Fundamentals of AI Flashcards
What is a neural network?
A machine learning model inspired by biological neural networks that constitute animal brains.
What are the layers of a neural network?
Structured in layers, including input layer, hidden layers, and output layer.
What is the function of neurons in a neural network?
They receive input, apply weights, and use an activation function to output a value.
What is the input layer in a neural network?
The first layer of neurons that receives the raw data.
What does each neuron in the input layer represent?
A feature or variable of the input data (e.g., pixels in an image).
What are hidden layers in a neural network?
Layers between the input and output that process data by applying transformations and computations.
What determines the number of hidden layers and neurons in a neural network?
The complexity of the problem.
What is the output layer in a neural network?
The final layer, which provides the result or prediction.
What type of result does the output layer provide in an image recognition task?
A classification label.
What are weights in a neural network?
Weights are crucial parameters that control the strength and direction of the connection between nodes (neurons)
How do weights affect the final output in a neural network?
Weights determine the impact of an input by being multiplied to the input
What is the significance of high positive weights in neural networks?
High positive weights indicate inputs that hold more importance
What role does bias play in a neural network?
Bias helps the model make better predictions by shifting the activation function
How does bias affect the activation function?
Bias allows the activation function to be shifted left or right, capturing patterns more effectively
What would happen without a bias term in a neural network?
The neural network would be restricted in learning patterns
What is the purpose of an activation function in a neural network?
It determines whether a neuron should activate based on its inputs
Why are activation functions essential for neural networks?
They enable the model to capture complex, non-linear patterns and support deeper learning
What is the sigmoid activation function used for?
The sigmoid function maps values to a range between 0 and 1, useful for binary classification
What is the output of the ReLU activation function for negative values?
Outputs zero for negative values
What is the output of the ReLU activation function for positive values?
Outputs the input value for positive values
In which types of networks is ReLU widely used?
ReLU is widely used in convolutional neural networks (CNNs) and feedforward networks
Fill in the blank: Weights are numerical values that determine the _______ of an input to a neuron.
importance
True or False: Without activation functions, neural networks can effectively model real-world data.
False
What is a Feedforward Neural Network (FNN)?
The simplest type of artificial neural network structured with layers of nodes in a sequential, unidirectional manner
How does information flow in a Feedforward Neural Network?
Information flows forward from input to output through hidden layers without any feedback loops
What components do each layer in an FNN consist of?
A set of neurons that process the data by applying weights, biases, and activation functions
What types of problems are FNNs ideal for?
Problems where inputs and outputs have no sequential dependency, such as image recognition, regression, and binary classification
What limitations do FNNs have?
They have limited ability to handle complex dependencies, time-series data, or structured hierarchies
In what way are FNNs used in more complex network architectures?
FNNs are often integrated as components within more complex network architectures
What is a Convolutional Neural Network (CNN)?
A specialized type of deep learning model primarily used for image data processing
What is the primary function of CNNs?
To automatically and adaptively learn spatial hierarchies in images for effective analysis of spatial and temporal patterns
What layers are involved in a CNN and what do they do?
Convolutional layers apply filters to capture specific features, while pooling layers reduce spatial dimensions
What is the purpose of pooling layers in CNNs?
To preserve important information while decreasing computation by reducing spatial dimensions
What tasks are CNNs highly effective in?
Image classification, object detection, and segmentation
Why is the architecture of CNNs well-suited for image data?
It is designed to handle the grid-like topology of image data
True or False: FNNs are suitable for tasks with sequential dependencies.
False
Fill in the blank: CNNs consist of convolutional layers that apply a set of ______ across the input data.
[filters or kernels]
What enables CNNs to learn complex visual patterns?
The feature extraction process as the network deepens
What is a key advantage of CNNs in computer vision applications?
Their ability to detect higher-level representations
What are recurrent neural networks (RNN)?
RNNs are designed to handle sequential data and temporal dependencies.
What tasks are RNNs particularly suited for?
RNNs are suited for tasks involving time series, natural language processing, and audio data.
How do RNNs differ from other neural networks?
RNNs have connections that loop back to previous nodes, allowing information to persist across time steps.
What enables RNNs to remember previous inputs?
The looping connections allow RNNs to remember previous inputs.
What problem do traditional RNNs face with long-term dependencies?
Traditional RNNs struggle with the vanishing gradient problem.
What is the vanishing gradient problem?
It is when gradients diminish as they backpropagate through time.
What are two specialized RNN variants that address long-term dependencies?
Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) networks.
What do LSTM and GRU networks use to retain relevant information?
They use gating mechanisms.
In what tasks do RNNs excel due to their gating mechanisms?
RNNs excel in tasks requiring an understanding of context over time, such as speech recognition and language translation.
What is one drawback of neural networks related to data?
