow Data API
- The tf.data API: creating efficient, scalable input pipelines
- Loading data from CSV, images, and TFRecord files
- Data augmentation strategies using tf.image and Keras preprocessing layers
- Batching, shuffling, prefetching, and caching for optimal GPU utilisation
- Handling class imbalance with oversampling, undersampling, and class weights
- Hands-on: building a production-ready image data pipeline with on-the-fly augmentation
Module 5: Convolutional Neural Networks for Computer Vision
- The convolution operation: filters, feature maps, padding, and stride
- Pooling layers, batch normalisation, and dropout in CNN architectures
- Classic architectures: LeNet, VGG, ResNet, and Inception — design principles and trade-offs
- Building a custom CNN for multi-class image classification from scratch
- Evaluating vision models: accuracy, precision, recall, F1-score, and confusion matrices
- Hands-on: training a plant disease classifier on an agricultural image dataset
Module 6: Transfer Learning and Fine-Tuning
- The transfer learning paradigm: feature extraction versus full fine-tuning
- Using Keras Applications: MobileNetV2, EfficientNet, ResNet50, and InceptionV3
- Freezing and unfreezing layers: strategies for staged fine-tuning
- Domain adaptation: adapting pre-trained models to new, smaller datasets
- TensorFlow Hub: discovering and integrating community-contributed model modules
- Hands-on: fine-tuning EfficientNetB0 on a medical imaging dataset with fewer than 2 000 labelled samples
Module 7: Recurrent Neural Networks and Sequence Modelling
- The sequential data problem: why feedforward networks fall short for time-series and text
- Simple RNNs: architecture, vanishing gradient problem, and practical limitations
- LSTM networks: cell state, gates, and long-range dependency capture
- Gated Recurrent Units (GRUs): a lighter alternative with comparable performance
- Stacked and bidirectional RNNs for richer sequence representations
- Hands-on: building a univariate and multivariate LSTM for energy consumption forecasting
Module 8: Natural Language Processing with Deep Learning
- Text preprocessing: tokenisation, vocabulary building, padding, and the Keras TextVectorization layer
- Word embeddings: Word2Vec, GloVe, and trainable Embedding layers in Keras
- Sequence classification with LSTM and CNN-LSTM hybrid architectures
- Sentiment analysis and document categorisation on African-language and multilingual corpora
- Introduction to the encoder-decoder architecture and attention mechanisms
- Hands-on: building a tweet sentiment classifier and visualising learned word embeddings
Module 9: Transformer Architectures and Pre-trained Language Models
- The self-attention mechanism: queries, keys, and values explained intuitively
- The Transformer architecture: positional encoding, multi-head attention, and feed-forward sublayers
- BERT and its variants: pre-training objectives and the fine-tuning paradigm
- Using Hugging Face Transformers with TensorFlow for text classification and NER
- GPT-style models and prompt engineering basics for generative tasks
- Hands-on: fine-tuning a DistilBERT model on a domain-specific question-answering dataset
Module 10: Regularisation, Optimisation, and Hyperparameter Tuning
- Regularisation techniques: L1, L2 weight decay, dropout, and label smoothing
- Batch normalisation versus layer normalisation: when each is appropriate
- Learning rate schedules: warm-up, cosine decay, cyclical rates, and one-cycle policy
- Hyperparameter search with Keras Tuner: Random Search, Hyperband, and Bayesian Optimisation
- Mixed-precision training with tf.keras.mixed_precision for faster GPU computation
- Hands-on: running a full hyperparameter search and comparing model variants on TensorBoard
Module 11: Model Interpretability and Responsible AI
- Why interpretability matters: regulatory requirements, bias, and stakeholder trust
- Grad-CAM and Grad-CAM++: visualising what a CNN attends to during classification
- SHAP (SHapley Additive exPlanations) for deep learning model explanations
- Fairness metrics: demographic parity, equalised odds, and calibration across subgroups
- Data privacy considerations: differential privacy and federated learning concepts
- Hands-on: producing Grad-CAM explanations for a medical image classifier and writing a model card
Module 12: Deploying Deep Learning Models to Production
- The model lifecycle: training, evaluation, packaging, and serving
- TensorFlow Serving: exporting SavedModels and configuring a gRPC and REST serving endpoint
- Building a lightweight REST API wrapper with FastAPI to expose a Keras model
- TensorFlow Lite: quantisation and conversion for deployment on mobile and edge devices
- Monitoring deployed models: data drift detection, prediction logging, and retraining triggers
- Hands-on: containerising a TensorFlow Serving deployment with Docker and testing with Postman
Module 13: Capstone Project: End-to-End Deep Learning Solution
- Problem scoping: translating a business challenge into a deep learning task definition
- Dataset acquisition, exploratory data analysis, and pre-processing strategy design
- Architecture selection rationale: justifying model choices against task requirements and constraints
- Training, validation, and iterative improvement cycle with experiment tracking in TensorBoard
- Model deployment and a live demonstration of the inference endpoint or mobile prototype
- Project presentation and peer review: communicating results, limitations, and next steps to a non-technical audience
Requirements & Prerequisites
Participants should have working proficiency in Python, including familiarity with NumPy, Pandas, and Matplotlib. A basic understanding of machine learning concepts such as regression, classification, and model evaluation is strongly recommended. Prior exposure to linear algebra and calculus at an introductory level will be helpful, though a focused review is provided in the opening module.
