1. Artificial Intelligence Fundamentals & Search
AI search algorithms are classified as uninformed (blind) — no domain knowledge beyond the problem definition, e.g. BFS, DFS, uniform-cost search — or informed (heuristic) — using a heuristic function to estimate distance-to-goal and guide the search more efficiently, e.g. Greedy Best-First Search (always expands the node estimated closest to the goal, fast but not guaranteed optimal) and A* search (combines actual cost-so-far with the heuristic estimate, f(n) = g(n) + h(n), guaranteed optimal if the heuristic is admissible — never overestimates the true remaining cost).
Knowledge representation in classical AI uses techniques like propositional/predicate logic, semantic networks and production rules (if-then rule systems, the basis of early expert systems). The syllabus expects awareness that modern AI increasingly relies on machine learning rather than hand-coded rules for many tasks, but the classical search/representation techniques remain the conceptual foundation tested here.
f(n) = g(n) + h(n)
g(n) = actual cost from the start node to node n (known exactly)
h(n) = heuristic ESTIMATE of cost from n to the goal (a guess)
f(n) = estimated total cost of the cheapest path through n
A* expands the node with the LOWEST f(n) first.
Guaranteed to find the optimal path IF h(n) is admissible:
h(n) must NEVER overestimate the true remaining cost to the goal.If a heuristic ever overestimates the true remaining cost, A* can be tricked into abandoning the actual optimal path in favour of one that only looks cheaper on paper — admissibility is the one condition that guarantees A* still finds the true optimum despite using estimates.
2. Machine Learning: Types & Core Algorithms
Supervised learning trains on labelled input-output pairs to predict labels for new data — classification (discrete labels, e.g. spam detection) or regression (continuous values, e.g. price prediction). Unsupervised learning finds structure in unlabelled data — clustering (K-means, hierarchical clustering) or dimensionality reduction (PCA). Reinforcement learning has an agent learn by taking actions in an environment and receiving rewards/penalties, gradually learning a policy that maximises cumulative reward — no labelled dataset at all, just trial-and-error feedback (used in game-playing AI and robotics).
Common supervised algorithms: linear regression (fits a straight line/hyperplane to predict a continuous value), decision trees (splits data on feature thresholds into a tree of decisions, human-readable), k-Nearest Neighbours (classifies a new point by majority vote of its k closest labelled neighbours, no explicit training phase), and Naive Bayes (a probabilistic classifier based on Bayes' theorem, assuming features are conditionally independent given the class — a strong, often unrealistic assumption that nonetheless works surprisingly well in practice, e.g. for spam filtering).
3. Cloud Computing & Internet of Things (IoT)
Cloud computing's three standard service models: IaaS (Infrastructure as a Service — rents raw compute/storage/networking, e.g. virtual machines; the customer manages the OS upward), PaaS (Platform as a Service — provides a managed runtime/platform for deploying applications without managing servers, e.g. a managed application-hosting platform), and SaaS (Software as a Service — a complete, ready-to-use application delivered over the internet, e.g. webmail). As you move IaaS → PaaS → SaaS, the provider manages progressively more of the stack, and the customer's control (and responsibility) progressively decreases.
Deployment models: public cloud (shared infrastructure, offered by a third-party provider to any customer), private cloud (dedicated to a single organisation), and hybrid cloud (a mix of both, e.g. sensitive data kept on a private cloud with burst capacity from a public cloud). An IoT (Internet of Things) system is typically described in layers: a Perception/Sensing layer (physical sensors/actuators collecting data), a Network layer (transmitting that data, e.g. via Wi-Fi, Zigbee, cellular), and an Application layer (processing the data into a useful service, e.g. a smart-home dashboard) — some models add a middle processing/middleware layer for data aggregation and analytics between the network and application layers.
Ask "what does the customer still have to manage?" IaaS: everything from the OS up. PaaS: only the application code and data. SaaS: nothing — just use the finished application. This single question resolves nearly every cloud-service-model exam item.
4. Hands-on Exercise
Classify ML scenarios and match cloud/IoT terms to examples
This unit is tested through recognition, so build your own quick-reference examples.
Part 1 — Machine learning classification:
- For each, name the ML type (supervised classification, supervised regression, unsupervised, or reinforcement) and the reasoning: (a) predicting house prices from square footage, (b) grouping news articles into topics with no predefined topic list, (c) a game-playing agent that improves by winning/losing matches, (d) detecting whether an email is spam using a dataset of pre-labelled emails.
