e

iv
I)Uncertainty in AI:

ad
Uncertainty in AI is the inability of an AI system to make perfectly accurate predictions or
an
decisions due to incomplete, ambiguous, noisy, or probabilistic information.
In Artificial Intelligence (AI), uncertainty refers to the lack of complete knowledge about the
environment, data, model, or outcomes, which makes predictions, decisions, or inferences less
than 100% certain. AI systems must account for uncertainty to make accurate and reliable
rR

decisions, especially in dynamic environments where information is inconsistent or evolving. AI

models handle uncertainty by using probabilistic methods, fuzzy logic, and Bayesian inference.
Proper uncertainty representation enables AI systems to make informed predictions and improve
reliability in real-world applications.
Sources of uncertainty in AI:
i
ih
M
 e
iv
ad
Techniques for addressing uncertainty in AI:

1. Probabilistic Logic Programming (PLP)

an
●​ Extension of logic programming where facts and rules are annotated with probabilities.
●​ Key Idea: Deals with randomness and likelihood of events.
●​ Advantages: Provides a structured way to represent uncertainty mathematically.
●​ Applications: Weather prediction, medical diagnosis, decision-making under risk.
rR

●​ Example:
○​ Fact: Rain(0.6). → There is a 60% chance of rain.
○​ Rule: WetGround :- Rain. → The ground is wet with probability 0.6.

2. Fuzzy Logic Programming (FLP)

i
ih

●​ Programming paradigm where truth values range between 0 and 1 instead of being only
True/False.
●​ Key Idea: Models vagueness and imprecision.
M

●​ Advantages: Handles linguistic variables like “hot,” “cold,” “tall,” etc.

●​ Applications: Control systems, recommendation systems, natural language processing.
●​ Example:
○​ Fact: Rainy(0.8). → Weather is 80% rainy.
○​ Rule: TakeUmbrella :- Rainy. → Decision to take an umbrella has strength 0.8.

3. Nonmonotonic Logic Programming (NMLP)

●​ Logic programming approach where conclusions can be withdrawn when new

information is added.
 ●​ Key Idea: Captures “commonsense reasoning” where knowledge is incomplete.
●​ Advantages: Adapts to dynamic knowledge bases.
●​ Applications: Expert systems, legal reasoning, planning systems.
●​ Example:
○​ Rule: Bird(X) → Flies(X).
○​ Fact: Bird(Penguin).
○​ New info: Penguin → ¬Flies. → Revises earlier belief.

4. Hybrid Logic Programming (HLP)

e
●​ Combines two or more logic paradigms (probabilistic, fuzzy, nonmonotonic) to deal with
complex uncertainty.

iv
●​ Key Idea: No single method is sufficient; hybrid models capture multiple kinds of
uncertainty.
●​ Advantages: More flexibility and robustness in reasoning.

ad
●​ Applications: Robotics, multi-agent systems, smart decision support.
●​ Example:
○​ Rain(0.7). (Probabilistic)
○​ Cold(0.6). (Fuzzy)
an
○​ Rule: StayHome :- Rain ∧ Cold.
○​ Inference: Stay home with weighted confidence.
i rR
ih
M

Inference using full joint distributions:

Inference using full joint distributions involves using a table or representation that
lists the probabilities of all possible combinations of variable values to calculate
probabilities of specific events or conditions. This method allows us to answer various
probabilistic queries by summing up relevant entries in the joint distribution table.
Probabilistic Inference:

e
iv
ad
an
Conditional probability:
i rR
ih
M

Refer eg from og doc

II)Bayes Theorem:
 e
iv
Applications of Bayes Theorem:

ad
Bayesian inference is very important and has found application in various activities, including
medicine, science, philosophy, engineering, sports, law, etc., and Bayesian inference is directly
derived from Bayes theorem. Some of the Key Applications are:
● Medical Testing → Finding the real probability of having a disease after a positive test.
an
● Spam Filters → Checking if an email is spam based on keywords.
● Weather Prediction → Updating the chance of rain based on new data.
● AI & Machine Learning → Used in Naive Bayes classifiers to predict outcomes.
rR

Semantics of Bayesian Network:

Bayesian Belief Network (BBN) is a graphical model that represents the probabilistic
relationships among variables. It is used to handle uncertainty and make predictions or decisions
based on probabilities.
● Graphical Representation: Variables are represented as nodes in a directed acyclic graph
i

(DAG), and their dependencies are shown as edges.

ih

● Conditional Probabilities: Each node's probability depends on its parent nodes, expressed as
● P(Variable | Parent)
● Probabilistic Model: Built from probability distributions, BBNs apply probability theory for tasks
M

like prediction and anomaly detection.

