Cheat Sheet for
Machine Learning
Machine Learning - answers A subset of AI that enables systems to learn and improve
from experience
Deep Learning - answers A subset of machine learning using neural networks with
multiple layers
Business Objectives - answers Define what problem you're solving
Data Preparation - answers Collect, prepare and label data
Data Augmentation - answers Increase dataset size and variety
Model Development - answers Choose model architecture
Model Training - answers Train the model with prepared data
Model Evaluation - answers Measure performance
Model Deployment - answers Implement in production
Model Monitoring - answers Continuous assessment
Business Objectives - answers Must relate to bottom line and business value
Specific, measurable goals - answers Define specific, measurable goals
100% accuracy necessity - answers Consider if 100% accuracy is necessary or if partial
improvement provides value
Current costs calculation - answers Calculate current costs of not solving the problem
Image Formation - answers Follow proper image acquisition fundamentals such as
camera position, field of view, resolution, lighting, and filters
Good image formation - answers Good image formation reduces required dataset size
Training set - answers Used to build the model
Validation set - answers Used to prevent overtraining
, Testing set - answers Used to evaluate performance
Image Classification - answers Labeling whole images (e.g., 'good part' vs 'bad part')
Object Detection - answers Finding items in images with bounding boxes
Segmentation - answers Identifying regions at pixel level
Instance Segmentation - answers Identifying individual objects at pixel level
Data Augmentation - answers Techniques to increase dataset size when limited data is
available
Augmentation methods - answers Methods include rotation, flipping, cropping, color
adjustments
Platform Selection - answers High-control platforms (F1 steering wheel) provide access
to all hyperparameters
User-friendly platforms - answers User-friendly platforms (normal steering wheel)
provide simplified interfaces
Hyperparameters - answers Values given to the training model that aren't derived from
data
Transfer Learning - answers Using pre-trained models as starting points, repurposing
models trained for one task to perform a related task
Confusion Matrix - answers Table showing predicted vs actual results
True Positive (TP) - answers Correctly identified positives
True Negative (TN) - answers Correctly identified negatives
False Positive (FP) - answers Incorrectly identified positives (Type I error)
False Negative (FN) - answers Incorrectly identified negatives (Type II error)
Precision - answers TP/(TP+FP) - How accurate positive predictions are
Recall - answers TP/(TP+FN) - How complete the results are
F-Score - answers 2*(Precision*Recall)/(Precision+Recall) - Combined metric
Model Deployment - answers Consider computational requirements for inference
Machine Learning
Machine Learning - answers A subset of AI that enables systems to learn and improve
from experience
Deep Learning - answers A subset of machine learning using neural networks with
multiple layers
Business Objectives - answers Define what problem you're solving
Data Preparation - answers Collect, prepare and label data
Data Augmentation - answers Increase dataset size and variety
Model Development - answers Choose model architecture
Model Training - answers Train the model with prepared data
Model Evaluation - answers Measure performance
Model Deployment - answers Implement in production
Model Monitoring - answers Continuous assessment
Business Objectives - answers Must relate to bottom line and business value
Specific, measurable goals - answers Define specific, measurable goals
100% accuracy necessity - answers Consider if 100% accuracy is necessary or if partial
improvement provides value
Current costs calculation - answers Calculate current costs of not solving the problem
Image Formation - answers Follow proper image acquisition fundamentals such as
camera position, field of view, resolution, lighting, and filters
Good image formation - answers Good image formation reduces required dataset size
Training set - answers Used to build the model
Validation set - answers Used to prevent overtraining
, Testing set - answers Used to evaluate performance
Image Classification - answers Labeling whole images (e.g., 'good part' vs 'bad part')
Object Detection - answers Finding items in images with bounding boxes
Segmentation - answers Identifying regions at pixel level
Instance Segmentation - answers Identifying individual objects at pixel level
Data Augmentation - answers Techniques to increase dataset size when limited data is
available
Augmentation methods - answers Methods include rotation, flipping, cropping, color
adjustments
Platform Selection - answers High-control platforms (F1 steering wheel) provide access
to all hyperparameters
User-friendly platforms - answers User-friendly platforms (normal steering wheel)
provide simplified interfaces
Hyperparameters - answers Values given to the training model that aren't derived from
data
Transfer Learning - answers Using pre-trained models as starting points, repurposing
models trained for one task to perform a related task
Confusion Matrix - answers Table showing predicted vs actual results
True Positive (TP) - answers Correctly identified positives
True Negative (TN) - answers Correctly identified negatives
False Positive (FP) - answers Incorrectly identified positives (Type I error)
False Negative (FN) - answers Incorrectly identified negatives (Type II error)
Precision - answers TP/(TP+FP) - How accurate positive predictions are
Recall - answers TP/(TP+FN) - How complete the results are
F-Score - answers 2*(Precision*Recall)/(Precision+Recall) - Combined metric
Model Deployment - answers Consider computational requirements for inference
