CAIC 9.4 Flashcards
What are some common marketing campaigns and advertising tactics used by retailers?
Direct marketing emails and digital advertisements
These tactics aim to attract customers with incentives or discounts based on demographics.
What is the primary goal of marketing campaigns in retail?
To achieve a high conversion rate while minimizing advertising costs and reducing customer disturbances.
How do ML models optimize marketing campaigns?
By using customer data and demographic factors to identify potential customers and determine appropriate messaging and incentives.
What is the purpose of customer segmentation in marketing?
To understand different customer segments and improve the effectiveness of marketing campaigns.
What is highly personalized marketing?
Marketing that creates individual profiles using behavior data to generate customized campaigns.
Fill in the blank: ML approaches to user-centric targeted marketing predict the conversion rate, known as _______.
[conversion probability]
What is contextual advertising?
A targeted marketing technique that displays ads relevant to the content on a web page.
How does ML assist in contextual advertising?
By identifying the context of an ad to ensure appropriate placement.
What is generative AI’s role in targeted marketing?
To create dynamically personalized content tailored to individual customer preferences.
Why is understanding consumer perception crucial for retail businesses?
It significantly impacts their success and helps in monitoring brand reputation.
What techniques do retailers use to assess customer sentiment?
Soliciting feedback and monitoring social media channels.
What is sentiment analysis?
A text classification problem that determines whether sentiment is positive, negative, or neutral.
What are common algorithms used for sentiment analysis?
ML algorithms, including deep learning-based algorithms.
How do retailers use sentiment analysis?
To gain insights into customer preferences and identify areas for improvement.
What do retailers rely on for inventory planning and demand forecasting?
To manage inventory costs while maximizing revenue and avoiding out-of-stock situations.
What limitations do traditional demand forecasting methods have?
Limitations in accuracy and reliability.
Which techniques are retailers turning to for improved demand forecasting?
Statistical and ML techniques such as regression analysis and deep learning.
Fill in the blank: Deep learning-based algorithms can produce accurate demand forecasts by incorporating multiple _______.
[data sources]
What is the system architecture of an autonomous vehicle composed of?
Perception and localization, decision and planning, control.
What role does perception play in autonomous driving?
It gathers information about surroundings and determines the vehicle’s position.
What sensors are used in the perception stage of autonomous vehicles?
RADAR, LIDAR, cameras, and ultrasonic systems.
What is the function of the decision and planning stage in autonomous vehicles?
To control the vehicle’s motion and behavior based on perception data.
What is the role of AI/ML in the decision and planning stage?
To analyze data and determine the optimal path for the vehicle.
What is the purpose of the control module in autonomous driving?
To translate decisions into physical actions that control the vehicle.
What techniques can be applied in the control module of autonomous vehicles?
Adaptive control systems and reinforcement learning techniques.
What does ADAS stand for?
Advanced Driver Assistance Systems.
What features do ADAS technologies provide?
Detecting potential hazards, issuing warnings, and taking corrective actions.
What is the objective function in ML optimization?
A business metric aimed at minimizing or maximizing discrepancies between projected and actual values.
What is the purpose of optimizers in ML?
To find optimal model parameters for minimizing the objective function.
What is gradient descent?
An iterative approach for optimizing neural networks and ML algorithms.
Fill in the blank: The learning rate is a hyperparameter that controls the magnitude of _______ in ML optimization.
[parameter updates]
What is gradient descent?
An iterative approach for optimizing neural networks and ML algorithms by calculating the rate of error change associated with input variables.
Gradient descent updates model parameters step by step to reduce error.
What is the learning rate in gradient descent?
A hyperparameter that controls the magnitude of parameter updates at each iteration.
It allows fine-tuning of the optimization process.
What are the key steps in the gradient descent optimization process?
- Initialize W randomly
- Calculate error using W
- Compute the gradient of the error
- Update W based on the gradient
- Repeat until gradient is zero
This indicates that the optimal value of W has been reached.
What is the normal equation in ML?
An alternative optimization technique that provides a one-step analytical solution for calculating coefficients in linear regression models.
Unlike gradient descent, it does not require iterative updates.
What are the two primary types of ML tasks?
- Classification
- Regression
Classification involves categorizing data, while regression involves predicting continuous values.
What is overfitting in machine learning?
When a trained model learns the training data too well but fails to generalize to new, unseen data.
Simpler algorithms with fewer parameters may help prevent overfitting.
What factors should be considered when selecting a ML algorithm?
- Problem type
- Dataset size
- Number and nature of features
- Computational requirements
- Interpretability of results
- Assumptions about data distribution
These factors aid in making informed decisions for algorithm selection.