Variables (nodes):
●​ Rain
●​ Sprinkler
●​ Wet Ground
Connections:
●​ Rain affects Wet Ground.
●​ Sprinkler affects Wet Ground.
Rain → Wet Ground ← Sprinkler
Inference in Bayesian networks:
Bayesian networks are graphical models that represent predictions, decisions or inferences in the
form of probability and show how actions and outcomes are connected and by how much.
A Bayesian Network is like a story map that shows how things are connected and how likely
they are to [Link]: If it rains, the ground gets wet. If the sprinkler is on, the ground also
gets wet.
Inference in a Bayesian Network = Using what we know to figure out what we don’t know.
The most common types of queries are:

e
1. Marginalization: Determining the probability distribution of a subset of variables, ignoring the
values of all other variables.

iv
2. Conditional Probability: Computing the probability distribution of a subset of variables given
evidence observed on other variables.

ad
Mathematically, if X are the query variables and E are the evidence variables with observed
values e, the goal is to compute P(X∣E=e)
The methods of Exact Inference are:
1. Variable Elimination (VE): Remove extra variables one by one.
Example:​
an
Suppose you want to know: What’s the probability it rains (R) given clouds (C)?​
But your network also has variables like wind (W) and temperature (T).
●​ VE says: "Let’s ignore (eliminate) W and T step by step by adding up their probabilities."
●​ Finally, you only get the relationship between R and C.
rR

2. Junction Tree Algorithm (JTA): Turn the network into a tree of groups (cliques) and pass
information.
Example: In a medical diagnosis BN, you have:
●​ Diseases → Symptoms → Test results.​
Instead of handling each separately, JTA groups them (e.g., {Disease, Symptom} as one
i

group, {Symptom, Test} as another).

ih

●​ These groups pass messages (probabilities) to each other like a tree.

●​ Result: Faster and more organized inference than VE.
3. Belief Propagation (BP): Nodes send messages (probabilities) to neighbors.
M

Example:​
Say your network is: Weather → Road Wet → Traffic Jam.
●​ If you see evidence: "Road is wet," that message is passed backward to update belief
about Weather, and forward to update belief about Traffic Jam.
●​ In a tree network, this gives exact answers.
●​ If the network has loops (e.g., Weather ↔ Traffic reports ↔ Road Wet), it only gives
approximate answers.
4. Global Semantics: Refer vid
Decision Theory:
Decision Theory is a branch of Artificial Intelligence that deals with making rational choices
under uncertainty. It combines probability theory (for modeling uncertainty) and utility theory
(for modeling preferences or outcomes).
Decision theory under uncertainty deals with making choices when the outcomes of those
choices are not fully predictable.
Unlike decision-making under risk, where probabilities of outcomes are known, decision-making
under uncertainty involves situations where probabilities are either unknown or cannot be
meaningfully assigned.
This lack of probability information makes assessing potential impacts and risks challenging.

e
In decision theory, utility is a numerical representation of a person’s preference or satisfaction
from outcomes. It allows us to compare, rank, and optimize decisions — especially when

iv
outcomes are uncertain.
A utility is a number that represents how much you like or dislike an outcome.

ad
It measures the usefulness, happiness, cost, or value of that outcome to you.
Higher utility = more desirable outcome & Lower (or negative) utility = less desirable or bad
outcome.E.G.: (Write more) use same eg for MUE and Decision Network
an
rR

Maximum Expected Utility (MEU) in Decision Theory

● Maximum Expected Utility (MEU) is a core concept in decision theory under risk, where
outcomes are uncertain but probabilities of those outcomes are known.
i
ih
M
 e
There are some Key Constraints associated with:
1. Completeness (or Comparability)
●​ Meaning: For any two options, you can always compare them. You either prefer A over

iv
B, B over A, or are indifferent.
●​ Example:

ad
○​ If choices are Pizza (A) and Burger (B) → You must be able to say:
■​ "I prefer Pizza" OR "I prefer Burger" OR "Both are equal for me."
2. Transitivity
●​ Meaning: If you prefer A over B, and B over C, then you must prefer A over C.
●​ Example:
an
○​ If Pizza > Burger and Burger > Salad, then logically Pizza > Salad.
○​ Without this, preferences would be inconsistent.
3. Continuity
●​ Meaning: If A is preferred to B, and B is preferred to C, then there exists some
rR

mixture/lottery of A and C that you would find equally good as B.

