KI-Glossar
Das vollständige Wörterbuch der Künstlichen Intelligenz
Graph Neural Networks (GNN)
Deep learning architecture specialized in processing data structured as graphs, enabling the learning of node and edge representations to solve combinatorial optimization problems.
Pointer Networks
Neural network architecture using an attention mechanism to point to specific positions in an input sequence, particularly effective for combinatorial optimization problems with variable output sizes.
Neural Combinatorial Optimization
Paradigm using neural networks to approximate solutions to NP-hard problems, replacing traditional algorithms with learned models capable of generalizing to new instances.
Reinforcement Learning for Optimization
Approach where an agent learns a decision policy to build optimal solutions through successive interactions with the optimization problem environment.
Attention Mechanism in Optimization
Technique allowing neural networks to selectively focus on relevant parts of the input, significantly improving performance on sequencing and assignment problems.
Transformers for Combinatorial Optimization
Application of Transformer architectures based on attention to solve combinatorial optimization problems, particularly effective for routing and sequencing problems.
Graph Embeddings for Optimization
Dense vector representations of graph structures that capture essential topological and structural properties to facilitate decision-making in optimization problems.
Seq2Seq Models for Optimization
Encoder-decoder architectures transforming optimization problem instances into sequences of decisions, enabling a unified approach for various combinatorial problems.
Graph Convolutional Neural Networks
Extension of CNNs to graph structures, using convolution operations on node neighborhoods to extract hierarchical features relevant for optimization.
Policy Gradient Descent for Optimization
Reinforcement learning algorithm directly optimizing policy parameters to maximize expected reward in combinatorial optimization problems.
Autoencoders for Combinatorial Problems
Unsupervised neural networks learning compressed representations of optimization problem instances, facilitating the discovery of hidden structures and patterns.
Q-Learning for Discrete Optimization
Value-based reinforcement learning algorithm learning a Q function to guide action selection in discrete state spaces of combinatorial problems.
Neural Architecture for the Traveling Salesman Problem
Network structures specifically designed to capture TSP constraints and symmetries, using attention mechanisms to model dependencies between cities.
End-to-End Learning for Optimization
Paradigm where a single neural network learns to directly transform problem inputs into optimal solutions without explicit intermediate steps.
Neural Combinatorial Solvers
Hybrid systems combining neural networks to guide search with traditional algorithms, leveraging the strengths of both approaches to improve efficiency.
Neural Monte Carlo Tree Search Methods
Combination of MCTS with neural networks to evaluate states and guide exploration, particularly effective for optimization problems with large search spaces.
Recurrent Neural Networks for Optimization
RNN architectures adapted to iteratively build solutions to combinatorial problems, maintaining a hidden state representing the progression of the construction.
Neural Particle Swarm Optimization
Hybridization between swarm optimization metaheuristics and neural networks to improve exploration and exploitation in complex solution spaces.