AI Project Cycle: Scoping, Acquisition & Planning

• The AI Project Cycle has five stages: Problem Scoping, Data Acquisition, Data Exploration, Modelling, and Evaluation.

Introduction to AI Project Cycle

• Imagine making a greeting card for your mother's birthday as an analogy for understanding the steps in the AI Project Cycle.
• Those steps include: researching ideas, listing required materials, gathering materials, making the card, correcting mistakes, and gifting the card.

Planning Tasks

• The steps in making the card reflect how humans plan tasks, consciously or subconsciously.
• When developing an AI project, the AI Project Cycle offers a framework to achieve set goals.

Problem Scoping

• It involves defining the project's goal and understanding the parameters affecting the problem.
• Involves looking at what parameters affect the problem so the full picture is understood.

Data Acquisition

• It involves gathering data to understand the parameters related to the problem scoping.
• It is done through collecting data from reliable and authentic sources.
• Collect data in large quantities, then represent it visually through graphs, databases, flow charts, or maps to interpret patterns.

Modeling

• Use the patterns found during data exploration to decide on the type of model to build.
• Test selected models to determine the most efficient one.
• The most efficient model becomes the base for the AI project, enabling algorithm development.

Model Testing and Evaluation

• The model, once complete, needs testing with new data to evaluate and improve it.
• After evaluation, the project cycle is complete, resulting in the AI project.

Problem Scoping in Detail

• Problem scoping involves identifying a problem and envisioning its solution.
• It may involve observing the immediate surroundings or considering the Sustainable Development Goals announced by the UN, aiming to achieve them by 2030.

4Ws Problem Canvas

• Use the 4Ws Problem Canvas, which identifies key problem elements through Who, What, Where, and Why blocks.
• "Who" analyzes the people affected, determining the 'Stakeholders' and what is known about them.
• "What" determines the nature of the problem and involves gathering evidence to prove its existence through newspaper articles, media, or announcements.
• "Where" focuses on where the problem arises and the context or situation in which stakeholders experience it.
• "Why" considers the benefits stakeholders would get from the solution and how it benefits them and society.
• After filling the 4Ws Problem canvas, summarization occurs using a Problem Statement Template to consolidate key points for future reference.

Data Acquisition Details

• Data refers to pieces of information or facts and statistics collected for reference or analysis.
• An Al project needs to be trained using data to accurately predict an output.
• In Al, previous data is known as training data, and what is predicted is known as testing data.
• Training data must be relevant and authentic for the AI project to be efficient.
• Focus on data features which involves identifying the data features required to address the problem statement.

Data Collection Methods

• Surveys: Involve collecting data through questionnaires or interviews.
• Web Scraping: It involves extracting data from websites.
• Sensors: It involves using devices to collect data.
• Cameras: It involves capturing visual data.
• Observations: It involves gathering data through direct observation.
• API (Application Program Interface): It involves using interfaces to acquire data from different applications.
• It's important to use reliable, open-source data and avoid extracting private data.
• Open-sourced websites hosted by the government can be reliable sources for general information.

Data Exploration Details

• It involves analyzing the data to visualize it so that it is user friendly so it is easy to understand.
• It helps users quickly grasp trends, relationships, and patterns within the data, define strategies for model selection, and communicate effectively.

Modeling Details

• Graphical representations make data understandable for humans, which makes it easy to discover trends and patterns.
• Machines need data in the most basic form of numbers (binary) for analysis, going in for mathematical representations to discover patterns and trends.
• The ability to mathematically describe the relationship between parameters is the heart of every AI model.

Al models

• AI models classifies as Learning Based, Machine Learning, Deep Learning, or follows a Rule Based approach.

Rule-Based Approach

• The rules are defined by the developer.
• The machine follows the rules defined by the developer.
• The learning is static, meaning the machine doesn't consider changes in the original training dataset.

Learning-Based Approach

• The machine learns by itself.
• The Al model gets trained on the data fed to it and this allows it to design a model which is adaptive to changes in data.

Supervised Learning

• The dataset is labelled.
• Grades students get on their exams is are labels which categorise the students according to their marks.

Supervised Learning Models

• There are two types of supervised learning models, classification and regression.

Classfication Models

• Where the data is classified according to the labels.

Regression Models

• Such models work on continuous data.

Unsupervised Learning

• Model works on unlabelled datasets.
• It identifies relationships, patterns, and trends out of the data which is fed into it.

Algorithms

• There are two types of unsupervised learning algorithmic processes: Clustering and Dimensionality Reduction.

Clustering

• unsupervised learning algorithm which can cluster the unknown data according to the patterns or trends identified out of it.

Dimensionality Reduction

• Various entities exist beyond 3-Dimensions, which means it cannot be visualised.
• As dimensions are reduced, the information which the data contains starts getting distorted.

Evaluation in Detail

• Model, once made and trained, requires proper testing to calculate its efficiency and performance.
• The model is tested with the help of Testing Data, which was separated out of the acquired dataset at the Data Acquisition stage.
• Efficiency of an Al model is calculated on the basis of parameters like Accuracy, Precision, Recall, and F1 Score.

Neural Networks

• They are loosely modelled after how neurons in the human brain behave.
• Key advantage is that neural networks extract data features automatically without needing input from the programmer.
• They are a system of organizing machine learning algorithms to perform certain tasks and are efficient and fast to solve problems for which the dataset is very large.
• Neural networks are divided into multiple layers where each layer is further divided into several blocks called nodes.
• The first layer of a Neural Network is the input layer, which functions to acquire data and feed it to the Neural Network.
• The layers next to input layers are the hidden layers in which the processing happens; its invisible to the user.
• Each node of these hidden layers has its own machine learning algorithm which executes on the data received from the input layer.
• There can be multiple hidden layers in a neural network system, with number depending on the complexity of the function.
• The last hidden layer passes the final processed data to the output layer, which gives it to the user as the final output; meant for user-interface.
• Neural Network systems are modelled on the human brain and nervous system.
• Neural networks extract data features automatically without needing input from the programmer.
• Every neural network node is essentially a machine learning algorithm.
• Neural Networks are useful when solving problems for which the data set is very large.