Intelligent System Models

• The text outlines system models with increasing intelligence, each with an environment, sensors, and effectors
• The models are differentiated by components and functions

Model Components

• Environment: The external context for system operation.
• Sensors: Devices gathering data from the environment.
• Effectors: Components acting upon the environment based on system decisions

Intelligent System

• Each model represents an intelligent system with varying capabilities for action selection and understanding context
• The systems evolve from reactive mechanisms to complex decision-making entities

Basic Intelligent System

• Components include sensors that collect environmental information like temperature and light
• Effectors act on the environment based on sensor data, such as turning on a heater
• The system responds directly to sensor input with effector output: a simple feedback loop without deeper context understanding

Context-Aware Intelligent System

• Features a "Context" component understanding the broader situation
• A utility function determines the best action based on sensor input and context
• The system evaluates context (time, occupancy) for more informed decisions, adding intelligence by considering multiple factors

Rule-Based Intelligent System

• Introduces "Rules" defining explicit conditions for system response to sensor input and context
• Rules create a structured action framework, following predefined logic
• Relies on human-defined rules for handling complex scenarios

Advanced Intelligent System (Learning and Adapting)

• Includes components like "Critic" evaluating actions and providing feedback
• Features "Learning" for adaptation over time by recognizing patterns and improving decisions
• "Generator" proposes new potential actions based on learned information
• The system learns from experience, refining behavior, anticipating needs, and optimizing responses

Progression Across Models

• First model is reactive, while the last is adaptive and proactive
• Context awareness starts in the second model
• The final model introduces feedback and learning, making system dynamic and evolving

Analogue to Digital Conversion

• Analog sensors require digitization for use by IoT devices
• IoT devices use binary ('0s and '1s) and ADCs (Analog-to-Digital Converters) to convert analog signals to digital signals.

ADC Bit Resolution

• The resolution of an ADC is determined by bits
• An 8-bit ADC represents discrete values
• With a 5V reference voltage, 256 levels correspond to 5V, 128 to 2.5V, and 0 to 0V

ADC Voltage Resolution

• The smallest change in output is known as the Least Significant Bit (LSB)

Quantisation Error

• Quantisation Error refers to the noise introduced during the quantization process in an Analog-to-Digital Converter (ADC)

Battery Capacity

• 9V: 500mAh
• AAA: 1000mAh
• AA: 2000mAh
• D: 12000mAh
• AA battery stores 2000mAh (or 2Ah) of charge

Battery Configurations

• Series: Batteries in series double the voltage while maintaining the same capacity
• Parallel: Batteries in parallel double the capacity while maintaining the same voltage

Flutter Widget Trees

• Flutter's widget tree is a hierarchical structure defining the layout and behavior of UI
• The Scaffold widget is a key component providing a material design layout framework

Key Components (Flutter)

• AppBar: Material design app bar for title, actions
• Body: Main content of the screen, typically a Center widget with a Column of widgets
• FloatingActionButton: Circular button triggering actions

BLE Weight Scale Service Overview

• Designed to provide weight measurement data to a collector device

Associated Services (BLE)

• Device Information Service: Provides device metadata
• User Data Service: Manages user-specific data
• Body Composition Service: Offers additional body metrics
• Battery Service: Monitors battery status

Unique Identification ID (UUID)

• 128-bit value
• characteristic + BLEbase
• Battery Service: 0000180F-0000-1000-8000-00805F9B34FB
• Battery Characteristic: 00002A19-0000-1000-8000-00805F9B34FB

Services and Characteristics (BLE)

• Services are offered to other devices
• Services contain Characteristics that reflect the device's capability.
• Characteristics report data in a defined format

Car Repair Agency ML Model

• A car repair agency wants to predict waiting time using a machine learning (ML) model
• The agency has a curated dataset

Key Steps for Data Processing

• Preprocessing: Preparing raw data for analysis, such as calculating the magnitude of acceleration values
• Segmentation: Converting continuous time series data into fixed-length vectors
• Feature Extraction: Identifying and extracting relevant features from segmented data
• Classification: Classifying data into distinct categories

Raw Data and Preprocessing

• Raw data is intially processed to remove noise and prepare it more easily for analysis
• Preprocessing involves synchronizing and removing artifacts through methods such as calibration, unit conversion, normalization, resampling, and synchronization
• Acquistion can occur from several sensors placed at different body locations with different sampling rates for reasons like energy efficiency
• Sensor data may be corrupted or contain errors

Predicting Waiting Time Using Logistic Regression

• A car repair agency wants to predict the waiting time with a machine learning algorithm
• Steps include data preprocessing, feature selection, model training, model evaluation, and prediction

Steps for Logistic Regression Setup

• Convert all features to numeric ones and assume how many values are there
• Normalize each feature to have zero mean and unit variance
• Step 2: Define the model parameters, θ as a vector
• Split the data into training and validation sets with stratified sampling and k-cross validation
• Gradient descent using training data (Step 5)
• Evaluate on the validation set (Step 6)

Human Activity Recognition (HAR) Task

• Classify activities into predefined categories with 3-D accelerometer signals (Objective)
• Linear classification creates a linear boundary between samples of different classes in the feature space

Data Processing Considerations

• Access to Dataset: Ensure you have access to the dataset for training and evaluating your ML model
• Training large models on extensive data requires significant compute and memory resources
• Communicate over networks, but may be prone to errors
• Consider the trade-offs involved in access, resources, and security

Examples For ML Data leakage

• Reveals the joint efficiency of the mechanics of a store
• If I know a person had a flat tyre on a particular day, I can find out where it happened by finding a possible match from this dataset

Summary of Model Evaluation

• Large and Balanced Dataset: Use random sampling
• Imbalanced Dataset: Use stratified sampling
• Small Dataset: Use k-fold cross-validation
• Imbalanced Data: Avoid using accuracy use metrics like F1 score

Key Performance Metrics

• Accuracy/Precision/Recall
• F1 Scores
• Sensitivity/Specificity

Evaluation Metrics

• Sensitivity and Specificity are important while evaluating the results of a classification model
• These are particularly relevant in medical diagnostics
• Sensitivity is the amount of correctly identified people with a diagnosis

Evalution Metrics

• Specificity is the amount of identified healthy people without a diagnosis
• Percision measures how many retrieved items are relevant
• Recall measures how many relevant items are retrieved

What is Deep Learning

• Deep Learning involves learning discriminative representations directly from data
• This often referred to as end-to-end learning
• Modeling techniques are are less dependent on task