Understanding Predictive Coding as an Adaptive Trust-Region Method
Not on the map. Why?
Abstract (arXiv)
Predictive coding (PC) is a brain-inspired local learning algorithm that has recently been suggested to provide advantages over backpropagation (BP) in biologically relevant scenarios. While theoretical work has mainly focused on showing how PC can approximate BP in various limits, the putative benefits of "natural" PC are less understood. Here we develop a theory of PC as an adaptive trust-region (TR) algorithm that uses second-order information. We show that the learning dynamics of PC can be interpreted as interpolating between BP's loss gradient direction and a TR direction found by the PC inference dynamics. Our theory suggests that PC should escape saddle points faster than BP, a prediction which we prove in a shallow linear model and support with experiments on deeper networks. This work lays a foundation for understanding PC in deep and wide networks.
Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Builds on 21 of its 31 references that are in the map. First posted on arXiv 29 May 2023.
Builds on
21 of 31 references.
References · 21
- 2013Exact solutions to the nonlinear dynamics of learning in deep linear neural networkscited by 763
- 2014Identifying and attacking the saddle point problem in high-dimensional non-convex optimizationcited by 289
- 2017How to Escape Saddle Points Efficientlycited by 285
- 2022The Forward-Forward Algorithm: Some Preliminary Investigationscited by 266
- 2016Equilibrium Propagation: Bridging the Gap Between Energy-Based Models and Backpropagationcited by 259
- 2015Escaping From Saddle Points --- Online Stochastic Gradient for Tensor Decompositioncited by 243
- 2020Predictive Coding Approximates Backprop along Arbitrary Computation Graphscited by 90
- 2021Predictive Coding: a Theoretical and Experimental Reviewcited by 77
- 2017Gradient Descent Can Take Exponential Time to Escape Saddle Pointscited by 74
- 2022Error-driven Input Modulation: Solving the Credit Assignment Problem without a Backward Passcited by 63
- 2020A Theoretical Framework for Target Propagationcited by 57
- 2022Predictive Coding: Towards a Future of Deep Learning beyond Backpropagation?cited by 45
9 more
- 2018Biologically Motivated Algorithms for Propagating Local Target Representationscited by 43
- 2016The Power of Normalization: Faster Evasion of Saddle Pointscited by 29
- 2017Revisiting Normalized Gradient Descent: Fast Evasion of Saddle Pointscited by 29
- 2022Backpropagation at the Infinitesimal Inference Limit of Energy-Based Models: Unifying Predictive Coding, Equilibrium Propagation, and Contrastive Hebbian Learningcited by 26
- 2016Efficient approaches for escaping higher order saddle points in non-convex optimizationcited by 23
- 2021On the relationship between predictive coding and backpropagationcited by 20
- 2022A Theoretical Framework for Inference Learningcited by 17
- 2022A Theoretical Framework for Inference and Learning in Predictive Coding Networkscited by 17
- 2021Predictive Coding Can Do Exact Backpropagation on Convolutional and Recurrent Neural Networkscited by 14
Built on by
Citing papers per year: 2023: 1.
1 paper.