0% found this document useful (0 votes)
20 views49 pages

5 Unit in AI

Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
20 views49 pages

5 Unit in AI

Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 49

• Chapter-5 (APPLICATIONS): Al applications - Language Models -

Information Retrieval - Information Extraction - Natural Language


Processing - Machine Translation - Speech Recognition - Robot -
Hardware – Perception - Planning – Moving.
Al applications
• Gaming: In India, AI is being utilized to develop sophisticated mobile and online games, with companies
like MPL (Mobile Premier League) using AI for enhancing user engagement and improving gaming
strategies.
• Natural Language Processing (NLP): Startups like Haptik and Vernacular.ai are harnessing NLP to create
chatbots and virtual assistants that understand and respond in multiple Indian languages, improving
customer service experiences.
• Expert Systems: Agriculture is a major sector in India, and AI-based expert systems are being developed
to help farmers with crop selection, pest control, and disease diagnosis, improving yields and
sustainability.
• Vision Systems: India's traffic and public safety departments are using AI-driven vision systems for
surveillance, crowd management, and traffic control, particularly in smart city projects.
• Speech Recognition: Organizations like the Indian government's e-governance services are
incorporating speech recognition to help citizens interact with services in a variety of Indian languages
and dialects.
• Handwriting Recognition: Banks and financial institutions in India are employing handwriting
recognition technologies to process cheques and forms faster and more accurately.
• Intelligent Robots: He al th c are i n Ind ia isse e ing t hei nt ro du c tio no f AI ro b o ts t hatassist in surgeries and
ht tp ://w w w.kn ow led g eg ate. in / g ate
patient care, while industries are using robots for tasks like assembly and warehouse management.
Language Models
• State Machines:
• Think of these like a game of stepping stones where each stone is a state. You
can move from one to another (transition) based on certain rules (inputs).
Examples include:
• Finite-State Automata (FSA): These are simple state machines that follow
strict rules to move from state to state.
• Finite-State Transducers (FST): These are like FSAs but can produce
outputs, not just state transitions.
• Rule Systems:
• These are sets of "if-then" rules. In linguistics:
• Regular Grammars: Used for simple structures like validating email
formats.
• Context-Free Grammars: Good for describing the structure of sentences
where context isn't needed.
• Feature-Augmented Grammars: These add extra information to
grammatical rules, like gender or number in languages.
• Logic:
• Logic models are like precise language for computers. They help computers
understand meanings and implications. Types include:
• First-Order Logic: Deals with objects and their relationships.
• Lambda Calculus: A way to express computations, used in functional
programming languages and semantics.
• Feature Structures: Like filling out a form, each feature (like 'color') has a
value (like 'red').
• Semantic Primitives: These are basic concepts used to build more complex
meanings.
• Probabilistic Models:
• These models are like weather forecasts for language, assigning probabilities
to different linguistic events. Examples are:
• Weighted Automata: State machines where each transition has a weight,
like a cost or likelihood.
• Markov Models: Predicts a sequence of events, where each event only
depends on the previous one.
• Hidden Markov Models (HMMs): Like Markov models, but the states are
hidden and only the outcomes are seen.
• Vector-Space Models:
• Imagine a giant space where each word is a point, and similar words are
closer together. This is useful for tasks like finding synonyms or grouping
similar words.
Information Retrievel

• Certainly, information retrieval (IR) is a vital component across various domains in


computer science, particularly in areas where large datasets are involved. Here's a
simplified explanation of your points:

• Huge Data/Document Collection:


• Think of IR systems as a vast library. Just like in a physical library where books are
categorized and indexed for easy retrieval, IR systems must manage and sift through
vast amounts of digital information to find what the user is looking for.

• Format of Query with Standard Query Language:


• When you want to find information, you ask a question or type a search. In some
systems, especially databases, you use a specific language with a particular syntax,
like SQL (Structured Query Language), to tell the system exactly what you're looking
for.
• The Generated Result Model:
• This is about how the system finds and picks the documents that answer your
query. It’s like having a smart assistant in the library who understands your question
and knows exactly where the answers are in the countless shelves of books.

