Download as PDF, PPTX
Uploaded byDan O'Leary
AI, Machine Learning, and Data Science Concepts
This document presents an introduction to artificial intelligence (AI), machine learning (ML), and data science for an audience unfamiliar with these concepts. It explores definitions, key components, current methodologies, and applications while also discussing the evolution and societal impacts of these technologies. The lecture emphasizes the importance of distinguishing between AI and ML, the significant data growth and computational advancements driving recent progress, and the benefits and limitations associated with data-driven approaches.
More Related Content
Similar to AI, Machine Learning, and Data Science Concepts
![[Redis Released]- FalkorDB - Redis + Graph Agentic Memory’s Secret Sauce](https://cdn.slidesharecdn.com/ss_thumbnails/redisreleased-falkordbslidedeck-1125-251115194922-e1c0046b-thumbnail.jpg?width=640&height=640&fit=bounds)
AI, Machine Learning, and Data Science Concepts
- 1. Artificial Intelligence & Machine Learning Concepts DataScience Society of Auburn November 16, 2020 Dan O’Leary dan.oleary@auburn.edu
- 2. Introduction • Introduction toAI/ML concepts and Data Science • Based on a lecture for undergrads in BET 2019 • No expectation of related knowledge • Expanded and deepened for this audience • Still “Big Picture” • Provide some context, detail for those of you in deeper studies • Introduce those of you new / interested • Disclaimers • Some references a little dated (yes, in 18 months) • Not as well cited as it should be • I am not an expert! DATA SCIENCE SOCIETY OF AUBURN 11/16/20 Data Science Society of Auburn // AI, ML, and DS Concepts // November 16, 2020 2
- 3. Topics • Attempt todefine Artificial Intelligence (fail) • Attempt to describe the field of Data Science (better) • Introduce fundamental concept of Machine Learning (pretty good!?) • Describe state of the art methods in 1 slide each (jury’s out) • Use cases and examples (great!) • Discuss why now, benefits, and limitations / pitfalls (ok) Only the tip of the tip of the iceberg Not a formal academic presentation 11/16/20 Data Science Society of Auburn // AI, ML, and DS Concepts // November 16, 2020 3
- 4. Background • 1992 –BS Mechanical Engineering • 2019 – Master of Engineering Management, Supervised Learning • 2020 – Grad Cert Modeling and Analytics for Operations (ISE) • ≈2022 – PhD, Industrial and Systems Engineering • Modeling / simulation, data science / machine learning • Teach undergrad courses in innovation and product development • Life-long fascination with modeling and simulation of all types • 2 x Entrepreneur: co-Founded n-Space in 1994, funding from Sony • 23 years, 45 games – concept to completion; major brands, publishers, and platforms; all genres and demographics 11/16/20 Data Science Society of Auburn // AI, ML, and DS Concepts // November 16, 2020 4
- 5. 11/16/20 Data ScienceSociety of Auburn // AI, ML, and DS Concepts // November 16, 2020 5 From GTC 2018 – GPU Tech Conference https://youtu.be/GiZ7kyrwZGQ Embedded videos have been replaced with screenshots and youtube links.
- 6. Question: What isArtificial Intelligence (AI)? If you can’t explain it to a six-year-old, you don’t understand it yourself. - Albert Einstein Answer: I don’t know. 11/16/20 Data Science Society of Auburn // AI, ML, and DS Concepts // November 16, 2020 6
- 7. What is AI? •Imprecise, overused term • Calculator? • Self-driving car? • Chatbot? • Definition is fuzzy, changes over time • Old, diluted, hyped term – backlash, cynicism • Generally used to describe machines doing tasks traditionally assigned to humans 11/16/20 Data Science Society of Auburn // AI, ML, and DS Concepts // November 16, 2020 7
- 8. Classic Definition ofAI An intelligently designed agent that perceives its environment and makes decisions to maximize the chances of achieving its goal. • Subfields: • Computer Vision • Robotics / Control Theory • Natural Language Processing https://medium.com/machine-learning-for-humans/why-machine-learning-matters-6164faf1df12 Now considered “Machine Learning” 11/16/20 Data Science Society of Auburn // AI, ML, and DS Concepts // November 16, 2020 8
- 9. AI Effect Once amachine takes over a task, humans tend to dismiss it as “not AI” It’s part of the history of the field of artificial intelligence that every time somebody figured out how to make a computer do something — play good checkers, solve simple but relatively informal problems — there was chorus of critics to say, ‘that’s not thinking’ Pamela McCorduck, 2004 AI is whatever hasn't been done yet. – Douglas Hofstadter 11/16/20 Data Science Society of Auburn // AI, ML, and DS Concepts // November 16, 2020 9
