1
Q
What is human intelligence?
A
Natural intelligence involving:
- Emotion
- Consciousness
- Creative and critical thinking
- Goal-directed actions
2
Q
What is Artificial Intelligence?
(AI)
A
- Machines performing tasks that typically require human intelligence.
- A branch of computer science focused on solving problems, performing tasks, and generating content.
3
Q
What is the Turing Test?
A
A test to determine whether a machine is considered “intelligent”.
The goal: for a machine to behave indistinguishably from a human, as judged by a human.
4
Q
Who was Alan Turing?
A
A British mathematician and cryptographer who proposed the Turing Test to assess machine intelligence.
5
Q
What are the shared features of AI definitions?
A
- Technology
- Intelligence
- Autonomy
- Goal-directed behavior
- Output
- Learning
- Human interaction
6
Q
What is a socio-technical system?
A
A system where humans shape AI/technology and are also shaped by it.
7
Q
For what reasons should AI teams be diverse and cross-functional?
A
To include perspectives from computer science, engineering, and social sciences for effective planning, design, and implementation.
8
Q
What is an algorithm in machine learning?
A
A set of instructions and rules to perform a task.
Example: ‘if X, then Y’.
9
Q
What is a corpus in machine learning?
A
A large collection of texts or data used by AI to find patterns and make predictions.
10
Q
What is inference in ML?
A
An ML model’s output, such as a decision or prediction.
11
Q
What is input data in ML?
Specifically, the data from which a model initially learns
A
Data provided to the ML model that forms the basis of its learning.
12
Q
What is labeled data?
A
Data that includes labels or tags giving context or meaning for the ML model.
13
Q
What is a machine learning model?
A
A learned representation of patterns and relationships in data.
14
Q
What is machine learning?
(ML)
A
A subfield of AI where algorithms learn from data to generate outputs, such as decisions or predictions.
15
Q
How does an algorithm work in ML?
A
It builds a model from input data and performs tasks on new data without being explicitly programmed.
16
Q
What is training data in ML?
A
Data used to train the model by identifying patterns and relationships.
17
Q
What are the four ML training models?
A
- Supervised
- Unsupervised
- Semi-supervised
- Reinforcement learning
18
Q
What is supervised learning?
A
An ML method where labeled data is used to train a model to classify inputs or predict values.
19
Q
Why does supervised learning use the word “supervised”?
A
Because a ‘supervisor’ (human) labels the data before it’s used to train the model.
20
Q
What is data labeling in supervised learning?
A
Enriching data with labels for training, validation, and testing purposes.
21
Q
Why is training data quality important?
A
Model accuracy and performance depend on the quality and volume of labeled training data.
22
Q
What is a classification model?
A
An ML model that identifies the correct categorical label, such as spam vs. ham.
23
Q
What is a regression model?
A
A supervised learning technique that predicts continuous values like house price or weight.
24
Q
What is Support Vector Machine used for?
(SVM)
A
Classification tasks in supervised learning.
25
What is **Support Vector Regression** used for?
| (SVR)
**Regression tasks** in supervised learning.
26
What is **unsupervised learning**?
| (As compared to supervised learning.)
Machine learning that processes **unlabeled data** to extract features or patterns **without human-labeled input**.
27
How does unsupervised learning **differ** from supervised learning?
Unsupervised learning **doesn't rely on labeled data** and is more useful for **exploratory data analysis**.
28
What are the two **subcategories** of unsupervised learning?
* Clustering
* Association rule learning
29
What is **clustering** in unsupervised learning?
A technique that **groups similar or identical data points** into clusters.
30
What is **association rule learning**?
A data mining method that **identifies patterns or relationships in data**.
## Footnote
Example: supermarket design **grouping items like bread and milk** that are often bought together.
31
What is **semi-supervised learning**?
An approach **combining labeled and unlabeled data** to reduce costs and improve reliability.
## Footnote
Example: Large Language Models (LLMs) using **small labeled datasets** and **large unlabeled datasets**.
32
What is **reinforcement learning**?
A machine learning method that **simulates motivation using rewards and punishments** with the goal to maximize reward.
33
What does **RLHF** stand for?
Reinforcement Learning from Human Feedback.
34
Does reinforcement learning use pre-labeled data?
No
## Footnote
It learns through **feedback** from **interaction with the environment**.
35
What is the **objective** of reinforcement learning?
To learn actions that **maximize cumulative reward** over time.
36
What are the **5 steps** in the reinforcement learning process?
