AI & Programmatic Advertising Master Class - Yang Han, StackAdapt
0 likes494 views
The document outlines a masterclass on AI and programmatic advertising led by Yang Han, detailing the workings of AI, its application in business, and its use in advertising. It discusses the complexities of AI, challenges faced, and how machine learning can optimize ad delivery by predicting user behavior based on extensive data analysis. The future of AI is also explored, highlighting the evolution of AI capabilities and its impact on marketing strategies.
AI & Programmatic Advertising Master Class - Yang Han, StackAdapt
- 1. Yang Han CHIEF TECHNOLOGY OFFICER, STACKADAPT TORONTO, ON ~ MAY 17 – 18, 2018 | DIGIMARCONCANADA.CA #DigiMarConCanada AI & Programmatic Advertising Master Class MASTERCLASS
- 2. Part 1. What is AI and how it works Part 2. AI and Business Part 3. Case Study in Advertising Part 4. AI in the Future Agenda
- 3. Part 1 What is AI
- 6. • Adaptable • Automated • Consistent • Solves the problem (better than a human) What is AI
- 7. A lot of data (Clean data) What Goes into AI?
- 8. How it works
- 10. Examples of AI
- 11. Neural Networks ● Part-of-speech Tagging ● Tokenization ● Named Entity Recognition ● Intent Extraction Recurrent Neural Networks ● Machine Translation ● Question Answering System ● Image Captioning Recursive Neural Networks ● Parsing sentences ● Sentiment Analysis ● Paraphrase detection ● Relation Classification ● Object detection Convolutional Neural Network ● Sentence / Text classification ● Relation extraction and classification ● Spam detection ● Categorization of search queries ● Semantic relation extraction Deep Learning Techniques
- 12. Foundations: Image recognition, Voice recognition, Natural language processing (NLP), Online user behavior Utilized to create a wide range of applications: • Automated Content Creation • Automated Suggestions • Content Personalization • Ad Personalization, Targeting, & Monetization • Chatbots, Smart assistants, Q&A • Concept extraction from text and images • Sentiment analysis • Dynamic A/B testing Keywords: Personalized, Automated, and Humanless AI in Marketing
- 13. • Data can be biased • AI doesn’t know the broader world • Complex models do not always equal to better results • AI can only do what you train it to do • Most of AI is a black box “A robot can learn to pick up a bottle, but if it has to pick up a cup, it starts from scratch.” “We can build these models, but we don’t know how they work” AI Challenges
- 14. Part 2 AI in Business
- 15. Building AI vs. Applying AI
- 16. • Has the problem been solved already? • Can I solve the same problem without ML? • Do I have the data and resources? When to Build AI?
- 17. Data Engineer Build data pipelines for training, and model application Data Scientist Model training, testing, and validation Hiring for AI
- 18. Part 3 A Case Study in Advertising
- 19. How ads are brought and sold online
- 20. Machine learning is used to determine which users to show specific ads to, and how to deliver the message • Maximizing an ad’s click-through-rate, time on site, and conversions • By understanding the user and who they are • Dynamically creative variations (headlines/imageneratingges) that provide optimized and personalized messaging How do we leverage AI?
- 21. Machine learning pipeline that analyzed billions of data points to predict user actions such as clicks and engagements Custom Segments NLP to understand what 1 billion users are reading about in real-time Thompson Sampling A method of automatically selecting bidding strategies, to improve the performance of campaigns. Age/Gender prediction Cross Device prediction What we built
- 22. Old Algorithm Machine Learning Algorithm Newest Algorithm (Not yet released)
- 24. User profiles based on browsing history of over 1 billion online users Stackadapt User Profile
- 25. • Platforms require the marketer to understand who to target, and how to communicate the message • Platforms track and identify actions (ads clicked, sites visited, products purchased) • Machine learning uses actions as data points, in addition to current attributes of a user (demographics, location, device, keywords on the page) • Optimizations result in amplifying your current actions. This is reactive. Advertising today
- 26. • Platforms will play a bigger role in automating messaging and targeting • Platforms will harness massive sequences of user actions and context to understand the individual and their behavior • Machines will be able to predict upcoming intent that seem unrelated to current actions. This is predictive. • Machines will know if you’re a customer, even before you do How it will work tomorrow
- 27. Part 4 AI in the Future
- 28. • The easiest problem is clear goals in a predictable environment. Example: Image recognition • A harder problem is clear goals in an unpredictable environment. Example: Self driving cars, Video • Another hard problem is indirect goals in a predictable environment. Example: Winning at a board game. • Hardest case: undefined or distant goals in an unpredictable environment. These can’t be solved by current AI at all. Current State of AI
- 29. How did image recognition become possible? Synergies: the foundation of AI evolves with • Evolution of Data Today’s examples: IoT, Sensors, Blockchain Ecosystems and interactivity are key • Computing Power Enterprises & industries evolve by applying AI Evolution of AI
- 30. Human vs AI
- 31. Thank you Q&A