- 10. AI Effect • 1997– IBM’s Deep Blue beats world chess champion Garry Kasparov • 259th most powerful supercomputer at the time • Planned 6-8 moves out; as high as 20+ • “Brute force methods… not real intelligence” • Changed the canonical example of human vs machines from Chess to Go • Simple rules, many more possible moves • More intuition, less susceptible to brute force • 2016 Google DeepMind beats Go Champ • 2019 DeepMind beats StarCraft II Pros 11/16/20 Data Science Society of Auburn // AI, ML, and DS Concepts // November 16, 2020 10
- 11. David Robinson: http://varianceexplained.org/r/ds-ml-ai/ 11/16/20Data Science Society of Auburn // AI, ML, and DS Concepts // November 16, 2020 11
- 12. Data Science • ArtificialIntelligence • Big Data • Statistical Learning • Predictive Analytics • Data Mining • Machine Learning • Pattern Recognition • Deep Learning Terms, usage, and interpretation vary Overwhelmingly expansive and fast-moving field 11/16/20 Data Science Society of Auburn // AI, ML, and DS Concepts // November 16, 2020 12
- 13. PracticeTheory MATH STATS ENG SCI BUSINESS Excel TableauMinitab MATLAB Python R Dan’s Crude Modelof Domains and Tools v0.01 Broad, Multi/Interdisciplinary Interest Figures Not To Scale! Source: I have only myself to blame for this slide. 11/16/20 Data Science Society of Auburn // AI, ML, and DS Concepts // November 16, 2020 13
- 14. The “Classic” Definitionof Data Science drewconway.com/zia/2013/3/26/the-data-science- venn-diagram Robinson, Emily, and Jacqueline Nolis. Build a Career in Data Science. Simon and Schuster, 2020. 2010 2020 11/16/20 Data Science Society of Auburn // AI, ML, and DS Concepts // November 16, 2020 14
- 15. Data Science inBroad Terms • Components (Skills) • Math and Stats – methods related to data literacy • Programming & Databases – coding, engineering, carpentry • Domain Knowledge – subject matter expertise • Applications (Jobs) • Analytics – create dashboards and reports that deliver data • Machine Learning – creates models that run continuously • Decision Science – creates analyses that create recommendations Robinson, Emily, and Jacqueline Nolis. Build a Career in Data Science. Simon and Schuster, 2020. 11/16/20 Data Science Society of Auburn // AI, ML, and DS Concepts // November 16, 2020 15
- 16. David Robinson: http://varianceexplained.org/r/ds-ml-ai/ 11/16/20Data Science Society of Auburn // AI, ML, and DS Concepts // November 16, 2020 16
- 17. Outcome-based… • Data Scienceproduces insights • Various types of insight – descriptive, exploratory, causal • Statistical inference, data visualization, and experiment design • Machine Learning produces predictions • Various types of predictions – regression, classification • Artificial Intelligence produces actions • Executed or recommended by autonomous agents • Includes game-playing, robotics / control theory, optimization, NLP, RL David Robinson: http://varianceexplained.org/r/ds-ml-ai/ 11/16/20 Data Science Society of Auburn // AI, ML, and DS Concepts // November 16, 2020 17
- 18. Example: Self-Driving Car •Machine Learning • Object recognition model trained using many photos of streetside objects • System predicts the presence of stop signs • Artificial Intelligence • Given varying road conditions and presence of a stop sign • Autonomous agent decides when / how to act, properly applying the brakes • Data Science • Analyzing test data developers gain insight about the cause of false negatives • They generate a report summarizing their findings / recommendations David Robinson: http://varianceexplained.org/r/ds-ml-ai/ 11/16/20 Data Science Society of Auburn // AI, ML, and DS Concepts // November 16, 2020 18
- 19. Another Angle –Archetypes • Four Components of Data Science • Analysis – insights • Modeling – prediction • Engineering – deployment • Mechanics – cleaning / prep* • Five Archetypes of Data Scientists • Generalist – proficient at everything • Detective – master of analysis • Oracle – master of modeling • Maker – master of engineering https://e2eml.school/data_science_archetypes.html 11/16/20 Data Science Society of Auburn // AI, ML, and DS Concepts // November 16, 2020 19
- 20. Not Shown • Fewall-around masters! • Everyone cleans data! https://e2eml.school/data_science_archetypes.html 11/16/20 Data Science Society of Auburn // AI, ML, and DS Concepts // November 16, 2020 20
