NEERAJ BEDI
FOUNDER MADHAWK GAMES
Briefly about Imitation Learning
which we are using in Esports
project!
Imitation Learning Also known as
learning by Demonstration
In Imitation Learning 
A specialist demonstrates how
to solve the task
Imitation Learning has two
approaches->
1.Behavioural cloning is a method
by which human skills can be
captured and reproduced in a
computer program. As the
human subject performs the
skill, his or her actions are
recorded
The learning program outputs a
bunch of rules that reproduce
the skilled behaviour
2.Inverse Reinforcement Learning
is the cycle of developing a
reward function from the
observed behaviour of the agent,
it sounds like  Reinforcement
learning but this is a different
technique 
accomplish 
"Reinforcement learning"
consumes rewards and
punishments to learn behaviour
but in Inverse Reinforcement
Learning the direction is
inverted, for example, a robot
will perceive a person's
behaviour to figure out what
goal we are trying to accomplish
In Games, we are using Imitation
learning over reinforcement
learning when such a reward
function is difficult to specify
and also we are using Imitation
Learning, in esports we are using
imitation learning where we
want an agent to perform like
human 

Briefly About Imitation Learning

  • 1.
    NEERAJ BEDI FOUNDER MADHAWKGAMES Briefly about Imitation Learning which we are using in Esports project!
  • 2.
    Imitation Learning Alsoknown as learning by Demonstration In Imitation Learning  A specialist demonstrates how to solve the task
  • 3.
    Imitation Learning hastwo approaches->
  • 4.
    1.Behavioural cloning isa method by which human skills can be captured and reproduced in a computer program. As the human subject performs the skill, his or her actions are recorded The learning program outputs a bunch of rules that reproduce the skilled behaviour
  • 5.
    2.Inverse Reinforcement Learning isthe cycle of developing a reward function from the observed behaviour of the agent, it sounds like  Reinforcement learning but this is a different technique  accomplish 
  • 6.
    "Reinforcement learning" consumes rewardsand punishments to learn behaviour but in Inverse Reinforcement Learning the direction is inverted, for example, a robot will perceive a person's behaviour to figure out what goal we are trying to accomplish
  • 7.
    In Games, weare using Imitation learning over reinforcement learning when such a reward function is difficult to specify and also we are using Imitation Learning, in esports we are using imitation learning where we want an agent to perform like human