- For scenario (a), name one specific algorithm suited to it and one suited to scenario (b).
Part 2 — Cloud/IoT mapping:
- Classify each as IaaS, PaaS or SaaS: a rented virtual machine with only an OS installed; a webmail service you just log into and use; a managed platform where you deploy your application code but never touch the underlying server.
- Sketch the three (or four) layers of a smart home IoT system, naming one real component at each layer (e.g. a temperature sensor, a Wi-Fi router, a mobile app).
5. Exam-Style Practice (UGC NET Pattern)
Five NTA-pattern questions closing out Paper 2's syllabus content.
Q1
A search algorithm that uses f(n) = g(n) + h(n) to select which node to expand next, and is guaranteed optimal if the heuristic never overestimates true cost, is:
A) Breadth-First Search
B) Greedy Best-First Search
C) A* Search
D) Depth-First Search
A search algorithm that uses f(n) = g(n) + h(n) to select which node to expand next, and is guaranteed optimal if the heuristic never overestimates true cost, is:
A) Breadth-First Search
B) Greedy Best-First Search
C) A* Search
D) Depth-First Search
Correct answer: C) A* Search. A* search specifically combines the actual cost-so-far g(n) with a heuristic estimate h(n), and is guaranteed to find the optimal path when h(n) is admissible (never overestimates the true remaining cost).
Q2
A machine learning approach where an agent learns a policy by taking actions and receiving rewards or penalties from its environment, with no labelled training dataset, is:
A) Supervised learning
B) Unsupervised learning
C) Reinforcement learning
D) Semi-supervised learning
A machine learning approach where an agent learns a policy by taking actions and receiving rewards or penalties from its environment, with no labelled training dataset, is:
A) Supervised learning
B) Unsupervised learning
C) Reinforcement learning
D) Semi-supervised learning
Correct answer: C) Reinforcement learning. Reinforcement learning is defined by trial-and-error interaction with an environment, using reward/penalty feedback to learn a policy — no labelled input-output dataset is involved, unlike supervised learning.
Q3
A cloud service where the provider manages everything up through the operating system, and the customer manages only their application code and data, is best classified as:
A) IaaS (Infrastructure as a Service)
B) PaaS (Platform as a Service)
C) SaaS (Software as a Service)
D) DaaS (Desktop as a Service)
A cloud service where the provider manages everything up through the operating system, and the customer manages only their application code and data, is best classified as:
A) IaaS (Infrastructure as a Service)
B) PaaS (Platform as a Service)
C) SaaS (Software as a Service)
D) DaaS (Desktop as a Service)
Correct answer: B) PaaS (Platform as a Service). PaaS provides a managed platform/runtime (including the OS) so the customer only has to manage their application code and data, without managing servers directly — this sits between IaaS (customer manages the OS upward) and SaaS (customer manages nothing).
Q4
In a typical layered IoT architecture, which layer is responsible for physically collecting data via sensors and actuators?
A) Application layer
B) Network layer
C) Perception/Sensing layer
D) Presentation layer
In a typical layered IoT architecture, which layer is responsible for physically collecting data via sensors and actuators?
A) Application layer
B) Network layer
C) Perception/Sensing layer
D) Presentation layer
Correct answer: C) Perception/Sensing layer. The Perception (or Sensing) layer is the physical layer of an IoT system, comprising the sensors and actuators that directly collect environmental data or act on the physical world.
Q5
Naive Bayes classifiers are based on Bayes' theorem and rest on which key simplifying assumption?
A) That all features are perfectly correlated with each other
B) That features are conditionally independent given the class label
C) That the dataset must be completely unlabelled
D) That only continuous numeric features can be used
Naive Bayes classifiers are based on Bayes' theorem and rest on which key simplifying assumption?
A) That all features are perfectly correlated with each other
B) That features are conditionally independent given the class label
C) That the dataset must be completely unlabelled
D) That only continuous numeric features can be used
Correct answer: B) That features are conditionally independent given the class label. Naive Bayes assumes conditional independence between features given the class — an assumption that is often unrealistic in practice, yet the classifier still performs surprisingly well on many real tasks, such as spam filtering.