What does linear regression aim to estimate?
The output value by calculating the weighted sum of input variables assuming a linear relationship.
This is expressed through a linear function of coefficients and input variables.
What is the goal of logistic regression?
To estimate the probability of an event occurring, effectively separating classes of data points.
It uses a logistic function to map input variables to a probability score.
What is a decision tree?
A hierarchical model that splits data based on features to classify or predict outcomes.
It uses algorithms like the Gini index and information gain for splitting.
What is a key advantage of decision trees?
Their ability to capture non-linear relationships and interactions between features.
Decision trees can handle both numerical and categorical features.
What is a limitation of decision trees?
They can be prone to overfitting, especially with a large number of features and noisy data.
Overfitting occurs when the model memorizes training data but performs poorly on unseen data.
How does a random forest improve upon decision trees?
By combining the decisions of multiple trees to enhance overall performance.
It utilizes majority voting for classification or averaging for regression.
What is gradient boosting?
A sequential algorithm that aggregates results from different trees, where each tree corrects the errors of the previous one.
It differs from random forests, which use parallel independent trees.
What is a key advantage of gradient boosting?
It excels in handling imbalanced datasets and can achieve higher performance with proper tuning.
It allows for custom loss functions, enhancing flexibility in modeling.
What is a limitation of gradient boosting?
It lacks parallelization capabilities, making it slower in training compared to algorithms that can be parallelized.
This sequential nature can hinder efficiency.
What is the main advantage of gradient boosting?
It has the potential to achieve higher performance than other algorithms when properly tuned.
What custom feature does gradient boosting support?
Custom loss functions.
What is a limitation of gradient boosting related to data?
It is sensitive to noisy data, including outliers.
What is XGBoost?
A widely-used implementation of gradient boosting.
How does XGBoost improve training times?
It enables training a single tree across multiple cores and CPUs.
What techniques does XGBoost use to mitigate overfitting?
Powerful regularization techniques.
What are some other popular variations of gradient boosting trees?
- LightGBM
- CatBoost
What does K-NN stand for?
K-Nearest Neighbors.
What is the underlying assumption of K-NN?
Similar items tend to have close proximity to each other in the feature space.
How does K-NN classify a new data point?
By majority voting among the K nearest neighbors.
What is a key advantage of K-NN?
Its simplicity and lack of need for training or tuning with hyperparameters.
What challenge does K-NN face as the number of data points increases?
Predictions can become slower.
What is a limitation of K-NN regarding dimensionality?
It is not suitable for high-dimensional datasets.
What does an artificial neuron do?
Processes inputs from another neuron, transforms them, and sends output.
What does the activation function in an artificial neuron do?
Modifies the output of the linear function.
What is a Multi-Layer Perceptron (MLP)?
A neural network that stacks multiple layers of neurons.
What is the purpose of backpropagation in neural networks?
To adjust the weights of each neuron based on the contribution to the error.
What are the types of data MLP can handle?
- Tabular data
- Images
- Text
What is clustering in data mining?
Grouping items together based on shared attributes.
What is the K-means algorithm used for?
To group similar data points together in clusters.
What is a drawback of K-means related to initial conditions?
It is sensitive to the initial placement of centroids.
What is a time series?
A sequence of data points recorded at successive time intervals.
What does trend refer to in time series analysis?
The long-term direction of the data.
What does seasonality capture in time series data?
Repeating patterns within a fixed interval.
What is stationarity in time series?
Statistical properties like mean and variance remain constant over time.
What is ARIMA used for?
Analyzing and predicting time series data.
What are the three components of ARIMA?
- Autoregressive
- Moving average
- Differencing
What is DeepAR?
A state-of-the-art forecasting algorithm based on neural networks.
What is a significant drawback of DeepAR?
Its black-box nature lacks interpretability.
What is a recommender system?
An essential machine learning tool used for personalized recommendations.
What is a significant drawback of DeepAR?
The black-box nature of the deep learning model, which lacks interpretability and transparency.
What is a major disadvantage of DeepAR regarding data?
DeepAR performs poorly when the dataset is small.
What is the primary function of a recommender system?
To predict a user’s preference for items based on user or item attribute similarities or user-item interactions.
In what industries has the recommender system gained widespread adoption?
- Retail
- Media and entertainment
- Finance
- Healthcare
What does collaborative filtering rely on to make recommendations?
The preferences and behaviors of similar users.
What is one of the major benefits of collaborative filtering?
It can provide highly personalized recommendations matched to each user’s unique interests.