* Perceive state
* Learn or improve action policy
* Take action
* Observe new state
* Receive reward
37
How does the **learning feedback loop** work in reinforcement learning?
* Action > reward > continue or refine, action > punishment > change action
* Rewards and penalties are proportionate to the action.
38
Why are **reward signals** in reinforcement learning difficult to interpret?
Because they provide **only partial information** and do not indicate which actions specifically led to success.
39
Where does reinforcement learning **work well**?
In games and simulations where **rules, feedback, and repetitions are clearly defined** and controlled.
40
What are the **challenges** of applying reinforcement learning in **real life**?
* Real-world time **cannot be sped up**
* Inputs and outputs are **complex**
* Failure has **consequences**, such as autonomous vehicle crashes
41
What is the main **goal** of a **discriminative model**?
To **classify data points** by mapping input features to specific class labels.
42
What is an **example** of a discriminative model task?
Telling whether an image is a **cat, dog, sloth, or koala**.
43
What does a discriminative model use to **make decisions**?
Patterns in data to draw a **decision boundary** between classes.
44
What is the main **goal** of a **generative model**?
To **generate new data points** by learning the underlying characteristics of the data.
## Footnote
Example: **Generating an image** of a cat based on its learned characteristics.
45
Is a generative model focused on **classification**?
No
## Footnote
It is focused on **generating new content** such as text, image, video, or audio.
46
What is a **neural network**?
A machine learning model **inspired by biological neurons**, includes **at least one hidden layer** and can model complex nonlinear patterns.
47
What is a **foundation model**?
A **large scale neural network** trained on massive data that **can be reused** for different but related tasks.
48
What is **transfer learning**?
A technique where **an algorithm learns one task** and **applies the knowledge to a different**, **related task**.
49
What is **fine tuning** in foundation models?
**Further training** of a pre-trained model on **specific data** to perform a **specialized task**.
50
What are the **advantages** of foundation models?
* Save time and resources
* Adaptable
* Generalizable, and
* Scalable
51
What are **different types** of foundation models?
* Large Language Models
* Vision models
* Scientific models
* Audio models
* Transformer models
* Multimodals models
52
What is an **example** of a scientific foundation model?
**[AlphaFold3](https://www.youtube.com/watch?v=J5XFYMsczy8)**, which is used for protein folding.
53
What are some examples of **specialized tasks** a foundation model might perform after being **fine tuned**?
* Summarize documents
* Write poetry
* Generate code
* Solve math problems
* Synthesize audio
54
What are the **4 high-level categories** of artificial intelligence?
* Artificial **Narrow** Intelligence
* **Broad** Artificial Intelligence
* Artificial **General** Intelligence
* Artificial **Super** Intelligence
55
What is Artificial **Narrow** Intelligence?
* Also called **weak AI**
* It performs **a single function under narrow constraints** at a high level.
## Footnote
Examples: Deep Blue (chess program, 1997), AlphaGo (2016); Check out the ["AlphaGo" documentary](https://www.youtube.com/watch?v=WXuK6gekU1Y&t=1s).
56
What is **broad AI** within the context of Artificial Narrow Intelligence (ANI)?
A **more advanced form of ANI** where multiple narrow systems work together.
57
What is Artificial **General** Intelligence?
* Also called strong or full AI
* It **matches human intelligence** and can generalize, infer from new data, and complete complex goals.
58
What is Artificial **Super** Intelligence?
AI with intellectual capabilities that **far exceed human intelligence**, consciousness, and emotional expression.
59
What is the **goal** of **expert systems**?
To **emulate human decision-making** and provide support rather than replace humans.
60
What is an **example** of an expert system?
Tax preparation software.
61
What are the **3 components** of an **expert system**?
1. Knowledge base
2. Inference engine
3. User interface
62
What is the function of the **knowledge base** in an expert system?
It **stores organized facts** about a specific domain.
63
What is the function of the **inference engine** in an expert system?
It **uses rules to locate facts** in response to a user prompt.
64
What is the function of the **user interface** in an expert system?
It allows the user to **input questions** and **receive output**.
65
What is **robotics** in the context of AI?
* A **multidisciplinary field** involving the design, construction, operation, and programming of robots
* Allows AI systems to **interact with the physical world**.
66
What is **Industry 4.0**?
The **Fourth Industrial Revolution** marked by robotics, automation, and **minimal human intervention in production**.
67
What are **key features** of Industry 4.0?
