Boltzmann machines are a type of stochastic recurrent neural network invented in 1983 by Geoffrey Hinton and Terry Sejnowski, and later popularized in cognitive sciences by Hinton, Sejnowski, and Yann LeCun. Inspired by their construction in statistical physics, Boltzmann machines aim to model the distribution of data in a way similar to the way particles in a physical system reach thermal equilibrium.

These machines consist of units or neurons that are either visible, representing the observed data, or hidden, which help to capture complex dependencies among visible units. The connections between these units are bidirectional and can be weighted, allowing for the representation of complex interactions.

Boltzmann machines use a stochastic process for training, typically employing a method called Gibbs sampling. This involves repeatedly adjusting the states of the neurons to minimize the difference between the observed data distribution and the model distribution. The process is computationally intensive and can be slow, which has limited the practical usability of Boltzmann machines, particularly for large-scale applications.

Despite these limitations, Boltzmann machines remain useful under certain conditions. They have been employed effectively in unsupervised learning tasks and for learning complex, multimodal data distributions. Restricted Boltzmann Machines (RBMs), a simplified and more efficient version of Boltzmann machines, have found applications in various domains, including collaborative filtering (e.g., recommendation systems), dimensionality reduction, and feature learning.

Moreover, Boltzmann machines have influenced the development of more advanced neural network architectures and learning algorithms. Their principles underpin deep learning methods and have contributed to the advancement of artificial intelligence research. Recent advancements in computational power and optimization techniques have rekindled interest in Boltzmann machines, potentially leading to new applications and insights in machine learning and cognitive science.