AIP-210 Sample Questions Answers

Caffe is a deep learning framework that is designed for speed and modularity. It can process large datasets of images efficiently and supports various types of neural networks. It also has a large and active developer community that contributes to its code base and documentation. Caffe is suitable for image processing tasks such as classification, segmentation, detection, and recognition

The optimal value of K is the one that minimizes the inertia or within-cluster sum of squares, while avoiding too many clusters that may overfit the data. The elbow plot shows a sharp decrease in inertia from K = 1 to K = 2, and then a more gradual decrease from K = 2 to K = 3. After K = 3, the inertia does not change much as K increases. Therefore, the elbow point is at K = 3, which is the optimal value of K for this data. References: How to Run K-Means Clustering in Python, K-means clustering - Wikipedia

Data lineage is the process of tracking the origin, transformation, and usage of data throughout its lifecycle. It helps to ensure data quality, integrity, and provenance. Data lineage also supports the Privacy by Design methodology, which is a framework that aims to embed privacy principles into the design and operation of systems, processes, and products that involve personal data. By understanding, documenting, and displaying data lineage, an ML system can demonstrate how it collects, processes, stores, and deletes personal data in a transparent and accountable manner3 .

Semi-supervised learning is a type of machine learning that uses both labeled and unlabeled data to train a model. Semi-supervised learning can be useful when:

Labeling data is challenging and expensive: Labeling data requires human intervention and domain expertise, which can be costly and time-consuming. Semi-supervised learning can leverage the large amount of unlabeled data that is easier and cheaper to obtain and use it to improve the model’s performance.

There is a large amount of unlabeled data to be used for predictions: Unlabeled data can provide additional information and diversity to the model, which can help it learn more complex patterns and generalize better to new data. Semi-supervised learning can use various techniques, such as self-training, co-training, or generative models, to incorporate unlabeled data into the learning process.

Reinforcement learning is a type of machine learning that involves learning from trial and error based on rewards and penalties. Reinforcement learning can be used to develop models for dynamic pathing, which is the problem of finding an optimal path from one point to another in an uncertain and changing environment. Reinforcement learning can enable the model to adapt to new situations and learn from its own actions and feedback. For example, a self-driving car company can use reinforcement learning to train its model to navigate complex traffic scenarios and avoid collisions .

Weighted class is a technique to rebalance a dataset by assigning different weights to each class, according to their frequency in the dataset. The weights are inversely proportional to the class frequency, meaning that rare classes have higher weights and common classes have lower weights. This helps to reduce the bias towards the majority class and improve the model performance on the minority class. References: 4. Data Validation - Building Machine Learning Pipelines, A guide to React design patterns - LogRocket Blog

Adversarial attacks are malicious attempts to fool or manipulate machine learning models by adding small perturbations to the input data that are imperceptible to humans but can cause significant changes in the model output. In general, models that perform their tasks more accurately are less robust against adversarial attacks, because they tend to have higher confidence in their predictions and are more sensitive to small changes in the input data. References: [Adversarial machine learning - Wikipedia], [Why Are Machine Learning Models Susceptible to Adversarial Attacks? | by Anirudh Jain | Towards Data Science]

The input format is the way that the data is structured, organized, and presented to the model. For example, the input format could be a CSV file, an image file, or a JSON object. The input format can affect how the model interprets and processes the data, and therefore how it makes predictions. When the input format changes, it may require retraining the model to adapt to the new format and ensure its accuracy and reliability. For example, if the weather prediction application switches from using numerical values to categorical values for some features, such as wind direction or cloud cover, it may need to retrain the model to handle these changes .

Preexisting bias is a type of bias that originates from historical or social contexts, such as stereotypes, prejudices, or discriminations. Preexisting bias can affect the data or the algorithm used for machine learning, as well as the outcomes or decisions made by machine learning. Preexisting bias can cause unfair or harmful impacts on certain groups or individuals based on their attributes, such as gender, race, age, or disability3. In this case, the software that uses machine learning generates lower employability scores for women and predicts that women, especially with children, are less likely to get a high-paying job. This indicates that the software has preexisting bias against women, which may reflect the historical or social inequalities or expectations in the labor market.

Lasso regression is a type of linear regression that adds a regularization term to the loss function to reduce overfitting and improve generalization. Lasso regression uses an L1 norm as the regularization term, which is the sum of the absolute values of the coefficients. Lasso regression can shrink some of the coefficients to zero, which effectively eliminates some of the features from the model. Lasso regression is preferable over ridge regression when there are many features with no association with the dependent variable, as it can perform feature selection and reduce the complexity and noise of the model.

A neural network without an activation function is equivalent to a form of a linear regression. A neural network is a computational model that consists of layers of interconnected nodes (neurons) that process inputs and produce outputs. An activation function is a function that determines the output of a neuron based on its input. An activation function can introduce non-linearity into a neural network, which allows it to model complex and non-linear relationships between inputs and outputs. Without an activation function, a neural network becomes a linear combination of inputs and weights, which is essentially a linear regression model.

Hyperparameter optimization is the process of finding the optimal values for hyperparameters that control the learning process of a given algorithm. Hyperparameters are parameters that are not learned by the algorithm but are set by the user before training. Hyperparameters can affect the performance and behavior of the algorithm, such as its speed, accuracy, complexity, or generalization. Hyperparameter optimization can help improve the efficiency and effectiveness of the algorithm by tuning its hyperparameters to achieve the best results.

