[English speaking lesson] Buổi 2 lớp FIT-LAB Spring 2026 về AI, ML, DL, CV, NLP và LLM tại NEU

Key Concepts

• Machine Learning Fundamentals: Understanding models, data interaction, loss functions, and the iterative process of learning.
• Supervised Learning: Utilizing labeled data for model training, specifically in classification and regression tasks.
• Loss Functions: Quantifying model performance and guiding optimization (MAE, MSE, Cross-Entropy).
• Model Evaluation: Assessing model performance using validation and test sets, and employing techniques like K-fold cross-validation.
• Softmax & One-Hot Encoding: Transforming raw model outputs into probability distributions and representing categorical data for loss calculation.
• Importance of Understanding ‘Why’: Prioritizing conceptual understanding over rote memorization of implementation details.
• English Proficiency in Tech: Recognizing the necessity of strong English skills for computer science professionals.

Course Introduction & Machine Learning Basics (Part 1)

The course aims to teach machine learning, deep learning, and AI in English, simultaneously developing both technical expertise and English language skills (listening and speaking). The instructor acknowledges potential language barriers and offers flexibility, switching to Vietnamese if needed, but emphasizes the importance of English proficiency for future career opportunities, including international collaboration and further education. Machine learning is presented as an analogy to human learning: receiving exercises (data), comparing to solutions (model prediction vs. actual value), receiving feedback (loss value), and iteratively improving. A model interacts with data, and the goal in supervised learning is to minimize the loss function. Two common loss functions for regression – Mean Absolute Error (MAE) and Mean Squared Error (MSE) – were introduced, with a discussion of how MSE is more sensitive to outliers. The distinction between classification (countable outputs, e.g., spam detection) and regression (uncountable outputs, e.g., house price prediction, crypto price prediction, predicting patient condition) was also reviewed. Examples included house price prediction as a recurring regression example, and spam email detection as a classification example. The instructor stressed that the interpretation of a loss value is context-dependent (e.g., a 5 million VND loss is acceptable for house prices, but not for chicken prices). The Softmax function was briefly introduced as a method to convert model outputs into probabilities for classification.

Deep Dive into Loss Functions & Model Evaluation (Part 2)

Building on the foundation laid in Part 1, the discussion focused on the intricacies of model evaluation and loss functions. The softmax function was explained in detail: it uses the exponential of each prediction (e^x) and normalizes it by dividing by the sum of all exponentials, ensuring the output is a probability distribution summing to one. The instructor highlighted the importance of understanding why the exponential function is used – to prevent negative probabilities. The concept of one-hot vectors was introduced as a way to represent ground truth labels for loss calculation, emphasizing its importance in IT job interviews. The cross-entropy loss formula (- Σ t_i * log(p_i)) was presented, where t_i is the true label and p_i is the predicted probability. The importance of model confidence (probabilities) beyond just correct/incorrect predictions was stressed.

The segment then transitioned to the crucial topic of data splitting. Data is divided into three mutually exclusive sets: a training set (for learning), a validation set (for evaluating performance during training and preventing overfitting), and a test set (for a final, unbiased evaluation). To improve the reliability of validation, K-fold cross-validation was introduced, where the data is divided into K folds, and the model is trained K times, each time using a different fold for validation. This provides a more robust estimate of generalization ability.

Conclusion

This course segment provided a comprehensive introduction to the core concepts of machine learning, emphasizing the importance of both technical understanding and practical language skills. The focus on why methods are used, rather than just how to implement them, sets a strong foundation for deeper learning. The discussion of loss functions, model evaluation techniques, and data splitting strategies highlights the critical steps involved in building and deploying effective machine learning models. Ultimately, the instructor conveyed that successful machine learning requires a blend of mathematical understanding, practical implementation, and the ability to communicate effectively – skills that this course aims to cultivate.