●​ Example:
○​ Suppose Pizza > Burger > Salad.
○​ Then a 50-50 chance of Pizza or Salad might feel just as good as Burger.
4. Independence (Substitutability)
i

●​ Meaning: If you prefer A over B, then adding the same third option C with equal
ih

probability should not change your preference.

●​ Example:
M

○​ If Pizza > Burger, then between:

■​ 50% Pizza + 50% Ice Cream
■​ 50% Burger + 50% Ice Cream
○​ You should still prefer the Pizza option.
5. Monotonicity (for lotteries)
●​ Meaning: If you have a lottery, and the probability of a better option increases, then that
lottery is preferred.
●​ Example:
○​ If you prefer Pizza > Salad, then:
 ■​ Lottery 1: 70% Pizza, 30% Salad
■​ Lottery 2: 50% Pizza, 50% Salad
○​ You should prefer Lottery 1 (higher chance of Pizza).
6. Non-satiation (Optional / Economic Axiom)
●​ Meaning: “More is better than less” (assuming no negatives).
●​ Example:
○​ If you like chocolate, then 2 chocolates > 1 chocolate.
Decision Network:
A Decision Network (also called an Influence Diagram) is an extension of a Bayesian Network
used in decision theory to model decision-making under uncertainty.

e
Components of Decision Network:

iv
Markov Decision Model:
ad
an
A Markov Model is a way to predict what happens next in a system where:
●​ The system can be in one of a few states (like sunny, rainy).
●​ It moves between states with certain probabilities.
●​ The future depends only on the present, not on the past history.​
rR

This special rule is called the Markov property.

Example: Weather Prediction
Let’s say we want to model the weather with two states:
●​ Sunny (S)
●​ Rainy (R)
i

We define transition probabilities (chances of moving from one state to another):

ih

●​ If Sunny today:
○​ Tomorrow Sunny = 0.8
○​ Tomorrow Rainy = 0.2
M

●​ If Rainy today:
○​ Tomorrow Sunny = 0.4
○​ Tomorrow Rainy = 0.6
How it works:
●​ Suppose today is Sunny (S).
●​ Probability tomorrow will also be Sunny = 0.8
●​ Probability tomorrow will be Rainy = 0.2
If tomorrow turns out Rainy (R), then the next day’s prediction will use the Rainy row:
 ●​ Day after tomorrow Sunny = 0.4
●​ Day after tomorrow Rainy = 0.6
Types of Markov Models
1. Markov Chain (Discrete-time)
○ Transitions happen in discrete steps.
○ Example: Weather model – Sunny → Rainy → Cloudy.
2. Continuous-time Markov Chain
○ Transitions can happen at any time.
3. Hidden Markov Model (HMM)
○ The actual state is hidden but you can observe some outputs (emissions).

e
○ Example: Speech recognition, where the true phoneme sequence is hidden, but you observe
sound signals.

iv
Applications:
1. Weather Prediction

ad
●​ Point 1: Forecasts future weather based on the current state (sunny, rainy, etc.).
●​ Point 2: Uses transition probabilities between weather states over time.
2. Speech Recognition
●​ Point 1: Models sequences of spoken words or phonemes as states.
●​ Point 2: Helps computers recognize speech by predicting likely next sounds.
an
3. Google PageRank
●​ Point 1: Web pages are treated as states; links are transitions.
●​ Point 2: The probability of moving from one page to another determines page
importance.
rR

4. Stock Market Analysis

●​ Point 1: Models stock prices or market trends as states (up, down, stable).
●​ Point 2: Helps predict future price movements using transition probabilities.
5. Robot Navigation / Path Planning
●​ Point 1: The robot's location is modeled as a state; moves to next positions are transitions.
i

●​ Point 2: Helps robots make probabilistic decisions in uncertain environments.

ih

III) Hidden Markov Model:

A Hidden Markov Model (HMM) is a statistical model in which the system being modeled is
M

assumed to follow a Markov process with hidden states.

