AI-ML-Virtual-Internship on new technology
- 1. AI-ML Virtual Internship Summary This report summarizes an AI-ML virtual internship completed by Y. Prasanth. It covers machine learning fundamentals, implementing ML pipelines, forecasting, and computer vision. by P.Dinakar
- 2. 1.)Introduction to Machine Learning Machine learning is a subset of AI focused on using data to train models for predictions. It includes deep learning, which uses neural networks inspired by human biology. AI Broad field of building machines to perform human tasks. Machine Learning Uses data to train models for making predictions. Deep Learning Uses neural networks inspired by human biology.
- 3. Business Problems Solved with ML Machine learning solves various business problems like spam filtering, product recommendations, and credit card fraud detection. These applications use trained models to make predictions. 1 Spam Filtering ML models trained on examples of spam and regular emails. 2 Product Recommendations Predicts products based on user behavior and purchase history. 3 Fraud Detection Identifies fraudulent transactions using patterns from historical data.
- 4. Machine Learning Process The ML process involves problem formulation, data preparation, feature engineering, model training, evaluation, and deployment. It's an iterative process requiring multiple passes to achieve optimal results. 1 Problem Formulation Define the business problem and convert it to an ML problem. 2 Data Preparation Extract, clean, and preprocess data from various sources. 3 Feature Engineering Select or create relevant features for training the model. 4 Model Training Train the model using prepared data and selected algorithm. 5 Evaluation Assess model performance using metrics and test data. 6 Deployment Deploy the model to deliver predictions in production.
- 5. 2.)Implementing ML Pipeline with Amazon SageMaker Amazon SageMaker facilitates ML pipeline implementation. It includes data collection, evaluation, feature engineering, training, hosting, and accuracy assessment. The platform manages infrastructure and integrates with AWS services. Data Collection Import and secure data from various sources. Data Evaluation Assess data quality and consistency. Feature Engineering Select and create relevant features for training. Model Training Train the model using SageMaker's algorithms. Model Hosting Deploy and manage the model in production.
- 6. Amazon SageMaker Definition: Fully managed machine learning (ML) service by AWS that allows developers and data scientists to build, train, and deploy machine learning models at scale. Main Components: : •SageMaker Studio: Integrated development environment (IDE) for ML. •SageMaker Ground Truth: Tool for creating high-quality training datasets with human labelers and automated workflows. •SageMaker Autopilot: Automates the process of building and training ML models. •SageMaker JumpStart: Pre-built models and algorithms for rapid deployment. •SageMaker Model Monitor: Monitors deployed models to detect anomalies or drift. Model Training: •Supports training with managed infrastructure, including distributed training. •Built-in Algorithms: Includes algorithms like XGBoost, Linear Learner, and more. •Supports bring-your-own-algorithm for custom model training.
- 7. 3.)Introducing Forecasting Forecasting predicts future outcomes based on historical data. It involves time series analysis, handling missing data, and choosing appropriate algorithms. Amazon Forecast simplifies this process. Time Series Patterns • Trend • Seasonal • Cyclical • Irregular Handling Missing Data • Forward fill • Moving average • Backward fill • Interpolation Amazon Forecast Steps 1. Import data 2. Train predictor 3. Generate forecasts
- 8. •Steps in Forecasting: •Data collection and pre-processing. •Exploratory data analysis. •Feature engineering and selection. •Model selection and training. •Evaluation using metrics like MAE, RMSE. •Hyperparameter tuning for accuracy. •Evaluation Metrics: •Root Mean Squared Error (RMSE) •Mean Absolute Percentage Error (MAPE) •Mean Absolute Error (MAE) Forecasting
- 9. Processing Time Series Data Time series data processing involves handling missing values, down sampling, and correlation analysis. Pandas library is useful for these tasks, offering functions for resampling and autocorrelation. Timestamp Handling Ensure consistent timestamp formats and handle time zones correctly. Down Sampling Convert finely-grained time data to less granular formats as needed. Correlation Analysis Identify relationships between time series, but be cautious of spurious correlations.
- 10. 4.)Computer Vision Overview Computer vision enables machines to extract information from digital images. It has applications in public safety, authentication, content management, autonomous driving, medical imaging, and manufacturing. Autonomous Driving Computer vision enables safer self- driving car navigation. Medical Imaging AI-assisted medical image analysis improves diagnosis accuracy and speed. Manufacturing Computer vision improves quality assurance in manufacturing processes.
- 11. Key Techniques: •Image Classification: Identifying objects or patterns in images. •Object Detection: Locating and identifying multiple objects within an image. •Image Segmentation: Dividing an image into regions or objects for analysis. •Facial Recognition: Identifying or verifying a person’s identity using facial features. •Optical Character Recognition (OCR): Converting images of text into machine-readable text. Common Applications: •Object detection •Image classification •Facial recognition •Medical imaging (X-rays, MRIs) •Autonomous vehicles
- 12. Analyzing Images and Videos Amazon Rekognition is a managed service for integrating image and video analysis into applications. It provides APIs for object detection, facial recognition, and scene analysis. API Function Object Detection Identifies objects in images Facial Recognition Detects and analyzes faces Scene Analysis Categorizes image content Text Detection Extracts text from images
- 13. Preparing Custom Datasets for Computer Vision Custom datasets for computer vision require proper labeling. For image classification, labels are assigned to entire images. For object detection, bounding boxes are used to identify object locations. 1 Image Classification Assign labels to entire images for categorization tasks. 2 Object Detection Use bounding boxes to identify object locations within images. 3 Manifest File Contains metadata about labeled images and bounding boxes.
- 14. •OpenCV: Open-source computer vision library. •TensorFlow/Keras: Frameworks for building CV models. •PyTorch: Deep learning library for CV tasks. •Detectron2: Facebook's framework for object detection. Tools and Libraries: Popular Models and Architectures: •Convolutional Neural Networks (CNNs): Backbone of most computer vision tasks. •YOLO (You Only Look Once): Real-time object detection. •ResNet: Deep neural network architecture for image classification. •Faster R-CNN: For object detection. •U-Net: For medical image segmentation
- 15. Conclusion Completing an AI/ML internship offers invaluable hands-on experience in developing and deploying machine learning models, working with large datasets, and leveraging AI frameworks and tools. Interns gain practical skills in data preprocessing, model training, and evaluation, while deepening their understanding of AI concepts such as deep learning, natural language processing, and computer vision. This experience not only enhances technical expertise but also cultivates problem-solving, teamwork, and the ability to apply theoretical knowledge to real-world challenges. Ultimately, the internship lays a strong foundation for a career in the rapidly evolving field of AI/ML. Additionally, an AI/ML internship helps interns build a professional network, connect with industry experts, and stay updated with cutting-edge trends and technologies, which can significantly enhance future career opportunities in the field.