IAI | HSLU | Jana Köhler

If several actions are possible in a state, this agent can evaluate their utility and make a deliberate choice

This agent can acquire new skills and reflect on its own performance to improve over time

Learning Agents can become more competent over time through learning and adaptability.

Yes, Learning Agents can operate in initially unknown environments and start with an empty knowledge base.

The components include:

1. Performance element (current learned model)
2. Learning element (improves performance through learning)
3. Critic (evaluates behavior and provides feedback)
4. Problem generator (suggests actions for informative experiences).

• Builds a model of the world and uses an explicit representation of goals
• Considers effects of actions on the world model before selecting an action to achieve a goal state - to choose among competing actions/goals utility function is needed

The goal-based agent selects actions based on their effects towards a goal in the future and can therefore also select actions that temporally lead to a state with worse utility

Agent architectures that we distinguish include Simple Reflex Agent, Model-based Reflex Agent, Goal-based Agent, Utility-based Agent, and Learning Agent.

The model-based reflex agent is more intelligent because it maintains an internal world model, allowing it to consider past percepts and anticipate action effects, which enables more informed and flexible decision-making.

The key difference is in their decision criteria. A utility-based agent maximizes expected utility, considering the desirability of outcomes, while a goal-based agent focuses on achieving specific goals or objectives.

The critic evaluates the behavior of the learning agent based on its performance and provides feedback to the learning element, assisting in the agent's learning and improvement process.

A problem generator suggests actions that lead to informative experiences, helping the learning agent explore and gather valuable data to improve its knowledge and decision-making.

To achieve human-level AI, we need sophisticated agent architectures that are capable of learning, adapting, and making intelligent decisions. A combination of learning agents and other advanced architectures may be necessary.

Complex Computer Applications:
Bezieht sich auf komplexe Anwendungen von Computertechnologie, die verschiedene Aufgaben und Funktionen in verschiedenen Bereichen unterstützen.
Enthält Anwendungen wie Datenbankmanagement, Finanzmodellierung und Simulationen.

Artificial Intelligence (Künstliche Intelligenz):
Eine spezialisierte Technologie im Bereich der Informatik, die Maschinen die Fähigkeit gibt, menschenähnliche Intelligenz auf Aufgaben wie Lernen, Entscheidungsfindung und Problembehandlung anzuwenden.
Enthält Techniken wie maschinelles Lernen, neuronale Netzwerke und natürliche Sprachverarbeitung.

Find a definition
- Describe a concept by listing explicitly its essential properties
• Challenge: what is essential?
- Apply automated reasoning procedures to infer that an object meets the
concept definition by verifying the explicitly stated properties

*Use many different (arbitrary) features to describe the concept
- Features can be given implicitly through pictures or unstructured text or explicitly as a list of feature-value pairs
- Show object examples to the system and let it learn a generalization pattern

 - Each example must be annotated if it belongs to the concept that the system must learn or not (positive and negative examples)
* If the generalized pattern is correct, the system has «learned» the concept without using an explicit definition

No feedback is given, the algorithm detects patterns in the sensory input data,

e.g., clustering and association algorithms

Algorithm processes example input–output pairs and learns a function that maps from input to output, e.g., decision trees and neural networks

Algorithm learns from a series of reinforcements (rewards or punishments)

returned from the environment when the agent executes actions, e.g., AlphaZero

•  Neuron fires when a linear combination of its inputs exceeds predefined threshold

• The bias weight (multiplied with input 1.0) is a constant to shift the result of the activation function towards the positive or negative side (offset the result)

Feed-forward networks are directed acyclic graphs, they have no internal state other than the weights

– Single-layer perceptron 

– Multi-layer perceptron

Consider a perceptron with a step function
– Can represent AND, OR, NOT, majority, etc., but not XOR
– Represents a linear separator in the input space
– Only a small fraction of all Boolean functions is linearly separable

• Input and output layers are usually fully connected via at least one hidden layer 

• Network architecture (number of hidden layers/edges typically chosen by hand)

  Description GPT:

• A Multi-Layer Perceptron (MLP) or Feed-forward Network is a widely used neural network model.
• It consists of at least three layers: an input layer, at least one hidden layer, and an output layer.
• These layers are typically fully connected, with each unit in one layer connected to every unit in the next layer.
• Data flows in one direction, from input to output, without feedback loops.

A Feed-Forward Network is a type of artificial neural network where data flows in one direction, from input to output, without feedback loops. To represent any Boolean function, at least one hidden layer is needed in the network.

A single, sufficiently large hidden layer in a Feed-Forward Network can accurately represent any continuous function of the inputs with arbitrary accuracy. It acts as a universal approximator for continuous functions.