Scribd
Upload
21 views46 pages

Machine Learning - Merged

The document discusses various concepts in machine learning, including the Self-Organizing Map (SOM) algorithm, layers in Convolutional Neural Networks (CNN), and the Markov Decision Process (MDP). It explains MDP as a framework for decision-making in environments with randomness and control, detailing its components and applications. Additionally, it outlines models of evolution in AI, such as Darwinian, Lamarckian, and Baldwinian evolution, and their applications in optimization and learning.

Uploaded by

sharmashish921
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
21 views46 pages

Machine Learning - Merged

The document discusses various concepts in machine learning, including the Self-Organizing Map (SOM) algorithm, layers in Convolutional Neural Networks (CNN), and the Markov Decision Process (MDP). It explains MDP as a framework for decision-making in environments with randomness and control, detailing its components and applications. Additionally, it outlines models of evolution in AI, such as Darwinian, Lamarckian, and Baldwinian evolution, and their applications in optimization and learning.

Uploaded by

sharmashish921
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/ 46

Q. Describe SOM algorithm using an example.

Q. Explain pooling layer, Convolutional layer and


fully connected layer.
POOLING LAYER
CONVOLUTION LAYER

FULLY CONNECTED LAYER

Q. APPLICATIONS OF CNN
Q27. Discuss Markov Decision Process. [K3]
A Markov Decision Process (MDP) is a mathematical model used to describe environments
in reinforcement learning where outcomes are partly random and partly under the control
of a decision-maker (agent).

Key Components of MDP:

1. States (S):
The possible situations or configurations the agent can be in.
2. Actions (A):
The set of all possible actions the agent can take in each state.
3. Transition Probability (P):
Probability of moving from one state to another after taking an action.

P(s′∣s,a)=Probability of reaching state s′ from state s using action aP(s'|s,a) =


\text{Probability of reaching state } s' \text{ from state } s \text{ using action }
aP(s′∣s,a)=Probability of reaching state s′ from state s using action a

4. Reward Function (R):


Immediate reward received after transitioning from state sss to s′s's′ by taking action
aaa.
5. Policy (π):
A strategy or mapping from states to actions that defines the agent’s behavior.

Goal of MDP:

To find an optimal policy π∗\pi^*π∗ that maximizes the cumulative reward over time.

Applications:

 Reinforcement learning algorithms (e.g., Q-learning)


 Robotics (e.g., robot navigation)
 Game playing (e.g., chess, Go)
 Operations research (e.g., inventory control)
Q. Explain models of evolution. [K2]
In machine learning and artificial intelligence, models of evolution refer to computational
approaches inspired by biological evolution. These models are used to optimize solutions
by mimicking the process of natural selection and genetic inheritance.

They are a foundation for evolutionary algorithms, such as genetic algorithms and
evolutionary strategies.

🔑 Key Models of Evolution:

1. Darwinian Evolution (Genetic Algorithms)

 Based on natural selection and survival of the fittest.


 Individuals (solutions) evolve through operations like:
o Selection: choosing the fittest individuals.
o Crossover: combining genes from parents.
o Mutation: randomly altering genes.
 Widely used in optimization problems.

2. Lamarckian Evolution

 Assumes that traits acquired during an individual’s lifetime can be passed to


offspring.
 In AI, it means local learning (e.g., backpropagation) is used to improve an individual,
and the improved version is inherited.

3. Baldwinian Evolution

 Similar to Lamarckian, but only the ability to learn is inherited, not the learned traits.
 Encourages selection of individuals that can learn effectively, improving population
over time.
📚 Applications:

 Optimization in complex search spaces


 Feature selection in machine learning
 Neural network architecture search
 Game strategy evolution
 Robot control and pathfinding

You might also like