Thuật ngữ AI
Từ điển đầy đủ về Trí tuệ nhân tạo
Boltzmann Machine
Stochastic undirected neural network composed of interconnected binary neurons, used to model the probability distribution of input data and learn internal representations.
Restricted Boltzmann Machine
Simplified version of the Boltzmann Machine where connections only exist between the visible and hidden neuron layers, but not within each layer.
Free Energy
Mathematical function that quantifies the energy of a particular state in a Boltzmann Machine, used to calculate the probability of each network configuration.
Visible units
Neurons in a Boltzmann Machine that correspond directly to the observed variables in the input data and serve as an interface with the outside world.
Hidden units
Internal neurons of a Boltzmann Machine that capture complex statistical dependencies between the visible units and enable the learning of abstract representations.
Boltzmann distribution
Exponential probability distribution that describes the thermodynamic equilibrium state of a system, on which the mathematical formulation of Boltzmann Machines is based.
Deep Boltzmann Machine
Extended architecture of Boltzmann Machines with multiple hidden layers, allowing for the hierarchical learning of increasingly abstract representations.
Persistent Contrastive Divergence
Variant of Contrastive Divergence where the Markov chain is not reset between iterations, improving the quality of sampling and the convergence of training.
Thermodynamic state
Specific configuration of all neuron activations in a Boltzmann Machine, characterized by its energy level according to the network's energy function.
RBM temperature
Parameter controlling the level of stochasticity in neuron activation, influencing the exploration of state space and the network's ability to model data.
Synaptic weights
Parameters of a Boltzmann Machine quantifying the strength and sign of connections between neurons, optimized during training to minimize the energy of probable configurations.
Neuronal bias
Individual parameters added to each neuron in a Boltzmann Machine, modifying its activation probability independently of its connections with other neurons.
Stochastic generative model
Class of models of which Boltzmann Machines are part, capable of generating new data by sampling from the learned probability distribution.
Unsupervised learning
Training paradigm where Boltzmann Machines learn the inherent structure of data without labels, by maximizing the likelihood of training data.