AIP-210 Questions and Answers

The transform-design pattern for ML pipelines aims to separate inputs from features, encapsulate the processing steps of ML pipelines, and represent steps in the pipeline with a DAG. These goals help to make the pipeline modular, reusable, and easy to understand. The transform-design pattern does not seek to isolate individual steps of ML pipelines, as this would create entanglement and dependency issues. It also does not transform the output data after production, as this would violate the principle of separation of concerns. 

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

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.

The Adversarial ML Threat Matrix is an open framework designed to help detect, respond to, and remediate threats in ML systems. The Adversarial ML Threat Matrix is inspired by the MITRE ATTandCK® Matrix1, which is a framework for describing cyberattacks across various stages of an attack lifecycle. The Adversarial ML Threat Matrix adapts this framework to address specific threats and vulnerabilities in ML systems, such as data poisoning, model stealing, model evasion, or model inversion2. The Adversarial ML Threat Matrix provides a structured way to organize and classify adversarial techniques, tactics, procedures, examples, and mitigations for ML systems2.

Underfitting is a problem that occurs when a model learns too little from the training data and fails to capture the underlying complexity or structure of the data. Underfitting can result from using insufficient or irrelevant features, a low complexity of the model, or a lack of training data. Underfitting can reduce the accuracy and generalization of the model, as it may produce oversimplified or inaccurate predictions. Some of the ways to fix underfitting are:

Add features to training data: Adding more features or variables to the training data can help increase the information and diversity of the data, which can help the model learn more complex patterns and relationships.

Increase the complexity of the model: Increasing the complexity of the model can help increase its expressive power and flexibility, which can help it fit better to the data. For example, adding more layers or nodes to a neural network can increase its complexity.

Train the model for more epochs: Training the model for more epochs can help increase its learning ability and convergence, which can help it optimize its parameters and reduce its error.

Getting more training data will not work to fix underfitting, as it will not change the complexity or structure of the data or the model. Getting more training data may help with overfitting, which is when a model learns too much from the training data and fails to generalize well to new or unseen data.

Ridge regression is a type of regularization technique that can help reduce collinearity among independent variables. It does this by adding a penalty term to the ordinary least squares (OLS) objective function, which shrinks the coefficients of highly correlated variables towards zero. This reduces the variance of the coefficient estimates and improves the stability and accuracy of the regression model. References: Multicollinearity in Regression Analysis: Problems, Detection, and Solutions - Statistics By Jim, A Beginner’s Guide to Collinearity: What it is and How it affects our regression model - StrataScratch

Semi-supervised learning is an approach that is best if a limited portion of your training data is labeled. Semi-supervised learning is a type of machine learning that uses both labeled and unlabeled data to train a model. 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. Semi-supervised learning can use various techniques, such as self-training, co-training, or generative models, to incorporate unlabeled data into the learning process.

Data skews happen in the ML pipeline when the distribution or characteristics of the live input data differ from those of the offline data used for training and testing the model. This can lead to a degradation of the model performance and accuracy, as the model is not able to generalize well to new data. Data skews can be caused by various factors, such as changes in user behavior, data collection methods, data quality issues, or external events. References: What is training-serving skew in Machine Learning?, Data preprocessing for ML: options and recommendations

Technical debt is a metaphor that describes the implied cost of additional work or rework caused by choosing an easy or quick solution over a better but more complex solution. Technical debt can accumulate in ML systems due to various factors, such as changing requirements, outdated code, poor documentation, or lack of testing. Some of the ways to decrease technical debt in ML systems are:

Documentation readability: Documentation readability refers to how easy it is to understand and use the documentation of an ML system. Documentation readability can help reduce technical debt by providing clear and consistent information about the system’s design, functionality, performance, and maintenance. Documentation readability can also facilitate communication and collaboration among different stakeholders, such as developers, testers, users, and managers.

Refactoring: Refactoring is the process of improving the structure and quality of code without changing its functionality. Refactoring can help reduce technical debt by eliminating code smells, such as duplication, complexity, or inconsistency. Refactoring can also enhance the readability, maintainability, and extensibility of code.

An activation function is a function that determines the output of a neuron in a neural network 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. Some of the common activation functions are:

Sigmoid: A sigmoid function is a function that maps any real value to a value between 0 and 1. It has an S-shaped curve and is often used for binary classification or probability estimation.

Hyperbolic tangent: A hyperbolic tangent function is a function that maps any real value to a value between -1 and 1. It has a similar shape to the sigmoid function but is symmetric around the origin. It is often used for regression or classification problems.

ReLU: A ReLU (rectified linear unit) function is a function that maps any negative value to 0 and any positive value to itself. It has a piecewise linear shape and is often used for hidden layers in deep neural networks.

