Introduction to Facial Recognition (EC49191) - PDF

Introduction to Facial Recognition (EC49191) Chapter 1: Artificial Intelligence Definitions 1 Introduction to Facial Recognition ARTIFICIAL DATA ACQUISITION MACHINE LEARNING INTELLIGENCE...

Introduction to Facial Recognition (EC49191) Chapter 1: Artificial Intelligence Definitions 1 Introduction to Facial Recognition ARTIFICIAL DATA ACQUISITION MACHINE LEARNING INTELLIGENCE TECHNIQUES DEFINITIONS OPENCV (COMPUTER FACIAL RECOGNITION ETHICAL GUIDELINES VISION) PROCESS AND FOR AI APPLICATION 2 What is Artificial Intelligence? 3  Artificial Intelligence Artificial Intelligence (AI), the ability of a digital computer or computer-controlled robot to perform tasks commonly associated with intelligent beings.  Source: https://www.britannica.com/technology/artificial-intelligence 4 Source: https://www.youtube.com/watch?v=0oRVLf16CMU&t=1s Artificial Intelligence Classic Algorithm: A step-by-step procedure for solving a problem. Algorithm Explicit step-by-step instructions. Input Programmable Output Computer 5 Artificial Intelligence AI Algorithm “Horse ” Input Programmable Output Computer 6 Artificial Intelligence ? Algorithm Horse Input Human Output Image Classification 7 Artificial Intelligence ? Algorithm Horse Input Computer Output Image Classification 8  Artificial Intelligence Tasks commonly associated with intelligent beings  Image understanding  Natural language processing,  Knowledge acquisition,  Text understanding,  Planning,  Robotics,  Forecasting,  And many others. Can a general system achieve all 9 these tasks? What is Machine Learning 10  Machine Learning Machine learning, in artificial intelligence, is a discipline concerned with the implementation of computer software that can learn autonomously.  source: https://www.britannica.com/technology/ma chine-learning Source: https://www.youtube.com/watch?v=qYNweeDHiyU 11 Machine Learning 12 Machine Learning (in layman’s term) Non-machine learning way: Machine learning way: Apply some pre-defined formula. Identify the relationship from repeated success/failures. Need to reprogram for new conditions; such as wind. No need to reprogram for new 13 conditions; just need more data. Facial Detection Definition True Positive (TP) is the number of faces that are detected from the algorithm. True Negative (TN) is the number of faces that are not detected. False Positive (FP) is the number of non- faces falsely detected as faces. False Negative (FN) is the number of non-faces rejected from the classifier. 14 Machine Learning Pros: Cons: Need for data; lots of it Autonomous: learns automatically from the data The relationship learnt is complex and is not easily No need for human subject explained matter expert to determine the rules Can be easily fooled with “bad data” Superhuman performance is possible for specific tasks (e.g. AlphaGo) AI vs. Machine Learning Artificial Intelligence Machine Learning Artificial intelligence originated around Machine learning originated around 1950s. 1960s. AI represents simulated intelligence in Machine learning is the practice of machines. getting machines to make decisions without being programmed. AI is a subset of Data Science. Machine learning is a subset of AI & Data Science. Aim is to build machines which are Aim to make machines learn through capable of thinking like humans. data so that they can solve problems. 16  Remember these models? Bar Chart Line Chart Scatterplot Tree Diagram Source: datavizcatalogue.com 17 Examples of Mathematical Models Logarithmic Exponential Linear A rule based approach means that we define the relationship. A machine learning approach means that we use machines to figure out the relationship for us, given the data. 18 Common ML algorithms Source: https://www.moogsoft.com/blog/aiops/understanding-machine- learning-aiops-part-2/regression-class-clustering-graph 19  Over-fitting Model Methods to resolve model over-fitting:  Decrease number of epochs or training iteration.  Apply data augmentation.  Apply regularization.  Add dropout layer. Over-fitting model  Any other methods (link (Green line trace the below) dataset too closely) https://medium.com/ml-research-lab/under-fitting-over-fitting-and-its-solution- dc6191e34250 Good fitted model (Black line trace the dataset optimally) 20 Source: https://en.wikipedia.org/wiki/Overfitting Under-fitting Model Under-fitting model Good fitted model (Red line unable to trace the dataset) (Blue line able to trace the dataset) Source: https://en.wikipedia.org/wiki/Overfitting Methods to resolve model under-fitting:  Increase the number of epochs or training iteration.  Increase the complexity of the model.  Decrease regularization.  Remove dropout layer.  Any other methods (link below) 21 https://medium.com/ml-research-lab/under-fitting-over-fitting-and-its-solution- dc6191e34250 22