●​ The actual states of the system are hidden (not directly observable).
●​ We can only observe outputs (or emissions) that depend probabilistically on these hidden
states.
1.​ States: The system has a set of hidden states (e.g., Rainy, Sunny).
2.​ Observations: For each state, there is a probability of producing certain observable
outputs (e.g., Umbrella seen = observation).
3.​ Markov Property: The next state depends only on the current state, not on past states.
4.​ Components of HMM:
 ○​ N: Number of hidden states
○​ M: Number of observation symbols
○​ Transition probabilities (A): P(state_t+1 | state_t)
○​ Emission probabilities (B): P(observation | state)
○​ Initial probabilities (π): P(state_0)
Algorithms in HMM:
Step 1: Define the State and Observation Spaces
●​ State space: All possible hidden states of the system (e.g., Sunny, Rainy).
●​ Observation space: All possible observations we can see (e.g., Umbrella, No Umbrella).
Step 2: Define Initial State Distribution

e
●​ Assign probabilities for each hidden state being the starting state.
●​ Example: P(Sunny at day 1) = 0.6, P(Rainy at day 1) = 0.4

iv
Step 3: Define State Transition Probabilities
●​ Probabilities of moving from one state to another.

ad
●​ Represented as a transition matrix.
●​ Example: P(Sunny→Rainy) = 0.3, P(Rainy→Rainy) = 0.6
Step 4: Define Observation Likelihoods (Emission Probabilities)
●​ Probabilities of seeing each observation from each state.
●​ Represented as an emission matrix.
an
●​ Example: P(Umbrella | Rainy) = 0.9, P(No Umbrella | Sunny) = 0.8
Step 5: Train the Model
●​ Estimate transition and emission probabilities from data.
●​ Use Baum-Welch (Forward-Backward) Algorithm for iterative learning until
rR

convergence.
Step 6: Decode the Most Likely Sequence of Hidden States
●​ Given observations, find the most probable hidden states using the Viterbi Algorithm.
●​ Helps in prediction, sequence classification, or pattern detection.
Step 7: Evaluate the Model
i

●​ Measure performance using metrics like accuracy, precision, recall, or F1 score.

ih

●​ Ensures the model predicts hidden states correctly and reliably.

Simple Example:
Problem: Predict the weather (hidden state) based on whether people carry umbrellas
M

(observation).
●​ Hidden states: Sunny (S), Rainy (R)
●​ Observations: Umbrella (U), No Umbrella (N)
●​ Example probabilities:
○​ Transition: P(R→R)=0.6, P(R→S)=0.4, P(S→S)=0.7, P(S→R)=0.3
○​ Emission: P(U|R)=0.9, P(N|R)=0.1, P(U|S)=0.2, P(N|S)=0.8
●​ Observation sequence: U, U, N → Use Viterbi to find the most likely weather sequence
(R, R, S).
Markov Decision Processes:

e
iv
ad
an
i rR

For Uncertainty:
ih
M

●​ States: Positions in a grid (S1…G)

●​ Actions: Move Up, Down, Left, Right
●​ Uncertainty: Actions may fail (e.g., 80% go intended way, 20% slip)
●​ Rewards: Goal = +10, Trap = -5, others = 0
●​ Policy: Choose actions that maximize expected reward
Example Path: S1 → S2 → G
●​ Moves Right from S1 → S2 (80% success)
●​ Moves Right from S2 → G (80% success, +10 reward)
 ●​ Avoids Trap (T) to minimize risk
IV)Gaussian Mixture Model
A Gaussian Mixture Model (GMM) is a probabilistic model that assumes that data points are
generated from a mixture of several Gaussian distributions with unknown parameters.
●​ Each Gaussian represents a cluster or component in the data.
●​ Useful for modeling complex, multi-modal data where a single Gaussian is insufficient.
1.​ Mixture Components: Each cluster has its own Gaussian distribution (mean μ, variance
σ²).
2.​ Weights: Each component has a weight (π) representing its proportion in the dataset.

e
3.​ Soft Clustering: Unlike k-means, GMM assigns probabilities of belonging to each cluster,
not just a hard assignment.​

iv
ad
4.​ Parameters: The model estimates means, variances, and mixture weights for each
an
Gaussian.
5.​ Maximum Likelihood Estimation: GMM parameters are usually learned using the
Expectation-Maximization (EM) algorithm.
Each Gaussian in the mixture is defined by:
rR

● Mean (μ): The center of the distribution.