Neural networks often require large datasets for accurate results.
What is a computational resource requirement for neural networks?
They need significant processing power, often utilizing GPUs.
True or False: Neural networks are always interpretable.
False
What is a challenge regarding the decision-making process of neural networks?
They are not always interpretable, making it hard to understand their decision-making process.
What is artificial intelligence (AI)?
A discipline that aims to create systems capable of performing tasks that usually require human intelligence, such as reasoning, decision-making, pattern recognition, and language comprehension.
What is machine learning (ML)?
A subset of AI that focuses on enabling machines to learn from data rather than being explicitly programmed.
What do ML models do?
Identify patterns within data using statistical techniques, helping systems make predictions or decisions based on that data.
What is deep learning (DL)?
A more advanced form of ML inspired by the structure of the human brain, using artificial neural networks with multiple layers.
What types of data is deep learning particularly effective at analyzing?
Unstructured data, such as images, audio, and text.
In what fields has deep learning achieved breakthroughs?
Computer vision and natural language processing.
What are the requirements for deep learning models?
Substantial amounts of data and computational power.
What tasks do deep learning models excel at handling?
Recognizing faces, translating languages, and generating original content.
Fill in the blank: Machine learning models help systems make predictions or decisions based on _______.
[data]
True or False: Deep learning uses simple algorithms to process data.
False
Fill in the blank: Deep learning is inspired by the structure of the _______.
[human brain]
What does Generative AI focus on?
Creating new content such as images, text, and music by learning patterns from existing data.
How does Generative AI differ from traditional Machine Learning?
Generative AI synthesizes information, while traditional ML mainly classifies or predicts.
List three fields where Generative AI is useful.
- Creative arts
- Content generation
- Personalized marketing
What are two models that underpin Generative AI?
- Generative adversarial networks (GANs)
- Variational autoencoders (VAEs)
What is one application of Generative AI in drug discovery?
Designing synthetic media.
True or False: Generative AI empowers systems to produce original outputs.
True
What is the primary focus of computer vision within AI?
Enabling machines to interpret and understand visual information from the world.
Fill in the blank: Generative AI is a specialized area of ML and DL that focuses on _______.
[creating new content]
What is natural language processing (NLP)?
An area of AI that enables machines to understand, interpret, and generate human language.
NLP combines machine learning (ML) and deep learning (DL) techniques.
What techniques does NLP use?
A combination of machine learning (ML) and deep learning (DL) techniques.
These techniques analyze text and speech.
What are some applications of NLP?
Applications include:
* Chatbots
* Translation services
* Sentiment analysis
* Voice recognition
These applications leverage NLP to facilitate human-computer interaction.
What are large language models (LLMs)?
Advanced NLP systems trained on vast amounts of text data.
LLMs can generate coherent and contextually relevant text based on input.
Name two examples of large language models.
Examples include:
* BERT
* ChatGPT
These models are capable of various NLP tasks.
What tasks can LLMs like BERT and ChatGPT perform?
Tasks include:
* Answering questions
* Summarizing text
* Engaging in extended conversations
LLMs excel in generating human-like responses.
What is labeled data?
Data that comes with predefined tags or annotations identifying characteristics or categories of each data point
For example, in image recognition, labeled data includes images with labels like ‘dog’ or ‘cat’.
Why is labeled data essential in supervised learning?
It allows the model to learn by example, associating specific input with the expected output.
In what applications is labeled data commonly used?
- Fraud detection
- Medical diagnosis
What is unlabeled data?
Data that lacks explicit labels, requiring the model to identify patterns or clusters without prior categorization.
In which learning paradigm is unlabeled data typically used?
Unsupervised learning.
What tasks can benefit from using unlabeled data?
- Customer segmentation
- Anomaly detection
True or False: Unlabeled data is more readily available and less costly to collect than labeled data.
True.
Fill in the blank: Labeled data is essential in _______ learning.
[supervised]
Fill in the blank: Unlabeled data is typically used in _______ learning.
[unsupervised]
What is structured data?
Highly organized data that fits neatly into predefined formats like rows and columns
What types of information does structured data usually include?
Quantitative information and clear attributes
Give an example of structured data.
A dataset of customer details with attributes like age, location, and purchase history
In which applications is structured data particularly useful?
Finance, healthcare, and business analytics
What tasks benefit from structured data?
- Predictive analytics
- Customer profiling
- Risk assessment
What is unstructured data?
Data that doesn’t have a predefined format and is more challenging to process
List examples of unstructured data.
- Text
- Images
- Audio
- Video files
Why is unstructured data valuable for AI tasks?
It requires the interpretation of complex content
Name three AI tasks that utilize unstructured data.
- Natural language processing
- Image recognition
- Sentiment analysis
What has driven advancements in AI related to unstructured data?