What is the cold-start problem in collaborative filtering?
Collaborative models struggle when new users or items with no ratings are introduced.
What is matrix factorization in the context of collaborative filtering?
A technique that involves learning vector representations for both users and items in the user-item interaction matrix.
What is the main goal of matrix factorization?
To approximate the original user-item interaction matrix by predicting missing entries.
What does MAB stand for in recommendation systems?
Multi-Armed Bandit.
What is the basic principle behind MAB-based recommendation systems?
To dynamically explore and exploit different recommendations to optimize user experience.
What do MAB algorithms struggle with?
Striking the right balance between exploration and exploitation.
What does computer vision refer to?
The ability of computers to interpret and understand visual representations, such as images and videos.
What are some tasks performed by computer vision?
- Object identification
- Image classification
- Text detection
- Face recognition
- Activity detection
What architecture is specifically designed for processing image data?
Convolutional Neural Network (CNN).
What is a key role of the convolutional layer in a CNN?
Feature extraction from input images.
What is the function of the pooling layer in a CNN?
To reduce the dimensionality of the extracted features.
What are the two commonly used pooling techniques in CNNs?
- Max pooling
- Average pooling
What is the vanishing gradient problem in CNNs?
Signals from initial inputs diminish as they traverse through multiple layers.
What does ResNet use to address the vanishing gradient problem?
Skip connections that allow signals to bypass certain layers.
What is the focus of Natural Language Processing (NLP)?
The relationship between computers and human language.
What are some common tasks within NLP?
- Document classification
- Topic modeling
- Speech-to-text conversion
- Language translation
- Reading comprehension
What does the Bag-of-Words (BOW) model do?
Counts the number of times a word appears in a text.
What are the two components of TF-IDF?
- TF (Term Frequency)
- IDF (Inverse Document Frequency)
What is embedding in the context of NLP?
A technique used to generate low-dimensional representations for words or sentences that capture semantic meaning.
How does embedding differ from BOW and TF-IDF?
Embedding captures the semantic meaning of words, while BOW and TF-IDF create large and sparse input vectors.
What is embedding?
A technique used to generate low-dimensional representations (mathematical vectors) for words or sentences that capture the semantic meaning of the text.
What does the proximity of embeddings in multi-dimensional space indicate?
Words or sentences with similar semantic meanings are closer to each other than those with different meanings.
What is cosine similarity?
A metric that measures how similar two vectors are by calculating the cosine of the angle between them.
Why are embeddings widely adopted in NLP tasks?
They offer more meaningful representations of the underlying text compared to other techniques like simple word counts.
Who created Word2Vec and in what year?
Thomas Mikolov created Word2Vec in 2013.
What are the two techniques used for learning embeddings in Word2Vec?
- CBOW (Continuous Bag of Words)
- Continuous-Skip-Gram
Describe the CBOW technique.
It tries to predict a word for a given window of surrounding words.
Describe the Continuous-Skip-Gram technique.
It tries to predict surrounding words for a given word.
What is the purpose of a sliding window in the CBOW technique?
To run across running text and choose one of the words as the target while the rest serve as inputs.
What is the structure of the MLP network used for training Word2Vec embeddings?
A straightforward one-hidden-layer MLP network.
What do the weights of the hidden layer represent after training Word2Vec?
The actual embeddings for the words.
What is a key advantage of using embeddings as features for downstream tasks?
They can be readily used for tasks like text classification or entity extraction.
What is the limitation of Word2Vec in terms of word meanings?
It produces a fixed embedding representation for each word, disregarding contextual variations in meaning.
What does BERT stand for?
Bidirectional Encoder Representations from Transformers.
What are the two main tasks BERT performs?
- Predicting randomly masked words in sentences
- Predicting the next sentence from a given sentence
How does BERT improve upon Word2Vec?
It generates context-aware embeddings that consider surrounding words.
What additional capability does BERT have regarding embeddings?
It generates embeddings at subword levels.
What is the term for learning embeddings using input tokens in BERT?
Token embedding.
What architectural component does BERT primarily use?
A transformer.
What are the two components of each encoder in a transformer?
- A self-attention layer
- A feed-forward network layer
What does the self-attention layer in a transformer do?
It calculates the strength of the connection between one token and all other tokens in the input sentence.
What NLP tasks can BERT be used for?
- Question answering
- Named entity extraction
- Text summarization
What is the significance of BERT achieving state-of-the-art performance?
It indicates its effectiveness in various NLP tasks when it was released.
What is fine-tuning in the context of using a pre-trained BERT model?
A technique used to adapt the pre-trained model for specific tasks.