● Covariance (Σ): Describes the spread and orientation.
● Mixing coefficient (π): Represents the proportion of each Gaussian in the mixture.
Algorithm (Simplified EM for GMM):
1.​ Initialization: Choose initial means, variances, and weights for each Gaussian component.
i

2.​ Expectation Step (E-step): Compute the probability that each data point belongs to each
ih

Gaussian.
3.​ Maximization Step (M-step): Update the means, variances, and weights based on these
probabilities.
M

4.​ Iterate: Repeat E-step and M-step until convergence.

Simple Example:You have heights of people in a class, but the class has boys and girls mixed.
●​ Boys’ heights ≈ Gaussian(mean=170 cm, std=5)
●​ Girls’ heights ≈ Gaussian(mean=160 cm, std=5)
You don’t know which height belongs to boys or girls.
GMM assumes the data comes from a mixture of two Gaussians and tries to:
●​ Figure out the two groups (boys & girls)
●​ Estimate their mean and variance
So after GMM:
●​ It can say, “This height likely belongs to boys” or “likely girls.

Real-World Applications:
1.​ Image Segmentation: Separating foreground and background regions in images.
2.​ Speech Recognition: Modeling distributions of acoustic features for phonemes.
3.​ Anomaly Detection: Detecting outliers in multi-modal data.
4.​ Finance: Modeling stock returns that follow multiple regimes.
5.​ Clustering: Soft clustering of complex datasets where clusters overlap.
To fit a Gaussian Mixture Model to the data we use the Expectation-Maximization (EM)

e
algorithm which is an iterative method that optimize the parameters of the Gaussian
distributions like mean, covariance and mixing coefficients. It works in two main steps:

iv
● Expectation Step (E-step): In this step the algorithm calculates the probability that
each data point belongs to each cluster based on the current parameter estimates

ad
(mean, covariance, mixing coefficients).
● Maximization Step (M-step): After estimating the probabilities the algorithm
updates the parameters (mean, covariance and mixing coefficients) to better fit the data.
These two steps are repeated until the model converges meaning the parameters no longer
change significantly between iterations. Here’s a simple breakdown of the GMM process:
an
1. Initialization: Start with initial guesses for the means, covariances and mixing coefficients of
each Gaussian distribution.
2. E-step: For each data point, calculate the probability of it belonging to each Gaussian
distribution (cluster).
rR

3. M-step: Update the parameters (means, covariances, mixing coefficients) using the
probabilities calculated in the E-step.
4. Repeat: Continue alternating between the E-step and M-step until the log-likelihood of the
data (a measure of how well the model fits the data) converges.
i

V)Cognitive Computing
ih

Cognitive computing refers to systems that simulate human thought processes to solve complex
problems. It uses technologies like artificial intelligence (AI), machine learning (ML), natural
language processing (NLP), and computer vision to mimic how humans reason, learn, and make
M

decisions.
CC is a technology mostly used to study human behaviour,human brain,and personality.
Cognitive computing is a technology approach that enables humans to collaborate with
machines.
To analyze in context all types of data, from structured data in databases to unstructured data in
text, images, voice, sensors, and video, it searches the human methodology, storage method,
method of data organization and data representation.
It is a type of machine learning to improve the performance and get accurate output.
Combination of Existing algorithms and CC which produces the advanced version of learning
methods.
Foundation of Cognitive Computing:
Cognitive computing is a technology approach that enables humans to collaborate with
machines.
To analyze in context all types of data, from structured data in databases to unstructured data in
text, images, voice, sensors, and video.
A cognitive system has three fundamental principles as described below:
Learn—A cognitive system learns. The system leverages data to make inferences about a
domain, a topic, a person, or an issue based on training and observations from all varieties,

e
volumes, and velocity of data.
Model—To learn, the system needs to create a model or representation of a domain (which includes

iv
internal and potentially external data) and assumptions that dictate what learning algorithms are used.
Generate hypotheses—A cognitive system assumes that there is not a single correct answer. A
cognitive system uses the data to train, test, or score a hypothesis.

ad
an
rR

Steps:
1.​ Data Collection → CC systems gather huge amounts of data from different sources (text,
images, speech, sensors, etc.).​
i

Eg: A hospital system collects patient records, scans, and doctor’s notes.
ih

2.​ Understanding → They use AI, NLP, and machine learning to understand the meaning of
data, even if it is unstructured or incomplete.​
Eg: The system reads medical reports and understands symptoms.
M

