02_Technology%20behind%20AI.pdf

Technology behind AI MGT00154 AI for Innovation and Entrepreneurship 23rd October, 2023 | TUM Audimax The Story behind... 2 3 4 5 6 7 8 Technology behind AI/Machine Learning under the Hood 23rd October Module Overview In the "Technology behind AI or Machine Learning under the Hood"...

Technology behind AI MGT00154 AI for Innovation and Entrepreneurship 23rd October, 2023 | TUM Audimax The Story behind... 2 3 4 5 6 7 8 Technology behind AI/Machine Learning under the Hood 23rd October Module Overview In the "Technology behind AI or Machine Learning under the Hood" module, students will delve deeper into the technical realms of AI. This session is designed to demystify the process of building machine learning algorithms and to explain the different machine learning styles, including an exploration of select approaches like deep learning. The aim is to equip students with a high-level conceptual understanding of the technical aspects of AI, shedding light on the intricate technologies that power artificial intelligence. Why Attend This Session This session is pivotal for students aiming to grasp the technological essence of AI and machine learning. It offers a unique opportunity to understand the predictive nature of AI and to learn about the diverse styles of machine learning. By attending this module, students will gain valuable insights into the construction of machine learning algorithms and the underlying technologies, enabling them to appreciate the technical intricacies of AI. This knowledge is crucial for anyone looking to explore AI applications, innovations, or developments, providing a solid technical foundation for further learning and exploration in AI. 9 Technology behind AI/Machine Learning under the Hood Learning Objectives 1. Insight into Machine Learning Algorithm Construction: Learn the essentials of building machine learning algorithms, gaining a foundational understanding of the processes involved in creating AI. 3. High-Level Conceptual Understanding of AI Technology: Achieve a conceptual grasp of the technical side of AI, forming a solid base for exploring and understanding advanced technical concepts in AI. . 2. Knowledge of Machine Learning Styles and Technologies: Explore the different styles of machine learning, acquiring a broad perspective on the different approaches and methodologies in AI (including deep learning via neural networks). 10 Agenda 1 Important concepts 3 Artificial Neural Networks 2 Machine Learning Styles 4 ML Skills Supervised Learning Unsupervised Learning (Semi-Supervised Learning) Reinforcement Learning 6 11 Building a ML Model Some theory 12 Prediction takes information you have to generate information you do not have. Prediction is a central input into decision making / automation 13 Join at slido.com #3592085 14 ⓘ Start presenting to display the joining instructions on this slide. 15 What is the predicted? 16 ⓘ Start presenting to display the poll results on this slide. 17 What is predicted? 18 ⓘ Start presenting to display the poll results on this slide. Instances & Features Features are also known as 'Examples', 'Data Points' or 'Observations' Features are the important properties of a given Subject, used to enable predictions. Color: Red Choosing the right set of features is crucial. Using a larger set of features than necessary is possible. Weight: 352g 19 Fruit as an Example 20 The Importance of Choosing the Right Features 21 Unlimited Dimensionality Often, more than two features are necessary. Dimensionality can go into the millions. Principles of ML remain, regardless of dimensionality of Data 22 What are the Features in Pictures? 23 What are the Features in Pictures? 24 What are the Features in Pictures? R002 G002 B002 25 Labeling Data 26 What is the correct label of this image? 27 What is the correct label of this image? ⓘ Start presenting to display the poll results on this slide. Building a simple ML Model Example: Linear Regression 29 Predict the number of ice cream scoops sold from temperature Input Feature Output X y Example from https://www.appliedai-institute.de/kostenfreie-online-kurse/ai-essentials 30 Linear Regression y =w *X+b =w * +b Model parameters 31 32 Regression model created by learning the parameters w1 33 Parameters are learned by minimizing an error to the data 34 Parameters are learned by minimizing an error to the data 35 Styles of Machine Learning 36 Supervised Machine Learning 37 Supervised Learning Definition Supervised Machine Learning uses labeled training data to learn the relation between a given input and output (label). Example: Wine Classifier 38 Starting with Wine Data 39 Seperating Yes from No 40 Optimizing or minimizing Errors 41 Choosing the right Algorithm 42 From Data to trained Model 43 Predicting Labels 44 45 Unsupervised Machine Learning 46 Sort the images into two groups 47 In what two groups would you sort the images? 