• Displaying Results Model:


• Once your assistant finds the books, they can show them to you in different ways.
Some systems just show you the titles or a brief summary; others give you more
detailed previews. And then there are those that might use fancy visual aids, like
maps or 3-D models, to present the information in an easy-to-understand manner.
Definitions and Scope of NLP:
• Natural Language Processing (NLP): This field studies the challenges in
processing and manipulating natural language, aiming to make computers
understand human language.
• Automatic Processing: NLP enables the automatic processing of human
languages, making it a subset of AI focused on human-computer
communication in natural languages like English.
Purpose and Need for NLP:
• Analyzing Text: NLP is used to analyze and represent natural texts at
various linguistic levels to achieve human-like language processing for
diverse tasks.
• Bridging Communication Gaps: It serves as a bridge between human
communication and machine understanding, facilitating more
intuitive interactions.
Key Areas in Language Generation:
• Content Determination: Deciding what information should be conveyed in
the response.
• Text Planning: Organizing the content into a coherent structure before
generating the actual text.
• Realization: Transforming the structured content into natural language
text.
Capabilities of a Full NLU System:
• Paraphrasing: Rewording the input text while retaining the original
meaning.
• Translation: Converting text into another language.
• Question Answering: Providing answers based on the content of the
text.
• Inference: Making logical deductions from the text.
Applications of Natural Language Processing
1. Machine Translation:
• Purpose: Helps translate texts like manuals, support content, or catalogs at
reduced costs, enabling access to multilingual information.
• Challenge: The main difficulty lies not in translating words but in understanding
and conveying the intended meanings of sentences.
2. Automatic Summarization:
• Need: Useful in managing information overload by condensing large
volumes of data into concise summaries.
• Applications: Provides quick overviews of long articles, such as news
stories or blog posts, and is useful in extracting sentiment and key
points from social media.
• 3. Sentiment Analysis:
• Goal: To detect and understand the sentiments expressed in text, even when
these emotions are not explicitly stated.
• Use Cases: Companies utilize this to gauge public opinion on products and
services, helping them to tailor marketing strategies and improve customer
experiences.
4. Text Classification:
• Functionality: Involves categorizing documents into predefined groups to
streamline information retrieval and organization.
• Examples: Common applications include email spam filtering and sorting
content into thematic folders.
5. Question Answering:
• Overview: Develops systems that can answer questions posed by humans in a
clear and accurate manner.
• Technologies: This includes interfaces like chatbots and virtual assistants,
which use NLP to provide timely and contextually relevant responses.
fundamental steps involved in Natural Language Processing (NLP)

1. Morphological Analysis
• Purpose: To break down words into their smallest meaningful units, known as
morphemes (e.g., "comes" breaks down into "come" + "s").
• Process: Words are analyzed to identify their root forms and affixes. This step
also involves distinguishing tokens, like punctuation, which are not part of any
words.
2. Syntactic Analysis
• Purpose: To arrange words into a grammatical structure that makes sense
according to the rules of the language.
• Process: This involves parsing sentences to identify grammatical structures,
determining the role of each word (such as subject, verb, object), and how
these words interrelate to form coherent sentences.
3. Semantic Analysis
• Purpose: To understand the meanings of individual words and how these
meanings combine in sentences to produce overall sentence meanings.
• Process: Semantic analysis takes the structures formed during syntactic analysis
and assigns meanings to them, ensuring the entire sentence makes sense
semantically.
4. Pragmatic Analysis
• Purpose: To interpret sentences based on their usage in specific contexts,
considering how meaning changes with different situations.
• Process: It goes beyond the literal meaning of sentences to consider factors like
the speaker’s intent, the listener's perception, and the context in which the
conversation occurs. This helps in determining what is meant from what is
simply said.
5. Discourse Integration
• Purpose: To consider the influence of preceding sentences on the interpretation
of the current sentence, and how the current sentence can affect subsequent
sentences.
• Process: This step looks at the text as a whole, rather than in isolated
sentences, ensuring that the interpretation of one sentence is consistent with
and influenced by others around it.
AI significantly enhances NLP by allowing machines to understand and interact with
human language in a natural and intuitive way. This involves several key capabilities and
approaches:
• Speech Recognition: AI enables machines to process spoken words, turning speech
into text, which is crucial for applications like voice-activated assistants.
• Natural Communication: AI programs communicate with humans using natural
language, making interactions with machines more user-friendly and accessible.
• Understanding Language: AI must understand the structure of language, including
vocabulary, syntax (arrangement of words), and semantics (meaning), to effectively
process human language.
Approaches to NLU (Natural Language Understanding)