* Increased interconnectivity
* Automation
* Robotics with little or no human intervention
68
What is **machine perception**?
The use of **sensors** and AI to enable machines to **perceive the environment** like humans.
69
What **types of sensors** are used in machine perception?
* 3D scanners
* Microphones
* Motion sensors
* Cameras
* Pressure sensors
* Thermal imaging
70
What is **Robotic Process Automation**?
**Software robots** that automate repetitive tasks such as data entry and forms processing.
71
What technologies **enhance** Robotic Process Automation?
* Machine learning
* Natural language processing
72
What do **linear and statistical models** do?
**Model relationships** between **two variables** such as temperature and ice cream sales.
73
What is a **decision tree**?
A model structured as a **question and answer flowchart**.
74
What is **deep learning**?
A subfield of artificial intelligence based on **neural networks with more than three layers**, often a black box.
75
What is a **neural network**?
A model with **layers of artificial neurons** used to detect patterns and make predictions.
76
What is **computer vision**?
Technology that allows AI to **recognize and interpret images and video**.
77
What is **natural language processing**?
A subfield of artificial intelligence that helps computers **understand, interpret, and generate human language**.
78
What is **speech recognition**?
A task within natural language processing that **converts spoken language into text**, used in voice assistants.
79
What are some tasks within **NLP**?
* Part of speech tagging
* Word sense disambiguation
* Sentiment analysis
* Speech recognition
80
What is a **large language model**?
| (LLM)
A type of **foundation model** that uses deep learning and is **trained on massive text datasets** to **understand and generate language**.
81
What does **large** mean in large language model?
It refers to the **number of parameters** in the model such as 175 billion in ChatGPT-3.
82
What is a **parameter** in machine learning?
An **internal variable** learned from training data that helps make predictions on new data.
83
What is the **difference** between **parameters** and **hyperparameters**?
* Parameters are **learned during training**
* Hyperparameters are **set before training** to control the process
84
What are **features** and **weights** in a neural network?
* Features are **input variables**
* Weights **determine strength of connections** between nodes and are adjusted during training
85
What are the **2 types** of large language models?
* Generative
* Discriminative
86
What does **GPT** stand for?
Generative Pre-trained Transformer
87
What does the **attention** technique do in GPT?
It **evaluates the importance of each word** in a sentence based on its context.
88
What does **multi-modal** mean in AI?
The ability to process and produce input and output across **multiple media** such as text, speech, image, video.
89
What is a **chatbot**?
An AI designed to **simulate human-like conversation** and interaction.
90
What are **examples** of popular chatbot models?
* **Claude** by Anthropic
* **ChatGPT** by OpenAI
* **Gemini** by Google
* **Llama** by Meta
91
What is a **small language model**?
A specialized AI model designed for natural language processing with a **compact architecture** and **fewer parameters**.
92
How does the parameter count of an SLM **compare** to an LLM?
SLMs have **fewer** parameters, while LLMs like GPT-3 and GPT-4 have **hundreds of billions**.
93
What are the **advantages** of small language models?
* Smaller file size
* Faster training time
* Cost-effective implementation
* Reduced data storage needs
* Enhanced security
94
Why do small language models **enhance security**?
Because they can be deployed on-premise or on private clouds, **reducing external exposure**.
95
What does **recommendation** mean in AI?
Proposing **suggestions or new content** based on user behavior or history.
96
What are **examples** of recommendation systems?
* Social media
* E-commerce
* Entertainment streaming
* Medical diagnoses
* Legal adjudication
97
What does **recognition** mean in AI?
**Identifying** images, speech, facial or palm features.
## Footnote
Goal: to determine if a specific object or concept is present. E.g., you upload a photo to Google Photos and it tags an image as a "dog".
98
What are **examples** of recognition use cases?
* Facial recognition
* Product matching
* Defect identification
* Plagiarism detection
99
What does **detection** mean in AI?
1. Determine **where** an object is
2. Identify statistical anomalies
100
What is the **difference** between recognition and detection?
* Recognition identifies **what** is in an image
* Detection identifies **where** the object is
101
What does **forecasting** mean in AI?
**Predicting future outcomes** such as sales, revenue, or demand.
102
What does **goal-driven optimization** mean?
**Determining the best path or steps** from one point to another.
103
What are **examples** of goal-driven optimization?
* Travel route planning
* Supply chain management
* Games like AlphaGo
104
What is **interaction support** in AI?
Helping users through **customer support tools** like virtual assistants or chatbots.
