Pattern Recognition Lecture 1 PDF 2024

Introduction Course: Pattern Recognition Lecture 1 Instructor: Nada Osman [email protected] Office hours: Tuesday 10:30 – 12:00 Credit: Prof. Marwan Torki Materials 2024  Course Grades ▪ Coursework: 60% ▪ Midte...

Introduction Course: Pattern Recognition Lecture 1 Instructor: Nada Osman [email protected] Office hours: Tuesday 10:30 – 12:00 Credit: Prof. Marwan Torki Materials 2024  Course Grades ▪ Coursework: 60% ▪ Midterm: 10% ▪ Sheets: 10% ▪ Assignments: 40% ▪ Final: 40% Pattern Recognition - What?  Pattern Recognition - What? https://www.freepik.com/premium-vector/sound-signal-absract-digital-record- voice-graph_29316016.htm  Pattern Recognition - What? https://www.freepik.com/premium-vector/sound-signal-absract-digital-record- voice-graph_29316016.htm  Pattern Recognition - What? https://www.freepik.com/premium-vector/sound-signal-absract-digital-record- voice-graph_29316016.htm  Pattern Recognition - What? https://www.freepik.com/premium-vector/sound-signal-absract-digital-record- voice-graph_29316016.htm https://deeplobe.ai/exploring-object-detection-applications-and- benefits/  Pattern Recognition - What? https://www.freepik.com/premium-vector/sound-signal-absract-digital-record- voice-graph_29316016.htm https://deeplobe.ai/exploring-object-detection-applications-and- https://research.open.ac.uk/news/ou-experts-ask- benefits/ whats-face Pattern Recognition – Overview & History Pattern Recognition – Overview & History Artificial Intelligence (AI) 1950s Pattern Recognition – Overview & History Artificial Intelligence (AI) 1950s Machine Learning (ML) 1980s Supervised Unsupervised Semi-Supervised Reinforcement Pattern Recognition – Overview & History Artificial Intelligence (AI) 1950s Machine Learning (ML) 1980s Supervised Unsupervised Semi-Supervised Reinforcement Neural Networks (NN) Pattern Recognition – Overview & History Artificial Intelligence (AI) 1950s Machine Learning (ML) 1980s Supervised Unsupervised Semi-Supervised Reinforcement Neural Networks (NN) Deep Learning (DL) 2010s Pattern Recognition – Overview & History Artificial Intelligence (AI) 1950s Machine Learning (ML) 1980s Supervised Unsupervised Semi-Supervised Reinforcement Neural Networks (NN) Deep Learning (DL) Generative AI 2010s LLMs DALL-E Stable GPT Diffusion 2020s Pattern Recognition - Pipeline  Pattern Recognition - Pipeline Learning Process (Training) Inference Data Model (Testing)  Pattern Recognition - Pipeline Learning Process (Training) Inference Data Model (Testing) Structured 𝒙𝟏 𝒙𝟐 𝒚 5.1 7.4 0 8.0 1.2 0 1.6 6.3 0 0.1 1.1 1  Pattern Recognition - Pipeline Learning Process (Training) Inference Data Model (Testing) Structured Unstructured 𝒙𝟏 𝒙𝟐 𝒚 5.1 7.4 0 8.0 1.2 0 1.6 6.3 0 0.1 1.1 1  Pattern Recognition - Pipeline Learning Process (Training) Inference Data Model (Testing)  Pattern Recognition - Pipeline Learning Process (Training) Inference Data Model (Testing) Linear Models Neural Networks Fully Connected Convolution Recurrent Transformer  Course Content ▪ Lecture 1: Learning Types & Learning Behaviours ▪ Lecture 2: Linear Classification & Linear Regression ▪ Lectures 3-6: Perceptron & Neural Networks ▪ Lectures 7: Convolutional Neural Network Midterm ▪ Lectures 8: Convolutional Neural Network – Con. ▪ Lectures 9-10: Recurrent Neural Network ▪ Lectures 11-12: Attention & Transformers  Course Content ▪ Lecture 1: Learning Types & Learning Behaviours ▪ Lecture 2: Linear Classification & Linear Regression ▪ Lectures 3-6: Perceptron & Neural Networks ▪ Lectures 7: Convolutional Neural Network Midterm ▪ Lectures 8: Convolutional Neural Network – Con. ▪ Lectures 9-10: Recurrent Neural Network ▪ Lectures 11-12: Attention & Transformers Learning Types  Learning Types Supervised Unsupervised Semi-supervised Reinforcement  Learning Types Supervised Unsupervised Semi-Supervised Reinforcement  Learning Types Supervised Unsupervised Semi-Supervised Reinforcement Learning Process (Training) Model Labelled Data Labels Each data point is accompanied with the Cat Dog ground truth label during the training  Learning Types Supervised Unsupervised Semi-Supervised Reinforcement  Learning Types Supervised Unsupervised Semi-Supervised Reinforcement Learning Process (Training) Learnt Categories Model Unlabelled Data Data is not accompanied with ground truth labels  Learning Types Supervised Unsupervised Semi-Supervised Reinforcement  Learning Types Supervised Unsupervised Semi-Supervised Reinforcement Learning Process (Training) Dog Well-Known approach is to Cat train the model on the labelled Model data first → predict pseudo labels of the unlabelled data → retrain the model Semi-labelled Data Part of the data is labelled while the reset are not  Learning Types Supervised Unsupervised Semi-Supervised Reinforcement Weakly-Supervised  Learning Types Supervised Unsupervised Semi-Supervised Reinforcement Weakly-Supervised Dog Learning Process (Training) Cat Model Dog The task is to predict the location of the cat or the Cat dog in the image. The image is labelled , , while the ground truth location is not given Weakly-labelled Data  Learning Types Supervised Unsupervised Semi-Supervised Reinforcement  Learning Types Supervised Unsupervised Semi-Supervised Reinforcement Learning Process (Training) Reward Action Agent (Model) Environment Learning Behaviours  Learning Behaviours Underfit  Learning Behaviours Underfit Uncapable Model Noisy Data Imbalanced Data Non-representative Data Small amount of data  Learning Behaviours Underfit Good fit  Learning Behaviours Underfit Good fit Good Model Adequate Data  Learning Behaviours Underfit Good fit Overfit  Learning Behaviours Underfit Good fit Overfit Complex Model Easy Task Small Data Thanks for your attention! Ask Questions? [email protected] Office hours: Tuesday 10:30 – 12:00 2024