• Keyword and Pattern Matching: Simple methods for identifying essential words
or patterns to interpret language.
• Syntactic and Semantic Analysis: More complex analyses that consider both
the structure and meaning of language to generate responses.
• Real-World Situation Matching: AI compares inputs to known real-world
scenarios to enhance understanding and response accuracy.
Machine Translation (MT)
• Refers to the automated process of translating text from one language (source
language) to another (target language) using computer algorithms.
• Definition: Machine translation automates the translation of text from one
language to another, converting sequences of symbols (words, phrases) from
the source language to equivalent sequences in the target language.
• Purpose: The primary goal is to enable communication across language barriers
without human translators, enhancing accessibility and efficiency in global
interactions.
Types of Machine Translation Systems
• Bilingual MT System:
• These systems focus on translating between two specific
languages. They are typically unidirectional, meaning they
translate from one language to another but not vice versa.
• Multilingual MT System:
• Capable of translating between multiple language pairs. These
systems can be either unidirectional or bidirectional, facilitating
translations across several languages.
Major Approaches to Machine Translation
• Direct MT Approach:
• Description: Direct translation from the source to the target language
without intermediate steps.
• Example: Translating "Buenos días" (Spanish) directly to "Good morning"
(English) without analyzing or restructuring the sentence's grammatical
components.
• Interlingua Approach:
• Description: Translation involves converting the source language into an
artificial intermediate language (Interlingua) and then into the target
language.
• Example: For translating "Bonjour" (French) to "Guten Morgen" (German),
the system first translates French into an Interlingua representation like
"Greeting(morning)" and then from Interlingua to the German equivalent.
• Transfer Approach:
• Description: Involves three steps: analysis of the source text, conversion to a
target-language representation, and generation of the target text.
• Example: To translate "Estoy cansado" (Spanish, meaning "I am tired") to
English, the approach would first analyze and understand the structure
("[subject] [verb] [adjective]"), map it to an equivalent English structure
("[subject] [verb] [adjective]"), and generate the target text ("I am tired").
• Empirical (Statistical or Neural) MT Approach:
• Description: Uses statistical models or neural networks trained on large
datasets of bilingual texts to predict translations.
• Example: Google Translate uses neural machine translation (NMT). When
translating "Je t'aime" from French to English, it analyzes vast amounts of
French-English text data to learn that the most probable translation of "Je
t'aime" is "I love you" based on statistical likelihood and context learned
from data.
Overview of Speech Recognition
• Speech recognition technology allows computers to recognize spoken language
and convert it into a digital format that machines can process. This capability can
transform spoken commands or dictations into text or other forms of actionable
data, depending on the application.
Commonly Used Algorithms for Speech Recognition:
• Hidden Markov Models (HMM):
• HMMs are statistical models that assume the probability of each state depends
only on the previous state. In speech recognition, they are used to model
sequences of speech units (like phonemes or words) and predict the most likely
sequence from the audio input.
• Neural Networks:
• Deep learning approaches use neural networks to model complex patterns in data.
These networks consist of layers of interconnected nodes that mimic the human
brain's structure and function. They process training data, learn to recognize
speech patterns, and improve accuracy over time through methods like gradient
descent.
Applications of Speech Recognition
• Automotive: Integration in vehicles for hands-free navigation, controlling media
playback, and other voice-activated features enhance driver safety.
• Technology: Virtual assistants (like Siri, Google Assistant, Alexa, Cortana) embedded in
smartphones and smart speakers help with tasks such as setting reminders, playing
music, and controlling smart home devices through voice commands.
• Healthcare: Doctors and nurses use speech-to-text applications for efficient
documentation of patient care, improving accuracy and saving time.
• Sales and Customer Service: Call centers use speech recognition to transcribe and
analyze calls, helping identify trends and common issues. Cognitive bots engage in
voice-based customer interactions, improving response times and service quality.
• Security: Voice authentication systems provide an additional layer of security, using
unique vocal characteristics to verify a user's identity.
Techniques for Speech Recognition
• Speech Analysis: This involves analyzing the physical and behavioral
characteristics of speech, such as pitch and tone, which can help differentiate
speakers and understand spoken commands.
• Feature Extraction: Key to improving performance, this technique involves
condensing speech into a set of meaningful data points that represent the
essential characteristics of the speech signal without the redundancy.
• Modeling: Techniques like speaker recognition and identification are used to
create models that can recognize and differentiate between different speakers
based on their unique vocal features.
Robots, robotics, and how they intersect with artificial intelligence (AI)
• What are Robots?
• Definition: Robots are artificial agents designed to perform tasks in the real
world. They can manipulate objects through actions like picking, moving,
modifying, or even destroying them to achieve specific goals, often in
repetitive roles to free up human labor.
• What is Robotics?
• Field Description: Robotics is an interdisciplinary branch that combines
electrical engineering, mechanical engineering, and computer science to
design, construct, and operate robots. It involves developing systems that
can perform complex actions autonomously or semi-autonomously.
Differences Between AI Programs and Robot Systems

