Do you like games? Are you curious about Artificial Intelligence? And how about joining these two great topics?!
This talk will introduce some general AI concepts as well as a case study on Super Mario Maker game. Let's discuss and understand the step by step how to apply Artificial Intelligence in games, shall we?
2. Speaker
Leonardo Mauro P. Moraes
2
Experience
• Machine Learning Engineer
(Sinch) Brazil, Nov. 2020 – Now
• Data Science Tutor
(USP) Brazil, Set. 2020 – Now
• PhD in Computer Sciences
(USP) Brazil, Mar. 2022 (Doing)
https://www.linkedin.com/in/leomaurodesenv/
3. 3
Agenda
• Games universe
• AI introduction
Introduction Datasets
Data forArtificial
Intelligence projects
Project
Case Study -
Super Mario Maker
ARTIFICIAL INTELLIGENCE
6. • The most popular form of entertainment,
reaching millions of players;
Digital Games
6
¹Newzoo’s Games Trends to Watch in 2021
• Universe of games is in constant ascendancy;
both in production and in consumption.
• Estimated U$ 189.3 billion for 2021¹;
• Also, it is cool!
7. 7
• eSports - professional competitions.
• joining Olympics in 2024.
• Streamers - produce online videos.
• audience hit 728.8 million in 2021²
• +10.0% from 2020.
Digital Games
• Game Influencers
• professional players and streamers.
²How Big Is the Game Live Streaming Audience?
8. 8
• Digital influencers of games;
• Exist since the popularization of social media;
• Publish online content;
• e.g., videos, blogs, forums.
• High influence in new trends.
Game Influencers
11. Artificial Intelligence (AI)
11
is intelligence demonstrated by machines. Its definition,
AI research as the study of "intelligent agents": any
device that perceives its environment and takes actions
that achieving its goals.
Russell et. al (2016)
Russell, S. J., Norvig, P. (2016). Artificial intelligence: a modern
approach. Malaysia; Pearson Education Limited.
12. Data Mining (DM)
12
is the process of discovering patterns in data sets (or datasets)
involving methods of machine learning, statistics, and database
systems, i.e., Artificial Intelligence; DM focus on extraction of
patterns in datasets.
Han and Kamber (2011)
Han, J., Pei, J., Kamber, M. (2011). Data mining:
concepts and techniques. Elsevier.
13. Machine Learning (ML)
13
is a research area from computer science
that can learn automatically through
experience and by the use of data.
Mitchell (1997)
Mitchell, Tom (1997). Machine Learning. New York: McGraw Hill.
15. 15
Fayyad, U. M., Piatetsky-Shapiro, G. S., & Smyth, P. (1996). P. and Uthurusamy, R.
Advances in Knowledge Discovery and Data Mining.
Knowledge Discovery
in Databases (KDD)
17. Where to find
a dataset?
17
• UCI Machine Learning Repository
http://archive.ics.uci.edu/ml/index.php
• Kaggle
https://www.kaggle.com/datasets
• GitHub
https://github.com
• etc...
18. 18
and ... games'
dataset?
Awesome Game Datasets
https://github.com/leomaurodesenv/game-datasets
Over 150 contents
“Awesome manifesto” is about finding
the awesome in the everyday.
20. Super Mario Maker
20
• Released on September 2015 - Nintendo
• Contains games from Super Mario Bros series.
• Video game console - Nintendo Wii U
• Player can:
• Create a game level;
• Play, clear, like …
21. Super Mario Maker
21
"If you played every level in #SuperMarioMaker for 1 minute
each, it would take you nearly 14 years to play them all!"
Nintendo (Twitter)
23. 23
SMMnet - kaggle
• over 115k games levels;
• over 880k players;
• over 7 millions of interactions.
• four game difficulties;
• four game styles;
• four countries (BR, CA, FR, DE);
• by quasi five months.
Moraes,L. M. P., Codeiro,R. L. F. (2019). Smmnet:A socialnetwork ofgamesdataset.Brazilian
Symposium on Databases(SBBD) - DatasetShowcase Workshop (DSW).Ceará,Brazil.
Super
Mario
Maker
Dataset
24. Super Mario
Maker Dataset
24
Social Network of Games
• Interactions: player → game
• play, like, etc.
• They can change over time
Moraes,L. M. P., Codeiro,R. L. F. (2019). Smmnet:A socialnetwork ofgamesdataset.BrazilianSymposium
on Databases (SBBD) - DatasetShowcaseWorkshop(DSW). Ceará,Brazil.
26. 26
Project Procedures
(a) Planning
• What to do?
PDCA Cycle
(plan–do–check–adjust)
Proposed by Shewhart,
Executed by Deming
For quality control
(b) Execution
• How to do?
(c) Evaluation
• Did I do it right?
27. 27
(a) Game Influencers
• Popular players who have
high influence in new trends.
Consequently
• Companies invest to endorse their products;
• Direct relevance in viral marketing.
Planning
28. 28
1. What to do?
• I want to find game influencers
in Super Mario Maker. (Problem)
2. Has anyone done something similar?
• Probably yes.
Planning
(a) Game Influencers
29. 29
• How to detect an influencer?
• What are the influencers’ characteristics?
Characteristics
• Publish many contents;
• Receive many “likes”; but not only that...
• Evolution of likes, trend of peaks.
Planning
(a) Game Influencers
30. • Evolution of likes, trend of peaks.
Planning
30
(a) Game Influencers
31. 31
Execution
1. How to do?
• Baseline – replicate similar works.
2. Enjoy!
• Creativity, explore your ideas!
Planning
(b) Execution
32. 32
1. Modeling
• Evolution of likes for each game level.
2. Feature Extraction
• Formulas to extract the
tendency of peaks in the streams.
Execution
Planning
(b) Execution
34. 34
3. Player Modeling
• Extract features from the game levels;
• A player is represented by the combination
of the characteristics of his/her games.
Execution
Planning
(b) Execution
36. 36
Did I do it right?
• Evaluation, Metrics…
Detection -> Classification
• Accuracy
• Precision
• Recall
• etc…
Execution
Evaluation
Planning
(c) Evaluation
37. 37
Dataset
• Canadian players - test/train.
• Manually labeled
the top-100 by the likes.
• 41 game influencers.
• Consensus: influencer who published in
popular websites of the SMM community.
Execution
Evaluation
Planning
(c) Evaluation
38. 38
• 28 algorithms evaluated;
• best: 87.1% accuracy and 90.3% precision.
Generality Test
• Is an algorithm trained in one country (CA)
capable of inferring influencers in
another nationality (FR)?
Yes, with 77.8% precision.
Execution
Evaluation
Planning
(c) Evaluation
39. 39
Is this the best way?
• I doubt it!
Remember: Cycle!
• Improve as much as you can.
Execution
Evaluation
Planning
(d) Adjust
41. 41
1. How to work with Artificial Intelligence
• Awesome Game Datasets
Moraes,L. M. P., Codeiro,R. L. F. (2019). Smmnet:A socialnetwork ofgamesdataset.BrazilianSymposium
on Databases (SBBD) - DatasetShowcaseWorkshop(DSW). Ceará,Brazil.
Moraes,L. M. P., Codeiro,R. L. F. (2019).Detecting Influencers in Very Large Social Networks ofGames.In
21stInternational Conferenceon Enterprise InformationSystems (ICEIS),Crete, Greece.
3. Your creativity is the limit!
2. Case Study - Super Mario Maker (code)
• Planning, Execution, Evaluation, Again?