Please enable JavaScript.

Coggle requires JavaScript to display documents.

Uncertainty, Probability, and Markov Chains in AI, Eg: A spam filter…

- - - - Used when outcomes are not fully predictable.
        
        Basis for decisions when full knowledge of the world isn’t available
        
        Eg: A self-driving car unsure whether a pedestrian will cross or not must evaluate multiple possible actions
  - - - Planning under uncertainty means preparing for different outcomes, even if not all are likely
        
        Eg: An AI system managing warehouse robots must plan alternate paths if a shelf gets blocked
- - - - How much each outcome is preferred
        
        How likely each is to occur
- - - - Eg: An AI doctor evaluates symptoms like cough and fever; it assigns probabilities to causes like flu, COVID-19, or allergies
  - - - Incomplete knowledge, lack of data, and theoretical gaps make logic insufficient
- - - - Eg: A Netflix recommender assigns a probability to how likely you are to enjoy a movie based on past viewing
  - - - Ranges from 0 (impossible) to 1 (certain)
- - - - Eg: A delivery drone evaluates whether to fly in light rain. The utility of completing delivery is weighed against the risk of damage
  - - - Utility measures how useful or desirable an outcome is. It helps evaluate trade offs between different outcomes.
- - - - Each row represents a current state.
        Each column represents a possible next state.
        
        Eg: A chatbot predicting emotional tone might use a matrix of user sentiment transitions (happy → sad, angry → neutral, etc.)
  - - - They are memoryless. only the current state matters for predicting the next
        
        Eg: A robot navigating rooms in a house. The probability of moving to the kitchen depends only on its current room
- - - - Eg: A public transport planner simulates crowd movement using Markov transitions between stations
  - - - Eg: Predicting which app a user will open next on their phone
- - - - Explore possible outcomes and estimate likelihoods
        
        Eg: AI simulating thousands of chess games to predict the best move
  - - - Instead of calculating everything exactly, MCMC uses repeated simulations to approximate the probabilities
- - - - Common in pattern recognition
        
        Eg: In speech-to-text systems, actual spoken sounds (states) are hidden; observed audio signals guide interpretation
- - - - represent end conditions or final outcomes in a process
        
        Eg: In a job application process, "hired" or "rejected" can be absorbing states. once you reach them, the process ends