Additive is not an activation function, but rather a term that describes a property of some functions. Additive functions are functions that satisfy the condition f(x+y) = f(x) + f(y) for any x and y. Additive functions are linear functions, which means they have a constant slope and do not introduce non-linearity.

Principal component analysis (PCA) is a technique that reduces the dimensionality of a dataset by transforming it into a set of new variables called principal components. The principal components are linear combinations of the original variables that capture the maximum amount of variance in the data. The first principal component explains the most variance, the second principal component explains the second most variance, and so on. The goal of PCA is to retain as much variance as possible while reducing the number of variables.

 Independence is an assumption of linear regression that states that the errors (residuals) of the model are independent of each other, meaning that they are not correlated or influenced by previous or subsequent errors. Independence can be violated when the data has serial correlation or autocorrelation, which means that the value of a variable at a given time depends on its previous or future values. This can happen when the data is collected over time (time series) or over space (spatial data). In this case, the data is collected over time from a group of subjects, which may introduce serial correlation among the errors.

Anomaly detection is an unsupervised learning technique that identifies outliers or abnormal patterns in data, which can be useful for fraud detection. Anomaly detection algorithms can learn the normal behavior of transactions and flag the ones that deviate significantly from the norm, indicating possible fraud.

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.

To calculate the minimum grade needed to achieve an 80% average, we can use the following formula:

minimum grade = (target average * number of tests - sum of grades) / (number of tests - 1)

Plugging in the given values, we get:

minimum grade = (80 * 7 - (76 + 81 + 78 + 87 + 75 + 72)) / (7 - 6)

minimum grade = (560 - 469) / 1

minimum grade = 91

Therefore, the student needs to score at least 91 on the last test to get an 80% average.

Highly interpretable models are models that can provide clear and intuitive explanations for their predictions, such as decision trees, linear regression, or logistic regression. Some of the statements that are true regarding highly interpretable models are:

They are usually easier to explain to business stakeholders: Highly interpretable models can help communicate the logic and reasoning behind their predictions, which can increase trust and confidence among business stakeholders. For example, a decision tree can show how each feature contributes to a decision outcome, or a linear regression can show how each coefficient affects the dependent variable.

They usually compromise on model accuracy for the sake of interpretability: Highly interpretable models may not be able to capture complex or non-linear patterns in the data, which can reduce their accuracy and generalization. For example, a decision tree may overfit or underfit the data if it is too deep or too shallow, or a linear regression may not be able to model curved relationships between variables.

Increasing the speed of analysis is a benefit of machine learning for solving business problems. Machine learning is a branch of artificial intelligence that involves creating systems that can learn from data and make predictions or decisions. Machine learning can help increase the speed of analysis by automating and optimizing various tasks, such as data processing, feature extraction, model training, model evaluation, or model deployment. Machine learning can also help handle large and complex data sets that may be difficult or impractical to analyze manually or with traditional methods.

Distributed artificial intelligence (DAI) is a subfield of artificial intelligence that studies how multiple intelligent agents can coordinate and cooperate to achieve a common goal or solve a complex problem. DAI relies on a distributed system that performs robust computations across a network of unreliable nodes, such as sensors, robots, or humans. DAI can handle large-scale, dynamic, and uncertain environments that are beyond the capabilities of a single agent. References: [Distributed artificial intelligence - Wikipedia], [Distributed Artificial Intelligence: An Overview]

k-means clustering is an algorithm that should be used by the market research team to obtain groups of patients with similar unmet needs. k-means clustering is an unsupervised learning technique that partitions the data into k clusters based on the similarity of the features. The algorithm iteratively assigns each data point to the cluster with the nearest centroid and updates the centroid until convergence. k-means clustering can help identify patterns and segments in the data that may not be obvious or intuitive. References: [K-means clustering - Wikipedia], [How to Run K-Means Clustering in Python]

IBM Watson is an AI technology that can take a question in natural language and return a precise answer to the question. IBM Watson is a cognitive computing system that can understand natural language, generate hypotheses, and provide evidence-based answers. IBM Watson can be applied to various domains and industries, such as healthcare, education, finance, or law.

Before log-transforming Y, we should add 1 to all of the Y values. This is because log transformation is undefined for zero or negative values, and some of the Y values may be zero. Adding 1 to all of the Y values can avoid this problem and ensure that the log transformation is valid and meaningful. Adding 1 to all of the Y values is also known as a log-plus-one transformation.

Harms modeling is a technique that helps product managers design AI solutions responsibly by evaluating both positive and negative outcomes. Harms modeling involves creating a shared taxonomy of potential negative impacts and conducting an assessment along vectors such as severity, impact, frequency, and likelihood. Harms modeling can help identify and mitigate any risks or harms that may arise from using AI solutions. References: [Harms Modeling for Responsible AI | by Google Developers | Google Developers], [Harms Modeling for Responsible AI - YouTube]