The fact that much of the world’s data is unstructured
What are some applications of AI that use unstructured data?
- Social media monitoring
- Customer feedback analysis
What is tabular data?
A subset of structured data represented in tables with rows and columns
What does each row in tabular data typically represent?
A unique instance
What does each column in tabular data typically represent?
An attribute
Where is tabular data commonly stored?
Databases and spreadsheets
In which applications is tabular data commonly used?
Traditional data science and machine learning applications such as:
* Regression analysis
* Classification
* Recommendation systems
What is time series data?
Sequences of data points collected or recorded at specific time intervals
Give examples of data points in time series data.
Hourly stock prices, daily temperature measurements, monthly sales figures
What is a key characteristic of each data point in time series data?
Each data point is timestamped
What fields widely use time series data?
Finance, healthcare, manufacturing
What is a common application of time series data in finance?
Stock price prediction
What unique challenges does time series data present?
Requires specialized models to account for the sequential nature of the data
What type of neural networks are commonly used with time series data?
Recurrent neural networks (RNNs)
Fill in the blank: Time series data is suitable for models that predict future trends or detect _______.
seasonal patterns
What are the three main categories of ML techniques?
Supervised, unsupervised, and reinforcement learning
What is supervised learning?
A technique where an algorithm is trained on labeled data
What does labeled data mean in supervised learning?
Each input comes with a corresponding output
What role does the ‘supervisor’ play in supervised learning?
Provides labeled data to help the algorithm learn associations
Give an example of supervised learning.
Email spam detection
How does unsupervised learning differ from supervised learning?
Works with unlabeled data
What is the main goal of unsupervised learning?
To find patterns, similarities, or structures in the data
What does an unsupervised learning algorithm do with the dataset?
Explores the dataset and groups the data into clusters based on similarity
In unsupervised learning, what does the algorithm identify instead of mapping inputs to outputs?
Clusters or groupings based on feature similarities
What is semi-supervised learning?
Semi-supervised learning bridges the gap between supervised and unsupervised learning by using the labeled data to guide the learning process on the larger unlabeled dataset.
A common application of semi-supervised learning is in speech recognition systems.
What is reinforcement learning (RL)?
Reinforcement learning (RL) is a type of machine learning where an agent interacts with an environment to maximize cumulative rewards over time.
Unlike supervised and unsupervised learning, RL relies on feedback through rewards or penalties instead of labeled or unlabeled data.
What inspires reinforcement learning?
Reinforcement learning is inspired by behavioral psychology, where learning occurs through rewards and punishments.
This inspiration helps in designing systems that mimic human learning behavior.
What are the key components of reinforcement learning?
In reinforcement learning, an agent takes actions within an environment, observes the outcome, and receives feedback in the form of rewards or penalties.
The goal is for the agent to learn an optimal policy.
What does an agent aim to achieve in reinforcement learning?
The goal is for the agent to learn an optimal policy—a series of actions that maximize the total reward.
What are the two strategies an agent uses in reinforcement learning?
The agent explores different actions (exploration) and exploits known strategies (exploitation) to learn effective behaviors.
What is a recommendation system?
The recommendation system provides dynamic suggestions based on the user’s preferences, especially in e-commerce use cases.
What is reinforcement learning with human feedback (RLHF)?
Reinforcement learning with human feedback (RLHF) is a specialized technique where human feedback is integrated into the reinforcement learning process to guide the agent’s learning.
How does RLHF differ from traditional reinforcement learning?
In traditional reinforcement learning, an agent learns through interaction with an environment, while RLHF introduces human evaluations into this loop, providing more nuanced and qualitative feedback.
What role does human feedback play in RLHF?
Human feedback acts as an additional reward signal, especially when defining a complex reward function or when the task involves subjective judgment.
What is a notable use case for RLHF?
A notable use case for RLHF is training AI chatbots like those used for customer support or personal assistants.
What is supervised learning best suited for?
Tasks with clearly defined goals and labeled data
Provides high accuracy for tasks like classification and regression.
What is the main value of unsupervised learning?
Data exploration when labels are unavailable
Enables insight into data structures and relationships.
How does reinforcement learning function?
An agent learns through interaction with an environment
Learns optimal policies through rewards and penalties.
What scenario is semi-supervised learning ideal for?
When labeled data is scarce but unlabeled data is abundant
Combines the benefits of both supervised and unsupervised learning to improve performance.
What does reinforcement learning with human feedback integrate?
Human evaluations into learning
Aligns AI behavior with human preferences and enhances model performance in complex, subjective tasks.
What is the first step in any ML project?
Data collection
What does data pruning involve?
Removing irrelevant or noisy data points
What is the purpose of imputation in data processing?
Handling missing values
Name two techniques used for imputation.