- 21. 11/16/20 Data ScienceSociety of Auburn // AI, ML, and DS Concepts // November 16, 2020 21
- 22. What is AI? •I’m still not sure… • Let’s go with this: • It includes ML and DL • Actions → AI • Predictions → ML • Usually • Ignore grander AI visions, claims, speculation Most of what we think of as “AI” today based on Deep Learning methods Much of AI’s imagined potential remains distant Machine Learning & Deep Learning are very real, here now, everywhere 11/16/20 Data Science Society of Auburn // AI, ML, and DS Concepts // November 16, 2020 22
- 23. Machine Learning • Gives“computers the ability to learn without being explicitly programmed.” – Arthur Samuel, 1959 • Identify patterns in observed data 11/16/20 Data Science Society of Auburn // AI, ML, and DS Concepts // November 16, 2020 23
- 24. Unsupervised Learning https://towardsdatascience.com/what-is-machine-learning-a-short-note-on-supervised-unsupervised-semi-supervised-and-aed1573ae9bb 11/16/20 DataScience Society of Auburn // AI, ML, and DS Concepts // November 16, 2020 24
- 25. Supervised Learning https://towardsdatascience.com/what-is-machine-learning-a-short-note-on-supervised-unsupervised-semi-supervised-and-aed1573ae9bb 11/16/20 DataScience Society of Auburn // AI, ML, and DS Concepts // November 16, 2020 25
- 26. Machine Learning • LinearRegression to Deep Neural Nets • Ingredient technology • “Macroscope” (inverted microscope) – sees things too big to view • Deep Neural Nets with tens of millions of parameters • Image data sets on the order of 1M x 1M+, video much larger • Entire USPTO archive (text and images), over 4M patents to 1976 • Many data sets much larger • Learns by finding statistical structure in training examples • Meaningful transformation / representations of data • Largely empirical methods 11/16/20 Data Science Society of Auburn // AI, ML, and DS Concepts // November 16, 2020 26
- 27. Transformations / Representationsof Data Example: Classification Chollet, François. Deep Learning with Python. Manning Publications Company, 2017. 11/16/20 Data Science Society of Auburn // AI, ML, and DS Concepts // November 16, 2020 27
- 28. Lots of waysto do it… Best method? It depends. • Bayesian • Decision Tree • Dimens. Reduction • Instance Based • Clustering • Regression • Rule System • Regularization • Neural Networks • Ensemble • Deep Learning SVM?! 11/16/20 Data Science Society of Auburn // AI, ML, and DS Concepts // November 16, 2020 28
- 29. “State of theArt” For Kaggle Contests, at least • Gradient Boosting • LightGBM, XGBoost • For structured data • Python or R • Deep Learning • Keras/TF, Fastai/PT • For perceptual problems • Python 11/16/20 Data Science Society of Auburn // AI, ML, and DS Concepts // November 16, 2020 29
- 30. Gradient Boosting in1 Slide • Series of decision trees* • Each improved by prior • Weights adjusted based on ease of classification • Repeat and combine results https://datascience.eu/machine-learning/gradient-boosting-what-you-need-to-know/ *Decision trees can be thought of as giant “if-then” structures converting inputs to outputs based on features 11/16/20 Data Science Society of Auburn // AI, ML, and DS Concepts // November 16, 2020 30
- 31. Deep Learning in1 Slide Bonus: Neural Network! Neural Network >1 → “Deep” 𝑓(𝒘 $ 𝒙 + 𝑏) 𝑥! 𝑦 𝑥" 𝑥# 𝑥$ 𝑥% 𝑥& 𝑓($) is the activation function Non-linear: sigmoid, tanh, etc. http://neuralnetworksanddeeplearning.com/index.html Chollet, François. Deep Learning with Python. Manning Publications Company, 2017. 11/16/20 Data Science Society of Auburn // AI, ML, and DS Concepts // November 16, 2020 31
- 32. Common Current ML/DLUse Cases • Natural Language Processing • Google Translate, Siri/Cortana/Alexa, Auto-correct • Recommendation Systems • Netflix, Amazon, Facebook • Customer Relationship Management • Direct marketing, mobile advertising, chatbots • Finance • Credit score, loan approval, fraud detection ($100B-$1T), algorithmic trading • Image Recognition • Pose detection, facial recognition, medical image processing Tip of the Iceberg – ML/DL is Everywhere! 11/16/20 Data Science Society of Auburn // AI, ML, and DS Concepts // November 16, 2020 32
- 33. 11/16/20 Data ScienceSociety of Auburn // AI, ML, and DS Concepts // November 16, 2020 33 https://youtu.be/ayPqjPekn7g
- 34. 11/16/20 Data ScienceSociety of Auburn // AI, ML, and DS Concepts // November 16, 2020 34 https://youtu.be/0FW99AQmMc8
