Loading Deep Networks Is Hard

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Deep learning in neural networks: An overview (2014, background): “Compare earlier terminology on loading deep networks (Sı́ma, 1994; Windisch, 2005) and learning deep memories (Gomez and Schmidhuber, 2005).”
Constructive algorithms for structure learning in feedforward neural networks for regression problems (1997, background): “Judd [11] and others (such as [12]–[14]) showed that the loading problem is in general NP-complete.”
Training a Single Sigmoidal Neuron Is Hard (2002, background): “Eventually, Judd (1988) provided a polynomialtime loading algorithm for restricted shallow architectures whose practical applicability was probably ruled out by the hardness result for loading deep triangular networks (Šı́ma, 1994), which was further extended to pyramidal networks (de Souto & de Oli”
Many-Layered Learning (2002, methodology): “For networks of a fixed architecture, trained by global error minimization, theoretical results have shown that loading data into such a network is an NP-complete problem, independent of the training algorithm and regardless of whether the network is shallow or deep (Judd, 1990; Blum & Rivest, 1988;”
Loading Deep Networks Is Hard: The Pyramidal Case (2005, methodology): “Šı́ma (2002) gives a more detailed survey of these and other results.”