48 ⓘ Start presenting to display the poll results on this slide. Group 1 Group 2 49 Dexter Elvis 50 Clustering 51 ML Algoritms Sidenote Aimed towards finding the 'underlying structure' of data without explicit given output structure (labels) Example: Explore data of different shops (sales, neighbourhood, opening hours, ...) to detect patterns. 52 Supervised vs. Unsupervised Learning Supervised: Used when a labeled dataset is available. Aims towards enabling the model to make accurate predictions on unseen data. Unsupervised: Used when a labeled Dataset is not available. Model aims to identify patterns like clusters allowing it to discover new, unseen ones. 53 54 Reinforcement Machine Learning 55 Learn iteratively to perform a task by continuously trying and receiving positive or negative rewards Example: Minimizing energy consumption of a data center 57 Artificial Neural Networks (Deep Learning) 58 59 Deep Learning How does it work ? Works by processing inputs and fitting Weights to generate an output. Backpropagation used to fit LOTS of parameters to allow accurate predictions. 60 61 62 ‹#› ML Skills 64 65 66 To master ML development, you need several talents 67 Audience Q&A Session 68 ⓘ Start presenting to display the audience questions on this slide. 69 Julia Gottfriedson, OroraTech Julia Gottfriedsen leads the Data Science team at OroraTech, which focuses on Earth observation data analysis and processing, machine learning applications, and new data product development. With a background in machine learning and environmental science, Julia has gained extensive experience within AI research and climate tech in different industries and research institutes: She worked at the Siemens AI Lab in Munich and California, investigated extreme weather events at the German Aerospace Center and is in research collaborations with ESA, LMU and TUM Munich. Her goal is to be at the forefront of technological innovation to create a carbon-neutral future. 70 Audience Q&A Session 71 ⓘ Start presenting to display the audience questions on this slide. Clair Helffer de Soyres, Nvidia Claire Helffer-de Soyres is leading Generative AI for Enterprise in the DACH region at Nvidia. Her role is to facilitate adoption of AI across organizations working closely with Nvidia partner ecosystem, as well as advising on accelerated computing. In the past years, she worked on multiple ML and AI projects as well as strategic partnerships in banking and insurance, and gained insights on best practices and pitfalls. Prior to this, she dedicated most of her career in information technology and network infrastructure for the financial industry across Europe, among others at Orange Business and Colt Technology. 72 Audience Q&A Session 73 ⓘ Start presenting to display the audience questions on this slide. Generative AI and Prompt Engineering 30th October Module Overview In the "Generative AI and Prompt Engineering" module, students will explore the burgeoning field of Generative AI and its transformative impact on the economy. This session will provide insights into conceptualizing and building applications leveraging Generative AI. The latter half of the lecture will transition into a hands-on exercise focused on prompt engineering, where students will learn various techniques, strategies, and tips to craft effective prompts. This module aims to acquaint students with the latest trends in AI and to enhance their proficiency in prompt creation, a skill integral to interacting with Language Learning Models (LLMs). Why Attend This Session This session is essential for students to stay abreast of the latest advancements in AI, specifically in the realm of Generative AI. By attending, students will not only gain insights into the applications and implications of Generative AI but also acquire practical skills in prompt engineering, a crucial component in utilizing LLMs effectively. Mastering the art of writing prompts is a valuable skill, beneficial throughout the course and beyond, especially for those aspiring to delve deeper into AI applications and innovations. This module will be particularly helpful for students in completing assignments that involve interaction with LLMs, providing them with the necessary tools and knowledge to excel. ‹#› Generative AI and Prompt Engineering Learning Objectives 1. Leverage generative AI’s application potential Get to know the application potential of Generative AI (especially large language models) and its transformative impact on the economy. 2. Be aware of Generative AI’s limitations Recognize the limitations of the state-of-the art Generative AI approaches and understand the need for governance mechanisms. 3. Incorporating Generative AI into an organization Understand both the business and technical aspects of incorporating LLMs into an organization. 4. Make rational, informed business decisions Learn how to navigate this new era of LLMs, enabling to thoroughly evaluate available opensource and closed-source LLM options, make strategic make-or-buy decisions, and consider process (re-)design. 5. Master the art of writing prompts Learn various techniques, strategies, and tips to craft effective prompts 75