- 35. 11/16/20 Data ScienceSociety of Auburn // AI, ML, and DS Concepts // November 16, 2020 35 Trains only using the score – reinforcement learning https://youtu.be/TmPfTpjtdgg
- 36. 11/16/20 Data ScienceSociety of Auburn // AI, ML, and DS Concepts // November 16, 2020 36 https://youtu.be/kSLJriaOumA
- 37. 11/16/20 Data ScienceSociety of Auburn // AI, ML, and DS Concepts // November 16, 2020 37 https://youtu.be/LBd5FZqhUVk
- 38. 11/16/20 Data ScienceSociety of Auburn // AI, ML, and DS Concepts // November 16, 2020 38 https://youtu.be/0jcigK65mpc
- 39. 11/16/20 Data ScienceSociety of Auburn // AI, ML, and DS Concepts // November 16, 2020 39 https://youtu.be/DjERMBnvTEE
- 40. Why Now? • 50+years of research • Algorithm / SW dev • Huge Investments • Democratization 11/16/20 Data Science Society of Auburn // AI, ML, and DS Concepts // November 16, 2020 40
- 41. 11/16/20 Data ScienceSociety of Auburn // AI, ML, and DS Concepts // November 16, 2020 41
- 42. Why Now? • Computepower • GPU – graphics processing unit • Originally developed for 3D graphics • Massively parallel matrix operations • Orders of magnitude better performance • Deeper Blue (1997) • 11.38 GFLOPS • ~ $100M • NVIDIA GTX 1080 (2016) • 8,873 GFLOPS • $499 MSRP • 150 million times more GF / $ 11/16/20 Data Science Society of Auburn // AI, ML, and DS Concepts // November 16, 2020 42
- 43. Why Now? • Accessto data • 175 zettabytes annually by 2025 • 1 zettabyte = 1 trillion gigabytes • The Internet • Infrastructure to facilitate • Instrumentation of everything https://www.digitalinformationworld.com/2018/06/infographics-data-never-sleeps-6.html 11/16/20 Data Science Society of Auburn // AI, ML, and DS Concepts // November 16, 2020 43
- 44. Benefits of DataScience • Domain independent technology; “metascience” • Informs decision making • Empowers organizational learning • Improves operational efficiency • Leverages underutilized by-product of work • Delivers actionable results • Automates scientific discovery … a user’s data can be purchased for about half a cent, but the average user’s value to the Internet advertising ecosystem is estimated at $1,200 per year. Credit: Predictive Analytics, Eric Siegel, p. 54 11/16/20 Data Science Society of Auburn // AI, ML, and DS Concepts // November 16, 2020 44
- 45. Limitations and Pitfalls •Accurate prediction (extrapolation) is generally not possible. • “Prediction is very difficult, especially if it’s about the future.” – N. Bohr • High value from relatively low predictive power; targeted optimization • Does not answer WHY or HOW. • Correlation does not imply causation; many models opaque, empirical • Value comes from the prediction, not understanding cause • Vast search / Multiple comparisons trap • Possibility of being fooled by randomness - real trend or random artifact • Importance of domain knowledge and disciplined research • Bias / Variance tradeoff • Fit vs Predictive Power 11/16/20 Data Science Society of Auburn // AI, ML, and DS Concepts // November 16, 2020 45
- 46. Takeaways • Data Scienceis a broad, fast-moving field, with hype and confusion • The “promise” of AI cannot be met by current or near future tech • We are surrounded by current use cases, many more emerging • Its recent growth is fueled by data, compute, algorithms, sw, and $$$ • Leverages existing data to improve operational efficiencies • Identifies unexpected connections but does not explore causation • It is not fool-proof and requires expert oversight • Cannot be fully explained (even introduced) in one short talk… 11/16/20 Data Science Society of Auburn // AI, ML, and DS Concepts // November 16, 2020 46
- 47. Resources 11/16/20 Data ScienceSociety of Auburn // AI, ML, and DS Concepts // November 16, 2020 47
- 48. Additional Resources forDeep Learning • http://neuralnetworksanddeeplearning.com/index.html – free, online only, starts with writing a simple backprop NN from scratch in Python • https://www.manning.com/books/deep-learning-with-python - build models in Keras and Tensorflow, written by creator of Keras, 2nd edition coming soon! • https://course.fast.ai – alternative to Keras built on PyTorch, all work is done inside Jupyter Notebooks 11/16/20 Data Science Society of Auburn // AI, ML, and DS Concepts // November 16, 2020 48
- 49. Thank You. Contact Information: DanO’Leary dan.oleary@auburn.edu Blog / Portfolio / Links: bit.ly/aboutdjo 11/16/20 Data Science Society of Auburn // AI, ML, and DS Concepts // November 16, 2020 49 DATA SCIENCE SOCIETY OF AUBURN