Statistical Learning
Course Overview
1
Outline
• What’s in the course
• Course logistics
2
Part I: What’s in the course?
3
Classify Images
[Link]
ai-research-and-possibly-the-world/ 4
Detect and Segment Objects
[Link]
5
Style Transfer [Link]
6
Generative Modeling
4.5 years of face generation
7
[Link]
Machine Translation
[Link]
8
[Link]
Text Synthesis
[Link]/berkeley-stat-157 9
OpenAI ChatGPT (in 2022):
me: why is carbon dioxide highly sought after?
ChatGPT: In recent years, CO2 has also become highly sought
after as a potential source of renewable energy. When burned,
CO2 releases a large amount of heat, which can be harnessed to
generate electricity. It can also be used to power fuel cells, which
are a type of clean energy technology that converts chemical
energy into electricity.
(Accessed in Dec 2022 by Jerry Zhu)
10
OpenAI ChatGPT (in 2022):
me: which molecules are produce when burning co2?
ChatGPT: When carbon dioxide (CO2) is burned, it reacts with
oxygen in the air to produce carbon monoxide (CO) and water
(H2O). The chemical reaction for this process can be written as:
CO2 + O2 -> CO + H2O
The exact products of the combustion of CO2 depend on the
conditions under which it is burned, such as the temperature,
oxygen concentration, and presence of other chemicals.
However, in general, CO and H2O are the primary products of
the burning of CO2. 11
(Accessed in Dec 2022 by Jerry Zhu)
OpenAI ChatGPT (in 2022):
me: why is salt highly sought after in the desalination process?
ChatGPT: Salt, or sodium chloride, is highly sought after in the desalination process because it is a readily
available and relatively inexpensive compound that can be used to remove dissolved salts from seawater.
Desalination is a process that is used to produce fresh water from seawater, and it is often necessary in
areas where there is a shortage of fresh water.
Salt is used in the desalination process because it has a unique chemical property that allows it to readily
dissolve in water. When seawater is mixed with salt, the salt ions will dissolve and become evenly distributed
throughout the water. This creates a solution that has a higher concentration of salt ions than the seawater.
The higher concentration of salt ions in the solution makes it possible to use a process called reverse
osmosis to remove the dissolved salts from the seawater. In reverse osmosis, the salty solution is forced
through a membrane that only allows water molecules to pass through, effectively separating the salt from
the water. The resulting water is fresh and drinkable, and the salt can be recovered and used again in the
desalination process.
Overall, salt is highly sought after in the desalination process because it is a key component in the reverse
osmosis process, which is the most commonly used method for removing dissolved salts from seawater.
12 (Accessed in Dec 2022 by Jerry Zhu)
Will AI kill me?
Improbable
• Sentient AI
ChatGPT on July 12, 2023
but…
• Sentient AI
• Dual use
[Dual use of artificial-intelligence-powered drug discovery. Urbina et al, 2022]
Spearphishing powered by AI
Your loved one … in car accident
3 seconds send gift card,
… been robbed
voice cloning wire transfer
… lost wallet
[Artificial Imposters—Cybercriminals Turn to AI Voice Cloning for a New Breed of Scam. Bunn, 2023]
Will AI take my job?
Maybe
• The more AI helps your job, the higher the replacement risk
High
...
Low
[GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models. Eloundou et al. 2023]
Will AI help my job?
GPT, large language model, generative AI, deep learning…
Oh my!
AI Venn diagram
Artificial Intelligence
AI Venn diagram
Artificial Intelligence
Machine learning
AI Venn diagram
Artificial Intelligence
Machine learning
Natural language processing
AI Venn diagram
Artificial Intelligence
Machine learning
Natural language processing
Computer vision
AI Venn diagram
Artificial Intelligence
Machine learning
Natural language processing
Computer vision
Robotics
AI Venn diagram
Artificial Intelligence
Machine learning
Deep learning with Artificial neural networks
Natural language processing
Computer vision
Robotics
AI Venn diagram
Artificial Intelligence
Machine learning
Deep learning with Artificial neural networks
Transformer (ANN structure)
Natural language processing
Computer vision
Robotics
AI Venn diagram
Artificial Intelligence
Machine learning
Deep learning with Artificial neural networks
Transformer (ANN structure)
Large
Language
Models
Natural language processing
Computer vision
Robotics
AI Venn diagram
Artificial Intelligence
Machine learning
Deep learning with Artificial neural networks
Transformer (ANN structure)
Large
Language
Models
GPT
Natural language processing
Computer vision
Robotics
AI Venn diagram
Artificial Intelligence
Machine learning
Deep learning with Artificial neural networks
Transformer (ANN structure)
Large
Language
Model
GPT
Natural language processing Generative AI
Computer vision
Robotics
Artificial Intelligence is not Magic
They rely on fundamental techniques in:
• Probability and Statistics
• Optimization
• Algorithms
• Logic
• Systems
• …
32
What you can learn from the course?
Learning Outcomes:
• Demonstrate an understanding of the conceptual underpinnings of various statistical learning
techniques in terms of how, why and when each method works in different real-life scenarios
• Critically evaluate the analytical strategies adopted in applying statistical learning techniques to
different areas
• Apply basic statistical learning methods to perform data analysis and build predictive models using the
statistical programming environment R
• Properly tune, select, and validate statistical learning models
• Interpret the results and discuss their implication
33
What you can learn from the course?
Focus more on WHY,
Learning Outcomes: rather than HOW
• Demonstrate an understanding of the conceptual underpinnings of various statistical learning
techniques in terms of how, why and when each method works in different real-life scenarios
• Critically evaluate the analytical strategies adopted in applying statistical learning techniques to
different areas
• Apply basic statistical learning methods to perform data analysis and build predictive models using the
statistical programming environment R
• Properly tune, select, and validate statistical learning models
• Interpret the results and discuss their implication
34
What you can learn from the course?
Content Themes:
• Foundational tools in Statistical Machine Learning: elements of Probability, Statistics, and
Linear algebra
• Basics of Machine Learning: supervised learning, unsupervised learning, reinforcement
learning; brief introduction to neural networks and deep learning
• Principles of Machine Learning: statistical learning framework, Bayes’ rule, MLE/MAP,
feature engineering, regularization, generalization
• Application Examples of Machine Learning in real-world settings
35
What you can learn from the course?
TL;DR Lots of useful stuff in statistical learning
•
•
36
What you can learn from the course?
Lecture Topics:
Foundations
37
What you can learn from the course?
Lecture Topics:
Supervised Learning
y2
x3 x2
x1 38
What you can learn from the course?
Lecture Topics:
Unsupervised
and Reinforcement
Learning
39
What you can learn from the course?
Lecture Topics:
Training Data
Food Image Classifier
Additional Topics
40
Part II: Course Logistics
41
Where to find everything?
• Moodle
• grades, and other private materials that should not be shared
• Slides, syllabus, policies
• Discussion, questions, announcements
42
Instruction Team
Instructor: Yingyu Liang
• Email: yingyul@[Link]
• Office: Room P307, Graduate House
• Office hours: Thursday 2-3pm
43
Instruction Team
TA: Yujin Han
• Email: yujinhan@[Link]
• Office: TBA
• Office hours: TBD
44
No Textbook
Useful References (reference is optional, but may be useful)
46
Grading scheme
• Continuous Assessment: 50% (homework + potentially midterm around mid Oct)
• Written Examination: 50% (final exam)
• Homework assignments:
• about 5 assignments, bi-weekly
• Mostly written questions
47
Thanks!
48
