HCI3 Exam 3 - Intro to AI and Machine Learning

patterns

AI finds ______ in structured and unstructured data and uses them to make predictions

structured

data in tables with rows and columns is an example of _______ (structured/unstructured) data

unstructured

data in formats like images, videos, free text, and sound are examples of ______ (structured/unstructured) data

true

T/F: brain waves and molecules are a types of unstructured data

statistics

focus on developing new estimators with key theoretical properties such as unbiasedness, consistency, efficiency, asymptomatic normality, and robustness

AI/ML

focus on predictive accuracy and computational speed with few theoretical guarantees such as evaluating predictive accuracy on out-of-sample data and comparison to existing predictive methods

rule-based AI

hard-coded logic that works with small datasets or no data at all and required heavy human involvement in design

ML AI

handles moderate amount of data to learn patterns and requires medium human involvement

deep learning

handles very large datasets for complex pattern recognition and deep neural networks

linear regression

a statistical method used to fit a linear model to a given data set

true

T/F: linear regression can be a relatively poor predictor of outcomes for complicated data patterns

decision trees

used to divide data into smaller groups to fit simple models for each group

interpretability

builds trust with patients, offers transparency, ensures accountability through informed decision making, and regulatory requirements in healthcare

boosted trees

uses multiple trees instead of one, building each new tree sequentially following the last tree with improvements; very powerful but difficult to compute and fine tune

Xgboost

powerful and popular algorithm built on boosted trees for structured data

neural networks

allow analyzation of unstructured data using multi-stage regression or classification model

explainability

a single tree model has more ______ than a boosted tree model

accuracy

a boosted tree model has more ______ than a single tree model

activation function

a neural network uses an ______ to discard useless results

image

neural networks are the most powerful ML tool for ______ classification

pixel

neural networks function by assigning a weight to each ______

performance monitoring

a process used to ensure model generalizability to fit both existing and new data well

70/15/15

what is a common split for data into training, validation, and test data to ensure performance monitoring

predefined iterations, cross-validation, and regularization

three examples of stopping rules to complete model training

K-fold cross validation

data is split into k mutual subsets and k number training/testing experiments are conducted

bias-variance trade-off

balancing model complexity and prediction accuracy

bias

model is too simple and misses important patterns - underfitting

variance

model is too complicated and captures noise from the training data making a bad predictor for new data - overfitting

define, collect, prepare, discover, train, evaluate, select, and launch/monitor/maintain

define the process for ML

Microsoft Azure, SAS Enterprise Miner (Viya), and Python Sci-Kit Learn

examples of structured data platforms

Amazon Rekognition, IBM Visual Recognition, and Google Vision AI

image apps for unstructured data platforms

IBM Watson Tone Analyzer and Amazon Comprehend

text apps for unstructured data platforms

false

T/F: it is advisable to incorporate multiple coding languages into your coding when analyzing unstructured data platformrs

reduce cost, manual labor, and subjectivity

use cases for AI/ML

descriptive (what happened?) and diagnostic (why did it happen?) analysis

when should ML NOT be used?

data quality, labeling, and model accuracy/confidence

why do most ML projects fail?