English
 
User Manual Privacy Policy Disclaimer Contact us
  Advanced SearchBrowse

Item

ITEM ACTIONSEXPORT

Released

Conference Paper

Bounded Rational Decision-Making in Feedforward Neural Networks

MPS-Authors
/persons/resource/persons192683

Leibfried,  F
Max Planck Institute for Biological Cybernetics, Max Planck Society;
Research Group Sensorimotor Learning and Decision-Making, Max Planck Institute for Biological Cybernetics, Max Planck Society;

/persons/resource/persons83827

Braun,  D
Max Planck Institute for Biological Cybernetics, Max Planck Society;
Research Group Sensorimotor Learning and Decision-Making, Max Planck Institute for Biological Cybernetics, Max Planck Society;
Research Group Sensorimotor Learning and Decision-making, Max Planck Institute for Intelligent Systems, Max Planck Society;

Locator

Link
(Any fulltext)

Fulltext (public)
There are no public fulltexts available
Supplementary Material (public)
There is no public supplementary material available
Citation

Leibfried, F., & Braun, D. (2016). Bounded Rational Decision-Making in Feedforward Neural Networks. In A. Ihler, & D. Janzing (Eds.), Uncertainty in Artificial Intelligence (pp. 407-416). Corvallis, OR, USA: AUAI Press.


Cite as: http://hdl.handle.net/21.11116/0000-0000-7A8E-8
Abstract
Bounded rational decision-makers transform sensory input into motor output under limited computational resources. Mathematically, such decision-makers can be modeled as information-theoretic channels with limited transmission rate. Here, we apply this formalism for the first time to multilayer feedforward neural networks. We derive synaptic weight update rules for two scenarios, where either each neuron is considered as a bounded rational decision-maker or the network as a whole. In the update rules, bounded rationality translates into information-theoretically motivated types of regularization in weight space. In experiments on the MNIST benchmark classification task for handwritten digits, we show that such information-theoretic regularization successfully prevents overfitting across different architectures and attains results that are competitive with other recent techniques like dropout, dropconnect and Bayes by backprop, for both ordinary and convolutional neural networks.