This course is aimed at learners who already learned the math and code of Machine Learning (ML)/Deep Learning (DL) models, and wish to practice the full stack: shipping a complete ML/DL project, from design to deployment. Learners will get first-hand experience with best R&D and industrial practices of all components of a deep learning project.
Key points that make this course special:
1) The course is advised, built, mentored, and directly taught by deep learning experts and leaders at companies and R&D centers, including COTAI, AINovation, Cinnamon, TMA Solutions, HCMUT, HCMUS, JVN, and more.
2) Learners will be introduced to a unified framework for understanding ML/DL projects and their life-cycle. Also the associated challenges and best practices, or “tricks of the trade”, for planning, setting up, and successfully delivering ML/DL projects.
3) Learners will practice developing and deploying state-of-the-art integrated AI systems with language/visual understanding and reasoning capabilities. Specifically, learners will practice using the most important building blocks of deep learning: CNNs; RNNs & sequence models; embeddings; GANs, VAEs & generative models; deep RL; attention & memory mechanisms; neural-symbolic integration with reasoning, planning, and search mechanisms.
4) Learners will understand state of the art in their domains of interest: what’s possible, what to try next, and what still needs more R&D. We will also introduce most promising research areas to help learners keep up with fast-moving research areas such as transfer, few-shot, meta learning, and disentangled representation learning.
- Working Professionals
- Science/Technology Diploma or Graduates
Students should be familiar with/ be equipted with:
- Programming (Python) (*)
- Linear Algebra, Calculus and Optimization (*)
- Logical and critical thinking (*)
- AI Specialist course (by VTC Academy)
(*) Entrance test required
|1||Introduction to planning and project setup for ML projects|
|2||Choosing machine learning project: project impact and project cost estimation|
|3||Planning and project setup: metrics, baselines, and surveying|
|4||Technical overview: recent advances in Computer Vision|
|5||Technical overview: recent advances in Natural Language Processing|
|6||Technical overview: recent advances in other fields (e.g. recommender system, data mining, etc.)|
|7||Lab instruction and guest lecture|
|8||Start course project: planning and project setup for course project|
|9||Data management for machine learning|
|10||Data-driven exploration and visualizations: tools and techniques|
|11||Data integration: entity resolution and data fusion|
|13||Data cleaning: error detection and repairing|
|14||Lab instruction and guest lecture|
|15||Model selection and construction strategies|
|16||Training strategies: optimization and regularization for deep learning|
|17||Training strategies: hyperparameter selection and experiment management|
|19||Lab instruction and guest lecture|
|20||Deploying machine learning projects|
|21||Maintaining, monitoring, and improving machine learning project|
|22||Lab instruction and guest lecture|
|23||Lab instruction and guest lecture|
|24||Lab instruction and guest lecture|
|25||Lab instruction and guest lecture|
|26||Advanced data annotating and sampling: hard sampling, online learning, and active learning|
|27||Problems with large-scale machine learning projects|
|28||Lab instruction and guest lecture|
|31||Deep generative models|
|32||Probabilistic graphical model and graph convolutional networks|
|33||Multimodal understanding and reasoning|
|34||Lab instruction and guest lecture|
|35||Lab instruction and guest lecture|
|36||Lab instruction and guest lecture|
|37||End course project|
|38||Course review and closing|
- 108 Hours, 12 Weeks, 3 sessions/week, 3 hours/session
Learning outcomes (Exit Competencies)
Upon completing this course, student should be able to:
Formulating the problem and estimating project cost. For example: how to prioritize projects & choosing goals; ii) how to choose/combine evaluation metrics; iii) how to choose appropriate baselines.
- Finding (e.g., using web crawler, social listening), cleaning, labeling, and augmenting data (data wrangling).
- Picking the right framework and compute infrastructure. Writing modular, hackable, and scalable code in major deep learning frameworks e.g. Tensorflow 2.0, PyTorch, Keras, etc.
- Training models in parallel to find the right model architecture. Set up continuous integration system for your codebase, which will check functionality of code and evaluate the model about to be deployed. Troubleshooting training and ensuring reproducibility.
- Deploying the model at scale. For example: package up the learned model as a REST API or deployable as a Docker container; deploy the learned model as a serverless function to Amazon Lambda.
- Setting up mechanisms that alert AI developers when new data is sufficiently available for continual training, or when the incoming data distribution changes that needs to retrain the models.