Mean substitution, predictive modeling
What is the purpose of scaling and normalization?
Adjusting features to a uniform scale for better model convergence
What is data splitting?
Dividing data into training, validation, and test sets
What is the typical ratio for data splitting in ML projects?
70:15:15
Fill in the blank: Data pruning is necessary to ensure the dataset is _______.
representative and manageable
What is feature engineering?
The process of creating new features or modifying existing ones to improve model performance.
This plays a crucial role in enhancing the predictive power of machine learning models.
What is feature selection?
Choosing the most relevant features to simplify the model and reduce overfitting using methods like recursive feature elimination or correlation analysis.
This process helps in improving model performance by eliminating irrelevant or redundant features.
What are polynomial features?
Creating interaction terms (e.g., a product of two features) for nonlinear relationships.
This is useful for capturing more complex patterns in the data.
What does encoding refer to in machine learning?
Converting categorical data into numerical forms using label encoding or one-hot encoding.
This step is essential for enabling machine learning algorithms to process categorical variables.
What is data augmentation?
Expanding the dataset using techniques such as rotation, flipping (for images), or paraphrasing (for text), which introduces variability and helps prevent overfitting.
This technique is particularly useful in training deep learning models.
What is model complexity?
Balancing complexity and interpretability (e.g., choosing between decision trees and neural networks).
This consideration is important for ensuring that the model is both effective and understandable.
What does algorithm suitability refer to?
Selecting models like convolutional neural networks (CNNs) for image data or gradient boosting for tabular data.
Choosing the right algorithm is crucial for achieving optimal performance based on the nature of the data.
What is model training?
The phase where the algorithm learns from the data.
This is a crucial stage in machine learning where the model adapts based on the input data.
What are optimization algorithms used for?
To minimize loss.
Common optimization algorithms include Gradient descent, stochastic gradient descent (SGD), and Adam optimizers.
What is early stopping in machine learning?
Halting training when performance on the validation set stagnates, preventing overfitting.
This technique helps to ensure that the model does not learn noise from the training data.
What are hyperparameters?
Model settings that are not learned from data but are set before training begins.
Hyperparameters influence the learning process and model architecture.
Give examples of hyperparameters.
- Learning rate
- Number of layers in a neural network
- Number of trees in a random forest
These values can significantly affect model performance.
What methods are commonly used to search through hyperparameter combinations?
- Grid search
- Random search
- Bayesian optimization
These methods help find the best hyperparameter settings for a model.
What is the purpose of tuning hyperparameters?
To avoid overfitting.
Tuning is often performed on a validation set to assess the model’s performance.
What is the final step in the machine learning process?
Evaluation.
This step assesses how well the trained model performs on new, unseen data.
Define inference in the context of machine learning.
The process of using the trained machine learning model to make predictions and generate results.
Inference is a key application of the trained model.
What is batch inference?
Takes multiple observation points simultaneously and produces a result.
It does not produce results in real time and usually runs inference weekly or hourly.
What is real-time inference?
Making predictions or decisions based on data in near-instantaneous time.
This process occurs immediately, or within a few milliseconds to seconds, after receiving new data.
What is asynchronous inference?
Asynchronous inference occurs when the inference request is queued and processed when the model is available.
How does asynchronous inference address the limitations of real-time inferencing?
It allows processing of requests without the need for an immediate response.
What happens to asynchronous inference requests?
They are pushed into a queue and processed when the resources are available.
True or False: Asynchronous inference holds resources until the request is processed.
False
hWhat does regression analyze in machine learning?
Regression analyzes how independent variables or features correlate with a dependent variable or outcome.
What type of modeling approach does regression serve as?
Regression serves as a predictive modeling approach.
What type of outcomes does regression predict?
Regression predicts continuous outcomes.
In the context of predicting house prices, what is the independent variable?
The independent variable is the area of the house.
In the context of predicting house prices, what is the dependent variable?
The dependent variable is the price of the house.
How can the relationship between area and house price be modeled?
The relationship can be modeled using a straight line.
What does a classification algorithm do?
A classification algorithm assigns a label or category to a given input based on its attributes.
What is the task of predicting the class of a new observation based on previously observed examples called?
The task is known as classification.
What type of learning is classification an example of?
Classification is an example of supervised learning.
What does supervised learning use for training?
Supervised learning uses labels and annotations for training.
What is clustering?
Clustering organizes the data points into groups.
What type of learning does classification represent?
Classification is an example of supervised learning.
What type of learning does clustering represent?
Clustering is an example of unsupervised learning.
When do we use clustering algorithms?
We use clustering algorithms when we want to divide the data into multiple categories without labeled data or defined labels.
Fill in the blank: Classification is to supervised learning as clustering is to _______.
unsupervised learning.
