Model Online / On-Site
Duration
Modules 5 modules

Course Objective & Scope

Most modern deep learning architectures, computer vision models and large language models are built on PyTorch. Rather than relying on black-box scripts, this course equips participants with a solid command of PyTorch's core building blocks—tensor algebra, dynamic computation graphs and modular network design—so they can build, debug and train models from scratch.

Participants manipulate tensors across CPU and GPU, compute gradients with autograd, write custom Dataset and DataLoader pipelines, implement neural networks with nn.Module, code complete training and validation loops, and save checkpoints via state_dict. They leave with a clean, reusable PyTorch training template.

Who Should Attend?

  • Data scientists moving from classical machine learning (scikit-learn) into deep learning
  • Software engineers who need to write, read and debug model training and fine-tuning code
  • Technical teams building a foundation for advanced computer vision, NLP and LLM projects
  • Prerequisite: basic Python programming and foundational linear algebra/probability

Key Takeaways

  • Create and reshape tensors, apply broadcasting and manage CPU-to-GPU (CUDA) memory transfers cleanly
  • Control the dynamic computation graph (autograd), backward() passes and torch.no_grad() blocks
  • Build custom Dataset and DataLoader pipelines for tabular, image or text data
  • Construct neural network architectures using nn.Module, loss functions and torch.optim
  • Track overfitting, save and reload model state_dict checkpoints and apply basic torch.compile acceleration

Curriculum

Tensors, Memory Layout and GPU (CUDA) Operations
  • Creating tensors, data types (float32, bfloat16), indexing and broadcasting rules
  • NumPy interoperability, contiguous memory and managing CPU/GPU placement with .to(device)
Dynamic Computation Graphs and Autograd
  • requires_grad, forward passes and automatic differentiation with loss.backward()
  • Gradient accumulation, optimizer.zero_grad() and torch.inference_mode()
Data Loading: Datasets, DataLoaders and Transforms
  • Wrapping custom data sources with torch.utils.data.Dataset
  • Batching, shuffling, num_workers, pin_memory and custom collate_fn in DataLoader
Building Neural Networks with nn.Module
  • nn.Linear, nn.Conv2d, nn.Embedding, activations, Dropout and BatchNorm/LayerNorm
  • Loss functions (CrossEntropyLoss, MSELoss) and AdamW with learning-rate schedulers
The Training Loop, Evaluation and Saving Checkpoints
  • model.train() vs model.eval(), validation metrics, early stopping and TensorBoard logging
  • Saving and loading model/optimizer checkpoints via state_dict and an introduction to torch.compile

Other Courses in This Category

  • TensorFlow and Keras Fundamentals for Deep Learning

    Deep learning fundamentals with TensorFlow 2 and Keras: tf.Tensor, GradientTape, tf.data pipelines, building models with Sequential/Functional/Subclassing APIs, callbacks, TensorBoard and SavedModel. For data scientists and engineers.

    Programming for AI Corporate
    Live Course
    5 modules
    View
  • R Programming for Data Science and Statistical Modelling

    R programming for data science, statistical analysis and machine learning: tidyverse (dplyr, tidyr, ggplot2), hypothesis testing and regression, ML workflows with tidymodels, Quarto reporting and Plumber APIs.

    Programming for AI Corporate
    Live Course
    5 modules
    View
  • Python for Data Science Python for Data Science

    Hands-on Python training covering data manipulation with Pandas and Polars, visualization with Matplotlib, Seaborn, and Plotly, exploratory data analysis (EDA), and machine learning modeling with Scikit-learn.

    Programming for AI Corporate, Individual
    Live Course
    4 modules
    View
  • Modern C++ Fundamentals and High-Performance AI Applications

    Modern C++ (C++17/20) for AI and high-performance systems: RAII and smart pointers, STL, move semantics, multithreading, SIMD, building Python extensions with pybind11 and the LibTorch/ONNX Runtime C++ APIs.

    Programming for AI Corporate
    Live Course
    5 modules
    View
  • Python Programming Python Programming

    Master Python from scratch: data structures, control flow, functional programming, Object-Oriented Programming (OOP), error handling, file I/O, virtual environments, and Pandas.

    Programming for AI Corporate, Individual
    Live Course
    9 modules
    View
  • Advanced Python for AI and Data Engineering

    Advanced Python for AI and data engineering: type hinting, decorators, memory-efficient generators, asyncio, multiprocessing and the GIL, Cython/Numba acceleration, profiling and packaging. For developers and data scientists.

    Programming for AI Corporate
    Live Course
    5 modules
    View

Get in touch