Objective 3.1 Collect and prepare a dataset to use for training and testing.

• Sources of data
  • Big data/data sets
    • Volume, variety, velocity
    • Data repositories
    • Data prep
    • Stream data
• Structure of data
• Extract, transform, and load (ETL)

Objective 3.2 Analyze a dataset to gain insights.

• Visualization
• Correlations between attributes
• Attribute characteristics
• Descriptive statistics applied to features
  • Standard deviation
  • Kurtosis
  • Mean, mode, median
  • Variance
  • Column distribution
• Classic data sets
• • Time Series
  • Bayesian

Objective 3.3 Clean data in preparation for use in machine learning.

• Data backups
• Data cleansing
• Typecasting
• Operations appropriate for different data types
• Data encryption

Objective 3.4 Engineer features of a dataset to prepare it for use in a machine learning model.

• Feature transformation
• Dimensionality reduction
• Center and spread measures
• Imputing missing values
• Duplicates
• Data binning
• Normalization and standardization
• String manipulation
• Summarization
• Embedded spaces

Objective 4.1 Select and implement an appropriate algorithm to solve a given business problem.

• Decision Tree
  • Hyperparameters
    • Maximum depth
    • Minimum sample per leaf node
  • Number of random splits
  • Information gain/entropy measure
  • GINI Index
  • Continuous variable discretization
  • Random Forest
    • Hyperparameters
      • Number of trees
      • Maximum depth
      • Minimum samples required to split node
      • Minimum samples required to be at leaf node
    • Variable selection
    • Assumption
    • Sampling is representative
• SVM
  • Kernels
    • RBF
    • Gaussian
    • Polynomial
    • Linear kernel
    • Sigmoid
    • TanH
  • Can tackle non-linearly separable data
  • Uses hyperplanes for separation
• Neural networks
  • ANN
    • Layers of a traditional network Input
    • Hidden
    • Input
    • Output
  • Activation functions
    • Tanh
    • Sigmoid
    • ReLU
  • CNN
    • Image processing
  • RNN
    • Natural language
    • LSTM
      • Natural language
      • Forecasting
  • GAN
    • Adversarial content creation
• Clustering
  • K-means clustering
    • Issues clustering circular data
    • Methods for deciding K-means
      • Elbow
      • Silhouette plot
• Classification
  • Logistic regression
  • Softmax regression
• Regression
  • Linear regression
• KNN
  • How do you decide the K?
  • Difference between K-Means and KNN

Objective 4.2 Select and implement the appropriate techniques for a given machine learning problem.

• Dimensionality reduction
  • Methods for DR
    • PCA
    • Lasso
    • Random Forest
• Feature engineering
  • When feature engineering is useful
  • Benefits of deriving custom features
• Feature expansion
  • Benefits to tackling nonlinear data
  • Methods of feature expansion
• Hyper-parameter optimization
  • Awareness of core parameters for key models
• Model tuning
• Types of hyperparameter searches
  • Random
  • Grid
  • Bayesian
  • Genetic
• Cross validation
  • Leave 1 out cross validation
    • Used for small data sets
  • N Fold cross validation
• Regularization
  • Types
    • L1
    • L2
  • Know the differences between L1 and L2
• Variance vs. bias
  • The variance bias tradeoff
• Model generalization
  • Overfitting
  • Use regularization
  • Relationship with variance and bias
• Embedded spaces
  • Embedded space extraction from CNN
  • Word vector open source tools
    • Fasttext
    • word2vec
    • doc2vec
• Data sets
  • Structured
  • Unstructured

Objective 4.3 Manage the time needed to train a model.

• Estimating the time needed to run a batch over a certain number of epochs
• Optimizing the development environment (e.g., scaling up GPU) vs. making compromises in the model to reduce processing time
• Awareness around processing costs

Objective 4.4 Train and tune a machine learning model.

• Model performance evaluation
  • Confusion matrix
  • Classifier performance measurement
  • Accuracy o Precision
  • Recall
  • Precision and recall tradeoff
  • F1 Score
  • ROC Curve
  • Thresholds
  • AUC
  • PRC
• Cross validation
• Model generalization
• Performance tuning

Objective 5.1 Communicate the findings of a machine learning project back to the organization.

• Translating ML results into potential business actions
  • Prediction or classification problems
• Data visualization
• Big data
• Internet of things
• Categorical
• Quantitative
• Types
  • Table
  • Graph
• Explain accuracy vs precision vs recall to non-ML practitioners
  • True positives
  • False positives
• Compromising on accuracy for model interpretability

Objective 6.1 Identify and describe the impact that propagating biases has within AI.

• Avoiding biases in data
  • Current data sources
  • Model explainability
  • Transparency
• Preventing propagation of preconceived notions
  • Open review of data sets and algorithmic approaches
• Recognizing that proxies may be indicative of larger social discriminations

Objective 6.2 Comply with laws and standards that are applicable to businesses that employ AI.

• Relevant data privacy laws
  • GDPR
  • California Consumer Privacy Act

Objective 6.3 Given a use case, identify and describe the ethical issues on sharing the data sources.

• The importance of Open Source Data Access and data integrity
• Respecting the privacy of data sources and targets
• Guidelines for protecting privacy when collecting, storing, and disposing of data

Objective 6.4 Comply with company policies to promote privacy, security, and ethical practices.

• Potential ethical issues resulting from AI practices
  • Recognizing that the output of AI even though derived can still infringe on privacy/copyright/intellectual property rights
  • Recognizing the importance of basic humanitarian principles before developing any AI capabilities
  • Ability to identify alternative uses of AI you create by bad actors