Fixed-weight on-line learning

How researchers see this paper

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

Meta-SGD: Learning to Learn Quickly for Few Shot Learning (2017, background): “Early studies show that recurrent neural networks (RNN) can model adaptive optimization algorithms [5, 26].”
Meta Networks (2017, neutral): “By following the success of the previous work (Mitchell et al., 1993; Younger et al., 1999; Andrychowicz et al., 2016; Ravi & Larochell, 2017), we studied the meta information in the form of the loss gradient of neural nets.”
Learning to reinforcement learn (2016, methodology): “Indeed, after an initial training period, the network can improve its performance on new tasks even if the weights are held constant (see also Cotter and Conwell, 1990; Prokhorov et al., 2002; Younger et al., 1999).”
Adaptive behavior with fixed weights in RNN: an overview (2002, background): “Sometimes none of the mappings have states , as in [ 3 ], [5] and [6].”
Learning to Learn Using Gradient Descent (2001, background): “The capability of recurrent networks to execute the subordinate system was proved and demonstrated in [3,19].”
Simple and conditioned adaptive behavior from Kalman filter trained recurrent networks (2003, background): “Younger, Conwell, and Cotter (1999) and Hochreiter, Younger, and Conwell (2001) have demonstrated related abilities for specially constructed recurrent networks and have used the term meta-learning for what we call adaptive behavior.”