AI/ML Fundamentals to advanced Slides by GDG Amrita Mysuru.pdf
- 1. Tech Winter Break: AI/ML Session Amrita Mysore Lasya S Charles Jeyaseelan
- 2. Today’s Agenda 1. Introduction to AI/ML 2. Types of Machine Learning 3. Introduction to Python for ML 4. Building a Simple ML Model 5. Feature Engineering and Neural Networks 6. Recap & QA
- 4. INTRO : What is AI and ML ? 1. AI refers to the simulation of human intelligence in machines that can perform tasks requiring human-like cognitive abilities. 2. ML is a subset of AI that enables machines to learn from data without being explicitly programmed.
- 6. Real World Applications Healthcare ,Finance Retail and E-commerce Transportation Manufacturing Education Agriculture Energy
- 9. What is the difference between labelled and unlabelled data ? Sources: https://www.kaggle.com/discussions/general/187644
- 10. TYPES OF DATA
- 12. Basic key concepts and fundamental steps Gather Data Prepare Data Choose Model Split data Test Model Train Model Build,fit the Model Evaluate Model Prediction
- 13. Let’s Implement
- 14. BUILDING A SIMPLE ML MODEL Use Case: A Simple Handwritten Digit Classification <Supervised Learning> <Multi-Class Classification>
- 15. Overview What is a Model? ● A mathematical representation of a process that learns from data. Steps to Build a Model: ● Data Preprocessing: Prepare data for analysis. ● Training: Teach the model using labeled data. ● Testing: Evaluate how well the model performs on unseen data.
- 16. Training and Evaluating a Model Dataset Used: ● DIGITS Dataset ● https://scikit-learn.org/1.5/modules/generated/sklearn. datasets.load_digits.html Steps in Practice: ● Split data into Training and Testing Sets. ● Select a Model (Logistic Regression). ● Train the model and make predictions. ● Evaluate performance using Accuracy or a Confusion Matrix.
- 17. Why Evaluation Matters? Helps identify: ● Underfitting (model performs poorly on training data). ● Overfitting (model performs well on training but poorly on testing). Metric Examples: ● Accuracy = Correct Predictions / Total Predictions. ● Confusion Matrix: A table to evaluate model performance. ● Precision, Recall, F1- Score (CLASSIFICATION) ● MAE, MSE, MAPE (REGRESSION) ● Silhouette Score, Jaccard Index (CLUSTERING)
- 19. LET’s IMPLEMENT
- 20. Feature Engineering and Neural Networks
- 21. Overview What is Feature Engineering? ● Process of selecting, modifying, or creating variables to improve model performance. Key Techniques: ● Handling Missing Data: Use mean, median, or imputation. ● Scaling and Normalization: Transform data for consistent scale (e.g., Min-Max Scaling). ● Encoding Categorical Data: Use One-Hot Encoding or Label Encoding.
- 22. Example of Feature Engineering (For MNIST Dataset) Dataset Example: Columns: Pixel Intensity Values (28x28), Target Label (Digit 0-9). https://www.tensorflow.org/datasets/catalog/mnist Apply Techniques: ● Scale Pixel Intensity Values to a 0-1 range. ● Flatten Images: Convert 28x28 pixel grids to a 1D array (784 values). ● Handle Corrupted Data: Replace missing pixel values with the average pixel intensity (if applicable).
- 23. Neural Networks – Basics What is a Neural Network? ● A collection of interconnected layers inspired by the human brain. Key Components: ● Input Layer: Accepts features. ● Hidden Layers: Process data with weights and biases. ● Output Layer: Generates predictions. Core Concepts: ● Activation Functions: Sigmoid, ReLU, Softmax. ● Training: Use gradient descent and backpropagation.
- 24. Comparison With Our Human Brain Source: https://towardsdatascience.com/the-differences-between-artificial-and-biological-neural-n etworks-a8b46db828b7
- 25. What happens inside a Single Artificial Neuron?
- 26. Activation Functions What are Activation Functions? ● Functions applied to neuron outputs to introduce non-linearity. ● Help neural networks model complex relationships in data. Common Activation Functions: ● ReLU (Rectified Linear Unit): ○ Outputs x if x > 0; otherwise 0. ○ Efficient and reduces vanishing gradient issues. ● Sigmoid: ○ Squashes values between 0 and 1. ○ Used for probability predictions. ● Softmax: ○ Converts outputs into probability distributions for multi-class classification. Sources: https://medium.com/@krishnakalyan3/introduction-to-exponential-linear-unit-d3e29 04b366c https://developer.apple.com/documentation/accelerate/bnnsactivationfunction/2915 301-softmax
- 27. Optimization What is Optimization? ● The process of adjusting weights and biases to minimize the loss/cost function. Common Optimization Techniques: ● Gradient Descent: ○ Updates weights by calculating the gradient of the loss function. Variants of Gradient Descent: ● SGD (Stochastic Gradient Descent): ○ Updates weights for each training example. ● Adam (Adaptive Moment Estimation): ○ Combines momentum and adaptive learning rates for faster convergence. Sources: https://www.analyticsvidhya.com/blog/2020/10/how-does-the-gradient-de scent-algorithm-work-in-machine-learning/
- 29. LET’s IMPLEMENT
- 30. Recap
- 31. Recap of Key Concepts What We Learned: ● How to build a basic ML model. ● Importance of feature engineering for better model performance. ● Neural network basics and their components.
- 32. Self-Study Recommendations Explore: ● Datasets: Titanic dataset, MNIST digits. ● Courses: Andrew Ng’s ML course, TensorFlow tutorials. ● Frameworks: TensorFlow, Keras, PyTorch. Try building: ● Regression models for house prices. ● Neural network for binary classification.
- 33. Any Questions?
- 34. THANK YOU