Introduction Flashcards
Relationship between AI, ML, Deep Learning
Deep learning ⊂ ML ⊂ AI
Flavour of AI
Symbolic AI
Neural AI
Statistical AI
Paradigm of AI
Modeling
Inference
Learning
Modeling
Modeling means getting the mathematical way of thinking for the real world problem
Real world → Model
Inference
Once the model is ready, we can ask the questions
Model → Predictions
Learning
Model without parameters + Data → Model with parameters
Make the low level intelligence to high intelligence diagram
- Reflex based model
-
State based model
1. Search Problem
2. Marcov decision process(MDP)
3. Adverserial Games -
Variable Based model
1. Constraint satisfaction problems
2. Marcov networks
3. Bayesian networks - Logic based models
Reflex based model
No uncertainty
The most common models in ML.
No back feed and no reasoning, fully feed forward
Reflex-based models are a type of reactive AI system that operates on a simple set of rules or conditions. These models directly map inputs to actions without any consideration of past experiences or future implications.
Characteristics of Reflex-Based Models
1.No Memory
2.Rule-Based
3.Real-Time Processing
4.Simple Architecture
State based models
Uncertainty
Sequence of actions matter
Ex: Chess game
State-based models represent systems where behavior depends on the current state and transitions between states occur based on specific inputs or conditions. These models keep track of state information to determine actions, making them more dynamic and adaptable than reflex-based models.
Characteristics of State-Based Models
1.Memory of State : These models maintain a notion of “state” to represent the system’s current context or status.
2.State Transitions :Decisions and actions are based on transitions between states triggered by inputs or events.
3.Finite or Infinite State Space: The set of possible states can be finite (e.g., a simple system) or infinite (e.g., complex systems like language models).
4.Dynamic Behavior : Adapt to changes in input or the environment by transitioning to different states.
Types of state based model
1.Search Problems: (No uncertainty) You control everything
2.Markov Decision process (MDP): Uncertainty by nature. Ex: dice throw
3.Adverserial games: Uncertainty by opponent. Ex: chess
Variable based models
Sequence of actions doesn’t matter
Ex: Sudoku
Variable-based models represent systems using variables that define different aspects or features of the problem. These models focus on relationships between variables and how their values change in response to inputs or conditions. Such models are foundational in statistical and probabilistic reasoning.
Characteristics of Variable-Based Models
1.Variables as Core Elements: Variables represent the properties or features of the system (e.g., temperature, position, or probabilities).
2.Interdependencies: Relationships between variables are defined explicitly, often using equations, probabilities, or constraints.
3.Dynamic or Static: Models can be static (values fixed during a process) or dynamic (values change over time).
4.Probabilistic or Deterministic:
Probabilistic: Models uncertainty (e.g., Bayesian Networks).
Deterministic: Uses fixed equations for outcomes (e.g., physical system modeling).
Types of variable based models
1.Constraint satisfaction problems: Hard constraints. Ex:Sudoku
2.Marcov networks
3.Bayesian networks: Soft dependencies. Ex: Tracking of car with censor
Difference between Reflex, state, variable and logic based model
state, cost, action, successor
