Work Experience and Education

• Assistant Professor (CSE), IIIT-D
• Research Scientist, GE Research Center, New York
• Internships at IBM Research, Bell Labs, Xerox Research, Stardog
• Software Engineer, CA Technologies, Hyderabad


• Ph.D from Wright State University, Dayton, OH, USA
• M.Tech from MNNIT, Allahabad
• B.Tech from JNTU, Hyderabad

• It is the study of the general principles of building intelligent agents.
• An agent is any device that can perceive its environment through sensors and react to it by taking action to achieve a stated goal.
• Mimicking human senses, i.e., sight, sound, touch, smell, taste, is a form of perception.
• Intelligent agents should be able to interpret and process other forms of input such as text, semi-structured, and structured data.
• Output of an intelligent agent could be in multiple forms such as movement (reaching a destination), decision taken, sound etc.

How do we reach The Indian School from IIIT-D?

• Multiple paths
• Which path is the best?
• Depends on multiple factors
• Distance, time, road conditions, time of the day
• Given an initial path and a final path, search algorithms find the best possible route

How can a machine engage with the world?

• Perception: Gather information about the environment and react to it
• Information is gathered with the help of sensors
• Temperature, pressure, movement, air quality, images etc.
• Machines analyze the sensor data and take actions (sometimes using actuators)

Perception

• Detecting the movement of a person using images from multiple different cameras (with different angles)
• Counting the number of tigers in India
• Lip reading: Transcribe the text for a video
• Automatic Speech Recognition
• Generating captions for images

How can a machine observe the world and improve?

• Learning: Machines can observe the world including its own experiences and improve its performance over time
• Types of learning based on the feedback
• Supervised learning
• Unsupervised learning
• Semi-supervised learning
• Reinforcement learning

Supervised Learning

• Agents learn from examples (labelled data)
• This is a cat
• This is a tiger
• Person holding an umbrella
• Dog is a mammal
• Is this a cat?
• I am 90% sure that this is a lion

Unsupervised Learning

• Learn patterns without any explicit feedback
• Clustering
• Taxi driver can learn traffic time and non-traffic time, without explicit feedback

Semi-supervised Learning

• Limited amount of labelled examples are given
• Labels might not be accurate
• Need to find patterns from unlabelled data
• Guess a person's age from their picture
• We can collect some limited labelled data
• People might lie about their age

Reinforcement Learning

• Agent learns from a series of reinforements - reward or punishment
• Eg: Tip to a barber
• A reward function exists. An action is taken if it leads to a reward
• Exploration vs Exploitation trade-off
• Picking the best possible ripe mango in 15 mins from a tall tree with many branches

Natural Language Processing

• Extracting knowledge expressed in the natural language (books, blogs, articles, tweets)
• Communicate with humans using natural language
• Agents should be able to parse sentences and make sense out of it
• Applications: Speech recognition (speech to text), machine translation (text to speech), information retrieval

Knowledge Representation and Reasoning (KRR)

• Techniques to capture knowledge about the world in a form that machines can understand
• A Car is a type of Vehicle
• Car has exactly four wheels
• Car has at least two doors and at most four doors
• Reasoning is the process of deriving new facts (knowledge) based on existing facts
• All birds fly
• Pigeon is a bird
• Can Pigeon fly?
• KRR is the field of AI that helps an agent to use what it knows (background knowledge) to decide what to do

Applications of KRR

• Unstructured data to structured data
• Search Engine Optimization: Schema.org
• Adding semantics to the data
• Data Integration
• Recommendations
• Drug recommendation based on symptoms
• Question Answering
• Who is the president of India who was also a scientist?

Structured Data

• <Delhi> <capitalOf> <India>
• RDF triple is called a RDF Statement
• Triple can be represented as a directed labelled graph

Knowledge Graph

Robotics

• Robots are physical agents that perform tasks in the physical world
• Robots are generally equipped with sensors, arms, wheels/legs, joints
• Social robots
• Robots interact with kids with autism spectrum disorder (ASD)
• Robots in industrial settings (assembly, welding, packaging and labelling)

Philosophical and Ethical Issues

• Autonomous vehicles
• Will we lose jobs?
• Will robots take over the world?