105
What does **personalization** mean in AI?
Adapting experiences or content to **match user preferences** or behavior.
106
What does **compute** mean in the context of AI?
The **processing resources of a computer** including CPU, GPU, memory, storage, and data handling.
107
What are the **4** **main components** of the **AI tech stack**?
1. Compute
2. Storage
3. Network
4. Software
108
What is a **supercomputer**?
A **very powerful computer** with specialized processors like CPU and GPU designed for **high-speed operations** measured in FLOPS.
109
What does **FLOP/s** stand for?
**Floating Point Operations per Second**, a measure of computational speed at a single moment.
110
What is the difference between **FLOP/s** and **FLOPs**?
* FLOP/s measures **speed per second**.
* FLOPs measures **total operations over time**.
111
What is **serverless compute**?
* A model where code **runs on various devices** without being tied to one server
* Features include **loose coupling** and **scaling**
112
What is **high-performance compute**?
**Isolated clusters of compute** that use **high-speed networking** and specialized chips to process data efficiently.
113
What is a **trusted execution environment**?
A **secure part of a processor** that protects the confidentiality and integrity of data and code.
114
Why do different storage stages have **different requirements**?
Each stage handles **different types and volumes** of data with varying speed and capacity needs.
115
What are important **storage considerations**?
* Expense
* File type, such as image or text or video
* Structured vs. unstructured data
116
What is the **difference** between **structured** and **unstructured** data?
* Structured data is easier to process because it is **organized**
* Unstructured data is **not organized** and is harder to process
117
What is required for **effective** AI networking?
**High-speed network** such as 10 gigabit Ethernet for data delivery, training, and inference.
118
Why are high-speed networks typically **located in the same data center**?
To **eliminate congestion** and improve performance by **minimizing latency**.
119
What are **alternatives** to centralized AI networking?
* Edge computing
* Internet of Things
120
What **network protocol** is commonly used for **data transmission**?
Transmission Control Protocol
| (TCP)
121
What is an **application** in the AI tech stack?
The **way the AI system is used**, such as in autonomous vehicles, chatbots, or finance.
122
What are **examples** of AI applications?
* Autonomous vehicles
* Chatbots
* Facial recognition
* Finance
* Health care
* Marketing
* Social media
123
What is an AI **platform**?
Software that helps **plan, design, develop, implement, and deploy** AI systems.
## Footnote
Examples: Amazon Web Services, Microsoft Azure, Google Cloud
124
What **functions** do AI platforms support?
* Data analysis
* Streamlining development and workflows
* Collaboration
* Automation
* Monitoring
125
What is **open source software** in AI?
Software with **viewable, modifiable code**, enabling decentralized development and innovation.
126
What are the **benefits** of open source software in AI?
It **supports experimentation, idea sharing, best practices, and accessibility** for those with limited resources.
127
What is **fine tuning** in AI?
**Customizing a foundation model** by training it on specific data to serve a targeted application.
128
**When** should you **fine tune** a model?
* To **adapt** to a new domain or data
* **Improve** task-specific performance, or
* **Customize** output
129
**How** do you fine tune a model?
* Start with pre-trained model
* Gather specific data
* Pass it through model
* Compare outputs to expected outputs
* Adjust parameters
* Repeat
130
What is **accuracy** in AI?
A **primary indicator of model performance** measuring correctness, effectiveness, and success in completing tasks.
131
How does **mobile technology** drive AI growth?
Through **consumer internet access and sensors** that generate large volumes of data.
132
What is the **metaverse**?
A **virtual space for interaction** and business that contributes data for AI.
133
How does cloud computing **support** AI?
By providing **on-demand, scalable infrastructure** for data storage and processing.
134
What is the **role of computer vision** in AI?
It enables computers to **interpret and process visual inputs** such as images and video.
135
What is **AR**?
**Augmented reality** overlays digital elements onto the real world.
136
What is the **Internet of Things**?
A **network of internet connected devices and sensors** that continuously generate data.
137
What are **privacy enhancing technologies**?
| (PETs)
Tools that allow data to be processed while **preserving privacy**, especially for personal information.
138
How does social media **contribute** to AI?
It **increases available data** on preferences, habits, lifestyle, and consumption patterns.
139
What is **blockchain**?
A technology that uses an **append-only transaction log** to record data securely.
140
What are the **4 storage stages** of the data pipeline?
* Ingestion
* Preparation
* Training
* Output
AIGP
flashcards
Decks in class (8)
# Cards