3.​ Reasoning → The system compares the input with past knowledge, finds patterns, and
evaluates different solutions.​
Eg: It matches symptoms with thousands of past cases to suggest possible diseases.
4.​ Learning → Cognitive systems improve over time by learning from new data and
feedback.​
Eg: The more cases it processes, the more accurate its diagnosis becomes.
5.​ Interaction → They interact with humans in natural language (text or voice) and explain
their suggestions.​
Eg: The system tells the doctor: “Based on symptoms and records, pneumonia is likely.”
 e
iv
ad
an
rR

Elements of Cognitive System: I Don’t Cook Tasty Omelets During Cold Late Party Celebrations
1.​ Infrastructure & Deployment
○​ The hardware, cloud, or on-premise setup where CC runs.
○​ Example: Servers, cloud platforms like IBM Watson Cloud.
i

2.​ Data Access & Management

ih

○​ Collecting, storing, and organizing data (structured + unstructured).

○​ Example: Database + text + images all stored in one system.
3.​ Corpus (Knowledge Base)
M

○​ A large collection of domain-specific data or documents.

○​ Example: Medical reports corpus for healthcare CC system.
4.​ Taxonomies
○​ Hierarchical classification of terms/concepts.
○​ Example: “Animal → Mammal → Dog → Labrador.”
5.​ Ontologies
○​ Relationships between concepts (more advanced than taxonomy).
○​ Example: “Doctor treats Patient” (relation between terms).
6.​ Data Analytics Services
 ○​ Tools that process and analyze data.
○​ Example: Predicting customer buying behavior.
7.​ Continuous Machine Learning
○​ System keeps learning and improving with new data.
○​ Example: Chatbot gets smarter as it talks to users.
8.​ Learning Process
○​ How the system models data, tests, and updates knowledge.
○​ Example: Training a CC model on patient history.
9.​ Presentation & Visualization
○​ Showing results in easy-to-understand graphs, charts, or dashboards.

e
○​ Example: Data insights shown in Power BI/visual charts.
10.​Cognitive Applications

iv
●​ Final apps that solve real problems using CC.
●​ Example: IBM Watson for healthcare, Siri, Google Assistant.

ad
Design Principles for Cognitive Systems:
Design principles are guidelines, biases and design considerations that designers apply with
discretion.”Some Smart Monkeys Simplify Really Clear Rules, Playing Very Intelligently”
Standardize: Keep things consistent so users don’t get confused.​
an
Example: All traffic lights worldwide use red = stop, green = go.
Use stereotypes: Follow common expectations of people.​
Example: A floppy disk icon = save, because people expect it.
Match controls to layout: Controls should match how things look.​
Example: Stove knobs arranged in the same pattern as burners.
rR

Simplify information: Don’t overload with too much detail.​

Example: Car dashboard shows a fuel gauge, not a full report of fuel chemistry.
Present information in right detail: Use the right style (digital, analog, graph).​
Example: Speedometer (analog) is quick to read, but digital clock gives exact time.
Present clear images: Info should be visible, clear, and easy to interpret.​
i

Example: Road signs are big, bold, and simple.

ih

Use redundancies: Show the same info in more than one way.​
Example: Stop sign = red color + octagon shape + word STOP.​
M

Use patterns: Patterns help people understand faster.​

Example: A bar chart is easier to read than a list of numbers.​
Provide variable stimuli: Change style to keep attention.​
Example: Fire alarm = loud sound + flashing light so people don’t ignore it.
Instant feedback: Tell users immediately if their action worked.​
Example: When you press an elevator button, it lights up to confirm.
Cognitive System Architecture:
 1.​ Environment
○​ The system is connected to the real world (data, situations, humans).
○​ Information flows in and actions flow out.

e
2.​ Sensors (Perception)
○​ Like human senses (eyes, ears, etc.).

iv
○​ The system collects raw data: text, images, sound, sensor readings.
3.​ Learning & Reasoning
○​ The system learns from data and past experiences.

ad
○​ It applies logic to make sense of new situations.
4.​ Knowledge & Models
○​ Stores what the system has already learned.
○​ Like a memory + library of models.
an
5.​ Planning & Control (Cognition)
○​ This is the “brain” that decides what to do next.
○​ It uses input from sensors + knowledge to plan actions.
6.​ Actuators (Action)
rR

○​ The system performs actions (like a robot moving, or software giving results).
7.​ Human Interaction
○​ The system communicates results/decisions back to humans.
○​ Humans can also give new feedback → system keeps learning.
VI) Natural Language Processing (NLP) in Cognitive Systems
i

NLP is a branch of AI that enables computers to understand, interpret, and generate human
ih

language.
NLP Techniques : %write more theory on ur own%
●​ Text Preprocessing – Cleaning text before analysis (tokenization, stopword removal).​
M

Example: "I am happy!" → ["I", "am", "happy"].

