Oldest Papers

Predictive coding (PC) offers a biologically plausible model of cortical functions, encompassing processes such as learning, prediction, encoding, and memory. However, predictive coding networks (PCNs) face significant challenges in stability and scalability, which constrain our capacity to elucidate cortical computation. Our study identifies instability in PCNs as a fundamental issue, focusing on the exponential growth of latent state norms and prediction errors after inference. These dynamics lead to exploding and vanishing gradients in PCNs. Moreover, the concentration of prediction errors near the input and output layer impedes effective learning, exacerbating performance degradation as network depth increases. To address these limitations, we propose stabilizing techniques for PCNs, including length regularization and sequential training with skip connection modules. This approach counteracts the exponential growth of latent states and makes the distribution of prediction errors more uniform across layers. Empirical evaluations demonstrate that our approach enhances stability and generalization, enabling the training of deeper networks more efficiently. This study deepens our understanding of complex dynamics in cortical networks, thereby advancing the practical application of predictive coding theory to its full potential.

No ratings yet

View paper →

PreviousPage 464 of 523Next

Network-based Active Inference for Adaptive and Cost-efficient Real-World Applications: A Benchmark Study of a Valve-turning Task Against Deep Reinforcement Learning | OpenReview

Harneet Singh, Jeong hwan Yoon, Pedro Fontana, Katherine Hanson, Whitney Sales, Varun Kamat, Karl Friston

openreview.net

This paper introduces Network-based Active Inference (NetAIF), a novel approach that integrates Active Inference (AIF) principles with network dynamics to enable adaptive, cost-efficient real-world applications. In benchmark tests against Deep Reinforcement Learning (DRL), NetAIF outperforms DRL in both computational efficiency and task performance. Leveraging random attractor dynamics, NetAIF generates real-time trajectories, allowing robots to adapt to complex, dynamic environments without the need for extensive pre-training. We demonstrate NetAIF's superiority in industrial valve manipulation, achieving over 99\% accuracy in goal position and orientation in untrained dynamic environments, with a 45,000-fold reduction in computational costs. NetAIF is approximately 100,000 times more efficient in iteration count than DRL, making it a highly robust and efficient solution for industrial applications.

No ratings yet

View paper →

Oldest Papers

The Continual Learning Problem

DeepSeekMath V2

Are ViTs As Global As We Think

Do Vision Transformers See Like Convolutional Neural Networks?

Are We on the Right Way for Evaluating Large Vision-Language Models?

Task-Driven Convolutional Recurrent Models of the Visual System

Crick1989

Climbing towards NLU: On Meaning, Form, and Understanding in the Age of Data - ACL Anthology

Towards Stable Learning in Predictive Coding Networks | OpenReview

Network-based Active Inference for Adaptive and Cost-efficient Real-World Applications: A Benchmark Study of a Valve-turning Task Against Deep Reinforcement Learning | OpenReview

Anthropic's Pilot Sabotage Risk Report

Neural Foundations of Mental Simulation: Future Prediction of Latent Representations on Dynamic Scenes