• Environment: AI programs typically operate in simulated virtual environments,


while robots operate in the real, physical world.
• Inputs: AI programs often receive inputs as symbols and rules, whereas robots
deal with analog signals such as sounds and images.
• Hardware Needs: AI programs run on general-purpose computers, while robots
require specialized hardware like sensors and actuators to interact with their
environment.
Integration of Robots and AI
• Enhanced Capabilities: AI equips robots with skills like computer vision,
allowing them to navigate and interact dynamically with their environment. AI
also enables robots to learn from experiences and make autonomous decisions,
using various forms of machine learning.
Integration of Robots and AI
Components of a Robot
• Power Supply: Includes batteries, solar power, hydraulic, or
pneumatic sources.
• Actuators: Convert energy into mechanical motion.
• Sensors: Equip robots with the ability to gather real-time data about
their surroundings, crucial for tasks requiring environmental
interaction.
Robot Locomotion
• Legged Locomotion: Uses multiple legs, suitable for varied terrain but
power-intensive and complex due to the need for coordination and
balance.
• Wheeled Locomotion: Involves fewer motors, generally easier to
implement, and more stable and power-efficient compared to legged
options.
• Slip/Skid Locomotion: Similar to a tank, using tracks that allow
stability and maneuverability on challenging surfaces.
Mobile Robot Hardware Overview
• A mobile robot integrates hardware and computational components to operate
autonomously or semi-autonomously in various environments. Key hardware
subsystems of a mobile robot include:
• Motion: This subsystem allows the robot to change its position relative to its
environment. It includes mechanisms like wheels, legs, or tracks that enable
movement.
• Sensing: Sensing technologies enable a robot to gather information about its
own state and its surroundings. This can include a range of sensors such as
cameras, infrared sensors, ultrasonic sensors, and lidar.
• Reasoning: This involves processing sensor data to make decisions. The
reasoning subsystem often includes onboard computers that execute
algorithms for navigation, obstacle avoidance, and task execution.

• Communication: Robots often need to communicate with human operators or


other machines. This subsystem can include wireless communication
technologies like Wi-Fi, Bluetooth, or even satellite communications in more
advanced robots.
Categories of Motion of Mobile Robots
• Mobile robots are designed to operate in various environments, each requiring
specialized modes of locomotion:
• Terrestrial:
• Environment: These robots navigate ground surfaces, leveraging the stability
of solid support.
• Locomotion: Terrestrial robots predominantly use wheels for simplicity and
efficiency. Other forms include legged robots for walking or climbing, tracked
robots for navigating rough terrains, and robots that slither, adapting to
diverse ground conditions.
• Airborne:
• Environment: Airborne robots are crafted to fly, often emulating the
designs of conventional aircraft or avian species.
• Locomotion: This category includes robotic helicopters, fixed-wing
aircraft, and inventive forms like automated parachutes and
dirigibles. Challenges for these robots include maintaining position
and navigating through varying air conditions.
• Aquatic:
• Environment: Aquatic robots function in water, suitable for both
surface and underwater operations.
• Locomotion: They mainly utilize propellers or jets for movement,
essential for conducting tasks in locations that are otherwise
inaccessible to humans, such as deep-sea exploration or
maintaining underwater structures.
• Space:
• Environment: Operating in the extreme conditions of outer space,
including microgravity and intense radiation.
• Locomotion: Space robots may operate independently as free-fliers
or move across larger vehicles like space stations. Their movement
typically relies on thrusters suited for the vacuum of space.
Types of AI in Robotics
• Weak AI: Often used for simulated human thought processes in
controlled environments, such as virtual assistants (e.g., Siri, Alexa).
• Strong AI: Employs sophisticated decision-making capabilities without
human oversight, used in autonomous vehicles and advanced
robotics.
• Specialized AI: Tailored for specific tasks, commonly found in
industrial robots performing repetitive actions like painting or
assembly.
Applications of Speech Recognition in Robotics
• Used extensively in fields like automotive for voice-activated systems,
healthcare for documentation, and customer service for interactive bots and
security systems, enhancing both efficiency and safety.

You might also like