ITCT101 Lecture 2.2 - AI and ML Applications PDF

ITCT101 Computer Technologies Module 2: AI, ML, and Data Science AI & ML Applications Thanapon Noraset (Nor) [email protected] Agenda - AI and ML Applications in your everyday life - Common AI/ML Problems - Ethical and other considerations AI methods are applied in many...

ITCT101 Computer Technologies Module 2: AI, ML, and Data Science AI & ML Applications Thanapon Noraset (Nor) [email protected] Agenda - AI and ML Applications in your everyday life - Common AI/ML Problems - Ethical and other considerations AI methods are applied in many fields and industries AI methods are applied in many fields and industries Algorithms behind AI/ML Applications Search: Google Maps Pathfinding Search in other applications Classification: Spam mail filter Classification in other applications  Types of Classification Errors PREDICTED - True Positives (TP): Hit P N - True Negative (TN): Correct rejection TP FN ACTUAL P - False Positive (FP): False alarm or Type I error - False Negative (FN): Miss or FP TN N Type II error Questions: Cancer screening vs Spam mail filter Recommendation: Netflix The two approaches of recommendation Regression: Keypoint Detection https://learnopencv.com/human-action-recognition-using-detectron2-and-lstm/  Clustering: Customer Segmentation https://medium.com/@ugursavci/step-by-step-customer-segmentation-using-k-means-and-pca-in-python-57338 22295b6 Large Language Models https://www.adventuresincre.com/chatgpt-bard-llama-2023/ Large language models are technology behind all modern chatbots and other applications today, including some creative industries: Automate Visualization of Data: LIDA Image Generation with DALL-E 3: DALL-E 3 How LLMs Work? A language model learns to predict the missing word (often the right most word). It uses a deep learning architecture named “Transformer” (2017). It learns from articles, books, forums, and even programming codes → billions for words in many languages Large Language Models Image from https://developer.nvidia.com/blog/scaling-language-model-training-to-a-trillion-parameters-using-megatron/ Why Large? (Brown et al., 2020 ) Large language models are few-shot learners (in-context learning) Ethical and other considerations The Trolley Problem What would you do? 1. Do nothing, the trolley kills five people. 2. Pull the lever, the trolley kills one person. Ethical and other considerations Potential Abuse Deepfake videos, autonomous weapons, driving assistance AI, especially Machine learning methods can have bias. Machine learning can learn to be bias if the data contain bias. Example: recidivism-risk assessment in the criminal justice system Fairness is subjective. Co-pilot helps us write code by copying from other open source. Self-driving cars save its owner by crashing to pedestrians.