●​ Syntax & Parsing – Understanding grammar structure of sentences.​
Example: "The cat sat on the mat" → cat = noun, sat = verb.
●​ Semantic Analysis – Finding meaning in text.​
Example: "Apple" → fruit OR company (depends on context).
●​ Information Extraction – Pulling entities & relations from text.​
Example: "Virat Kohli plays for India" → Entity: Virat Kohli, Relation: plays for.
●​ Text Classification – Categorizing text into classes.​
Example: "This phone is awesome!" → Positive sentiment.
 ●​ Language Generation – Creating text automatically.​
Example: Google Translate converts English → Hindi.
●​ Speech Processing – Converting between speech and text.​
Example: Saying "Hey Siri" → system converts to text & acts.
●​ Question Answering – System answers user queries.​
Example: Q: "Who is the CEO of Tesla?" → A: "Elon Musk".
●​ Dialogue Systems – Conversational systems like chatbots.​
Example: Swiggy chatbot helps track your order.
●​ Sentiment & Emotion Analysis – Detecting emotions in text.​
Example: Tweet: "I lost my phone " → Emotion: Sadness.

e
Main Steps in NLP
1.​ Text Input – Collect text or speech from humans.

iv
2.​ Tokenization – Break sentences into words. (e.g., "I like apples" → ["I", "like", "apples"])
3.​ Stop-word Removal – Remove common words like "is", "the", "and".

ad
4.​ Stemming / Lemmatization – Reduce words to their root form. (e.g., "running" → "run").
5.​ Part of Speech Tagging – Identify nouns, verbs, adjectives.
6.​ Parsing / Syntax Analysis – Understand sentence structure.
7.​ Semantic Analysis – Understand meaning/context.
8.​ Output – Give results like translation, chatbot reply, or summary.
an
VII)Advanced analytics of cognitive computing
Advanced analytics is the use of smart techniques like machine learning, statistics, AI, and
data mining to find hidden patterns, predict outcomes, and guide better decisions.
It is useful for handling big, complex, and fast data.
rR

Some of the specific statistical techniques used in advanced analytics include decision tree
analysis, linear and logistic regression analysis, social network analysis, and time series analysis.
1. Descriptive Analytics

●​ It is the first stage of analytics that deals with summarizing and interpreting past data.
i

●​ It helps organizations understand what has already happened in their business or process.
ih

●​ Uses techniques such as data aggregation, data visualization, and reporting.

●​ Example: A company checking last month’s sales reports or a hospital analyzing patient
admission statistics.
M

2. Diagnostic Analytics

●​ Goes a step further than descriptive analytics by asking “Why did it happen?”
●​ Involves identifying causes, relationships, and dependencies in the data.
●​ Often uses statistical analysis, data mining, and correlation studies.
●​ Useful for root cause analysis in case of failures or unexpected results.
●​ Example: If sales dropped, diagnostic analytics helps find out whether it was due to
higher prices, low marketing, or competition.
3. Predictive Analytics

●​ Focuses on forecasting future outcomes based on past data and trends.

●​ Relies heavily on machine learning, statistical models, and AI techniques.
●​ Answers the question “What is likely to happen?”
●​ Often uses regression analysis, classification models, and time-series forecasting.
●​ Example: Banks use predictive analytics for credit scoring, e-commerce platforms use it
for product recommendations.

4. Prescriptive Analytics

e
●​ The most advanced stage of analytics.

iv
●​ Goes beyond predicting outcomes by suggesting the best possible action.
●​ Answers “What should we do?” and provides recommendations.
●​ Uses optimization techniques, simulations, reinforcement learning, and decision theory.

ad
●​ Example: Uber or Ola uses prescriptive analytics to suggest optimal driver routes; airlines
use it for dynamic ticket pricing.
an
i rR
ih
M