Chapter 2: Infrastructure and Tools For Ai Flashcards
What does an operator take care of?
Conditions and everything that affects them
It consists of actions, preconditions, and changes resulting from taking actions.
What is a rational agent?
An entity that takes actions to achieve certain goals
The text discusses making rational agents more intelligent.
What are common techniques to impart intelligence to an agent?
- Machine learning
- Stored knowledge
- Rules
The focus is on machine learning in the text.
What is the role of labeled data in machine learning?
To solve a given problem by allowing machines to learn patterns
Machines extract patterns and relationships from labeled data.
What are the steps that an intelligent agent takes when using machine learning?
- Sensor perceives input
- Sends to feature extraction block
- Trained inference engine performs prediction
- Inference engine sends decision to actuator
This process results in taking required action in the real world.
Name some applications of machine learning.
- Image recognition
- Robotics
- Speech recognition
- Predicting stock market behavior
Machine learning is widely used in various fields.
What are the two types of models in AI?
- Analytical models
- Learned models
The text discusses the transition from analytical to learned models.
What characterizes analytical models?
Derived using mathematical formulation and human judgment
They are often simplistic and inaccurate.
What defines learned models?
Obtained through training, analyzing many examples of inputs and outputs
These models are complex and accurate.
What is a significant advantage of machine learning over analytical models?
No need to derive the underlying mathematical formula
Machines derive the formula based on data.
What is the first command to enter the Python command prompt?
$ python3
This command starts the Python interpreter.
What did Alan Turing ask in 1950?
Can machines think?
This question remains relevant in AI discussions today.
What does ML stand for?
Machine Learning
It is a key area of focus in applied AI.
What is the difference between strong AI and weak AI?
- Strong AI: Refers to AGI (Artificial General Intelligence)
- Weak AI: Refers to ANI (Artificial Narrow Intelligence)
Weak AI is what is generally meant in most product contexts.
True or False: AGI exists today.
False
AGI does not exist; current AI is narrow in application.
What is the main characteristic of ANI?
Narrow application of AI that is good at specific tasks
Its expertise does not extend to other areas.
What is a rules-based engine?
A system that replicates human decision-making without learning
It is considered AI but not ML.
What is the main challenge in AI adoption mentioned in the text?
Finding clear information about what makes a product AI
Marketing often complicates the understanding of AI.
What does AGI stand for?
Artificial General Intelligence
AGI refers to a type of AI that can understand, learn, and apply intelligence across a wide range of tasks, similar to human cognitive abilities.
True or False: Sentient AI currently exists.
False
Sentient AI does not exist, and the statement emphasizes the importance of ethical AI applications.
What are the consequences of unethical applications of AI?
They cause minor inconveniences and major upsets
Unethical applications of AI can lead to significant harm in various contexts.
What are the two essential components of Machine Learning (ML)?
Models and training data
The models are used to learn from historical data points.
What is the role of data scientists and ML engineers in ML?
They choose, build, tune, and maintain models for optimized performance
Their work is crucial for ensuring that ML models function effectively.
What are the four major learning categories in ML?
- Supervised learning
- Unsupervised learning
- Semi-supervised learning
- Reinforcement learning
Each learning type has specific models and algorithms associated with it.
What differentiates Deep Learning (DL) from Machine Learning (ML)?
DL is based on neural network algorithms, while ML encompasses other algorithms
DL is considered a subset of ML.
What does the term ‘feature selection’ refer to in the context of DL?
The neural networks learn to pick up patterns or features from the data autonomously
This contrasts with traditional ML where engineers select features.
What is supervised learning?
Humans label the data, and machines attempt to label current or future data points
This method involves continuous training to improve model accuracy.
Name one application of supervised learning models.
- Classification models (e.g., spam filters)
- Regression models (e.g., predicting trends)
These applications require continuous training and updating.
What is the Naive Bayes algorithm used for?
Classification problems
It considers each feature in the dataset as an independent variable.
What does SVM stand for?
Support Vector Machine
SVM is used for classification by splitting datasets into two classes.
What is the purpose of linear regression?
To predict future data points based on one or more variables
It finds the best line that fits the data.
What does logistic regression predict?
A future binary categorical state
Examples include predicting loan defaults.
How does a decision tree model work?
It uses nodes and branches to learn from past data to predict future values
Decision trees are popular due to their interpretability.
What is a random forest?
An ensemble of decision trees used for both categorical and numerical predictions
It averages or votes for predictions based on multiple trees.
What characterizes unsupervised learning?
The data is unlabeled, and machines find patterns independently
This method is often used when the correct answers are unknown.
Name one application of unsupervised learning.
- Clustering
- Dimensionality reduction
Clustering identifies groups in data, while dimensionality reduction simplifies datasets.
What does PCA stand for, and what is its purpose?
Principal Component Analysis; reduces dimensions without losing information
PCA is useful for analyzing large datasets.
What is semi-supervised learning?
A model trained with both labeled and unlabeled data
It helps guide the model towards finding patterns when labeled data is scarce.
What is the primary challenge of overfitting in ML?
When a model fits too well to a specific dataset and performs poorly on new data
This is a common issue in data science.
What is semi-supervised learning?
A learning method that uses a small amount of labeled data alongside a larger amount of unlabeled data to improve model performance
The process involves predicting on unlabeled data and checking accuracy against the labeled data.
How does reinforcement learning operate?
It learns through trial and error, adapting its approach based on past behavior and optimizing for rewards
Commonly used in robotics to help machines adjust to real-world parameters.
What are the main components of operationalizing AI/ML?
- Choosing models and use cases
- Managing production
- Updating models
- Keeping data fresh and clean
- Organizing experiments
- Validating and testing
This process is complex and often a challenge for companies outside the tech industry.
What is the purpose of an ETL pipeline?
To make layers of data and metadata available for AI/ML models to ingest and offer insights from
ETL stands for Extract, Transform, Load.
What are the four major components of the continuous maintenance process in AI?
- Testing/validating code and components
- Continuous code updates
- Continuous learning from new data
- Monitoring model performance
This ensures that models do not become stale and maintain effective performance.
Why is proper data storage critical for AI/ML products?
It impacts the performance of models and the overall product
Poor data storage can lead to significant operational issues.
What are the differences between a data lake, database, and data warehouse?
- Data Lake: Stores raw data in its native format
- Database: Structured data for easy access and querying
- Data Warehouse: Centralizes structured data for analysis and insights
Each serves different purposes based on organizational needs.
What is the benefit of using a data warehouse for AI/ML?
It allows for quick leverage of insights and trends across various business units
Essential for organizations looking to implement AI/ML across multiple functions.
True or False: A relational database is ideal for combining various datasets for advanced analytics.
False
Relational databases can struggle with aligning schemas when combining data from different sources.
Fill in the blank: Continuous maintenance in AI involves ______ to ensure models are effective.
iterating the process
Stagnation can lead to outdated models and ineffective performance.
What is a potential consequence of improper model maintenance?
Loss of jobs or unfair outcomes, such as incorrect mortgage rates or prison sentences
This highlights the importance of responsible AI management.
