Evolve19 | Giancarlo Berner & Brett Butterfield | AI & Adobe Sensei

Information about Evolve19 | Giancarlo Berner & Brett Butterfield | AI & Adobe Sensei

Published on August 15, 2019

Author: evolve_conference

Source: slideshare.net


1. #evolve19 AI / MACHINE LEARNING & ADOBE SENSEI Brett Butterfield – Adobe Sensei Giancarlo F. Berner – Kleiber Digital August 6 - 2019

2. #evolve19 2 SESSION OVERVIEW • Introduction into AI/Machine Learning • Live Demo of AI Techniques, powered by Adobe Sensei

3. #evolve19 3 WHAT IS ARTIFICIAL INTELLIGENCE? Artificial Intelligence is the concept and process of defining and building systems that act like humans, think like humans, learn like humans and decide like humans. G.F.Berner - 1983 (Today I want to add “and perform tasks like humans”)

4. #evolve19 4 1956 – AI workshop at Dartmouth College (NH) – Beginning of AI research 1956 – 74 – Reasoning Algorithms (games), Natural Language, Robotics 1980 – 87 – Expert Systems, Knowledge Base, Pattern Recognition 1993 – 2011 – Deep Blue, Autonom. Driving, Robotics, Speech Recognition, Search Engines, Cognitive Systems, Computational Intelligence 2011 – today – Machine/Deep Learning, Big Data, Speech/Image Recognition 1983 – Work on USA – Universal System Analyzer (Digital Analysis System) 1985 – Matrix Matching Concepts (OCR, Image) 1987 – Work on Dr. Hey – Expert System for Pharmacies 1989 – Analysis of Chess Algorithms A LITTLE HISTORY

5. #evolve19 55 AI Machine Learning Deep Learning Expert Systems Knowledge Base Pattern Recognition Image Recognition Robotics Speech Recognition Animation Simulation Diagnosis Analyzers Games Predictions

6. #evolve19 6 MACHINE LEARNING - DEFINITION Machine Learning is the practice and science of building systems that continuously improve the outcome of a process, not by modifying the code, but by processing more and better data. G.F.Berner - 1992

7. #evolve19 7 Common Styles • Supervised Learning – A model’s mapping functions are trained by labelling the input data and knowing the output data. • Unsupervised Learning – The model only knows the input variables, but not the corresponding output variables. Used e.g. to structure the data (“similarities”). New Styles • Reinforcement Training – “Trial & Error”, decide the best next action (“what gives most points”). MACHINE LEARNING STYLES COMMON AND NEW LEARNING STYLES

8. #evolve19 8 Machine Learning Style • Classification – Predict the outcome if the output variables are in form of categories, e.g. input label is “headache”, output is “healthy” or “sick”. • Regression – Predict the outcome if the output variables are in form real values, e.g. predict the age of a person. Popular Algorithms • Linear Regression (determine output variable of an input and quantify relationship) • Logistic Regression (transformation function to determine probability) • Naïve-Bayes (probability that an event will occur based on past events) • CART (Classification and Regression Trees) • KNN (K-Nearest Neighbors) SUPERVISED LEARNING FORMS AND ALGORITHMS

9. #evolve19 9 Machine Learning Style • Association – Find probability of co-occurrence of items in a data set. Popular in E- Commerce, e.g. “Buy a printer, there is 85% probability to buy toner”. • Clustering – Group samples so that objects in the same cluster are more similar than to objects of another cluster. • Dimensionality Reduction – reduce the number of variables of a data set but still maintaining the relevant information. Feature Extraction and Feature Selection methods Popular Algorithms • Apriori (transactional database, popular for commerce) • K-means (iterative algorithm, groups similar data into clusters) • Principal Component Analysis (PCA) (Feature Extraction) UNSUPERVISED LEARNING FORMS AND ALGORITHMS

10. #evolve19 10 DEMO ADOBE SENSEI MACHINE LEARNING & AI • Voice Search • Image Recognition • ~ 10 more Live Demos Brett Butterfield – Director Adobe Sensei

11. #evolve19 THANK YOU! Brett Butterfield – [email protected] Giancarlo Berner – [email protected]

#evolve19 presentations

Related presentations

Other presentations created by evolve_conference