Hidden attacks are malicious activities that aim to compromise or manipulate an ML system without being detected or noticed. Hidden attacks can target different stages of an ML workflow, such as data collection, model training, model deployment, or model monitoring. Some examples of hidden attacks are data poisoning, backdoor attacks, model stealing, or adversarial examples. Continuous monitoring of bias and variance can help ML engineers to prevent hidden attacks, as it can help them detect any anomalies or deviations in the data or the model’s performance that may indicate a potential attack.

Workflow design patterns for machine learning pipelines are common solutions to recurring problems in building and managing machine learning workflows. One of these patterns is to represent a pipeline with a directed acyclic graph (DAG), which is a graph that consists of nodes and edges, where each node represents a step or task in the pipeline, and each edge represents a dependency or order between the tasks. A DAG has no cycles, meaning there is no way to start at one node and return to it by following the edges. A DAG can help visualize and organize the pipeline, as well as facilitate parallel execution, fault tolerance, and reproducibility.

Normalization is a method that can help us to solve the problem of features being on different scales when using the k-nearest neighbors algorithm. Normalization is a technique that rescales the values of features to a common range, such as [0, 1] or [-1, 1]. Normalization can help reduce the influence or dominance of some features over others, as well as improve the accuracy and performance of the algorithm2.

An L1 norm is a measure of distance or magnitude that is defined as the sum of the absolute values of the components of a vector. For example, if x and y are two components of a vector, then the L1 norm of that vector is |x| + |y|. The L1 norm is also known as the Manhattan distance or the taxicab distance, as it represents the shortest path between two points in a grid-like city.

Generative adversarial networks (GAN) are a type of AI technology that can create fake videos, images, audio, or text that are realistic and indistinguishable from real ones. GAN consist of two neural networks: a generator and a discriminator. The generator tries to produce fake samples from random noise, while the discriminator tries to distinguish between real and fake samples. The two networks compete against each other in a game-like scenario, where the generator tries to fool the discriminator and the discriminator tries to catch the generator. Through this process, both networks improve their abilities until they reach an equilibrium where the generator can produce convincing fakes.

Emergent bias is a type of bias that arises when an AI model encounters new or different data or scenarios that were not present or accounted for during its training or development. Emergent bias can cause the model to make inaccurate or unfair predictions or decisions, as it may not be able to generalize well to new situations or adapt to changing conditions. One possible cause of emergent bias is seasonality, which means that some variables or patterns in the data may vary depending on the time of year. For example, if an AI model for merchandise sales prediction was trained in winter and applied in summer, it may produce biased results due to differences in customer behavior, demand, or preferences.

Word embedding is a task in natural language processing (NLP) where words are converted into numerical vectors that represent their meaning, usage, or context. Word embedding can help reduce the dimensionality and sparsity of text data, as well as enable various operations and comparisons among words based on their vector representations. Some of the common methods for word embedding are:

One-hot encoding: One-hot encoding is a method that assigns a unique binary vector to each word in a vocabulary. The vector has only one element with a value of 1 (the hot bit) and the rest with a value of 0. One-hot encoding can create distinct and orthogonal vectors for each word, but it does not capture any semantic or syntactic information about words.

Word2vec: Word2vec is a method that learns a dense and continuous vector representation for each word based on its context in a large corpus of text. Word2vec can capture the semantic and syntactic similarity and relationships among words, such as synonyms, antonyms, analogies, or associations.

GloVe: GloVe (Global Vectors for Word Representation) is a method that combines the advantages of count-based methods (such as TF-IDF) and predictive methods (such as Word2vec) to create word vectors. GloVe can leverage both global and local information from a large corpus of text to capture the co-occurrence patterns and probabilities of words.

 Transparency is an ethical principle that describes the degree to which an AI system can provide clear and understandable information about its inputs, outputs, processes, and decisions. Transparency can help increase trust and confidence among users and stakeholders, as well as enable accountability and responsibility for the system’s actions and outcomes. Logging and storing historical predictions for potential, future access requests is an example of transparency, as it can help provide evidence and explanation for the system’s recommendations, as well as facilitate auditing and feedback.

Skip-gram and TF-IDF are both text vectorization methods that convert text into numerical feature vectors. Skip-gram is a prediction-based word embedding method that learns vector representations of words from their contexts in a large corpus of text. TF-IDF is a frequency-based word weighting method that assigns scores to words based on their importance in a document and in a corpus of documents. References: Text Vectorization and Word Embedding | Guide to Master NLP (Part 5), What Is Text Vectorization? Everything You Need to Know - deepset

Personal data is any information that relates to an identified or identifiable individual, such as name, address, email, phone number, or biometric data. Personal data should not be disclosed, made available, or otherwise used for purposes other than specified, except with:

The consent of the person it is collected from: Consent is a clear and voluntary indication of agreement by the person to the processing of their personal data for a specific purpose. Consent can be given by a statement or a clear affirmative action, such as ticking a box or clicking a button.

The authority of law: The authority of law is a legal basis or obligation that requires or permits the processing of personal data for a legitimate purpose. For example, the authority of law could be a court order, a subpoena, a warrant, or a statute.

Data privacy is the right of individuals to control how their personal data is collected, used, shared, and protected. It also involves complying with relevant laws and regulations that govern the handling of personal data. Data privacy is especially important when extracting business intelligence from primary data captured from the public, as it may contain sensitive or confidential information that could harm the individuals if misused or breached .