Training Methodology
- Guided coding labs in Jupyter Notebooks where participants build and train models step-by-step on real datasets
- Structured code-along sessions led by the instructor, followed by independent exercises that reinforce each concept
- Case study analysis drawn from healthcare, finance, agriculture, and e-commerce contexts relevant to African and Middle Eastern markets
- Mini-project sprints at the end of each two-day block, culminating in a deployable model and brief peer presentation
- Debugging workshops where participants diagnose and fix deliberately broken training pipelines to build troubleshooting intuition
- Hyperparameter tuning competitions using Keras Tuner, with teams comparing model performance on a shared leaderboard
- Model interpretability exercises using Grad-CAM, SHAP values, and attention visualisations to connect predictions to business explanations
- Capstone project in the final two days: participants select a domain problem, build an end-to-end solution, and present findings to a panel
Organisational Benefits
- Accelerated development of in-house AI capabilities, reducing dependence on external consultants
- A team equipped to prototype and validate deep learning solutions rapidly, shortening time-to-insight
- Improved quality and reliability of predictive models deployed across business functions
- Stronger competitive positioning through the adoption of state-of-the-art AI architectures
- Reduced risk in AI projects through staff who understand both technical trade-offs and deployment constraints
- Better cross-functional collaboration between data science, engineering, and business stakeholders
- A foundation for building an internal AI centre of excellence and knowledge-sharing culture
- Enhanced capacity to evaluate vendor AI solutions critically and negotiate from an informed position
Where Is Deep Learning with TensorFlow and Keras Training Offered?
This course runs in-person across: Abuja, Lagos, Zanzibar, Cape Town, Dubai, Nairobi, Nakuru, Kigali, Riyadh, Mombasa, Kisumu, Accra, Addis Ababa, Naivasha, Arusha, Kampala, Pretoria, Johannesburg, Dar es Salaam, plus Live Virtual (online) cohorts. Browse live dates for your city below.
Upcoming Schedules
| Location | Format | Dates | Price |
|---|
| Live Virtual |
Online |
19 Oct 2026 – 30 Oct 2026 |
KES 160,000 |
| Live Virtual |
Online |
24 Oct 2026 – 13 Dec 2026 |
KES 160,000 |
| Abuja, Nigeria |
In-Person |
26 Oct 2026 – 4 Nov 2026 |
USD 5,600 |
| Lagos, Nigeria |
In-Person |
26 Oct 2026 – 4 Nov 2026 |
USD 5,000 |
| Zanzibar, Tanzania |
In-Person |
26 Oct 2026 – 4 Nov 2026 |
USD 5,000 |
| Cape Town, South Africa |
In-Person |
26 Oct 2026 – 4 Nov 2026 |
USD 7,600 |
| Dubai, United Arab Emirates |
In-Person |
26 Oct 2026 – 4 Nov 2026 |
USD 8,200 |
| Nairobi, Kenya |
In-Person |
26 Oct 2026 – 4 Nov 2026 |
KES 230,000 |
| Nakuru, Kenya |
In-Person |
26 Oct 2026 – 4 Nov 2026 |
KES 240,000 |
| Kigali, Rwanda |
In-Person |
26 Oct 2026 – 4 Nov 2026 |
USD 3,800 |
| Riyadh, Saudi Arabia |
In-Person |
26 Oct 2026 – 4 Nov 2026 |
USD 9,000 |
| Mombasa, Kenya |
In-Person |
26 Oct 2026 – 4 Nov 2026 |
KES 240,000 |
| Nairobi, Kenya |
In-Person |
2 Nov 2026 – 13 Nov 2026 |
KES 230,000 |
| Kigali, Rwanda |
In-Person |
2 Nov 2026 – 13 Nov 2026 |
USD 3,800 |
| Abuja, Nigeria |
In-Person |
2 Nov 2026 – 13 Nov 2026 |
USD 5,600 |
+228 more dates — view live schedule →
Frequently Asked Questions
Do I need prior deep learning experience to attend?
No prior deep learning experience is required, but participants should be comfortable coding in Python and have a basic grasp of machine learning concepts such as classification and regression. The first module provides a focused review of the mathematical foundations so that all participants start from a common baseline.
Will I receive a certificate upon completion?
Yes. Participants who attend the full programme and complete the capstone project will receive a TrainingComplex Certificate of Completion in Deep Learning with TensorFlow and Keras. This certificate can be shared on professional profiles such as LinkedIn.
What equipment do I need to bring?
Participants should bring a laptop with at least 8 GB of RAM and a modern browser. All hands-on labs are run in Google Colab, which provides free GPU access, so no specialised local hardware is required. TrainingComplex will provide login credentials for any additional cloud resources needed during the course.
Is the course suitable for someone who works primarily with tabular data rather than images or text?
Absolutely. While the programme covers computer vision and NLP in depth, the core skills — building networks, tuning models, interpreting results, and deploying to production — apply equally to tabular and structured data problems. Several labs explicitly use tabular datasets from finance and operations domains.
Can my organisation send a group and have the content customised to our industry?
Yes. TrainingComplex offers group enrolments and can tailor the case studies, datasets, and capstone project themes to your organisation's specific sector and use cases. Please contact our corporate training team to discuss a bespoke delivery arrangement.
