DMPCN: Dynamic Modulated Predictive Coding Network with Hybrid Feedback Representations

2025-04-20

AI Summary

Traditional predictive coding networks struggle to jointly model local details and global context, lack input-adaptive capacity, and employ no dedicated loss function optimized for prediction error. To address these limitations, we propose the Dynamic Modulation Predictive Coding Network (DM-PCN). DM-PCN introduces a hybrid error feedback mechanism that integrates local recurrent and global cross-layer feedback to enable fine-grained and coarse-grained representation co-adaptation. A complexity-aware dynamic modulation module adaptively adjusts network depth and computational paths conditioned on input characteristics. Furthermore, we design a customized prediction-error-weighted loss function to precisely suppress reconstruction deviations in salient regions. Evaluated on CIFAR-10/100, MNIST, and Fashion-MNIST, DM-PCN achieves 23–37% faster convergence and improves classification accuracy by 1.8–3.2 percentage points on average over state-of-the-art predictive coding models.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Computer Vision: Multi-modal Vision
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Traditional predictive coding networks, inspired by theories of brain function, consistently achieve promising results across various domains, extending their influence into the field of computer vision. However, the performance of the predictive coding networks is limited by their error feedback mechanism, which traditionally employs either local or global recurrent updates, leading to suboptimal performance in processing both local and broader details simultaneously. In addition, traditional predictive coding networks face difficulties in dynamically adjusting to the complexity and context of varying input data, which is crucial for achieving high levels of performance in diverse scenarios. Furthermore, there is a gap in the development and application of specific loss functions that could more effectively guide the model towards optimal performance. To deal with these issues, this paper introduces a hybrid prediction error feedback mechanism with dynamic modulation for deep predictive coding networks by effectively combining global contexts and local details while adjusting feedback based on input complexity. Additionally, we present a loss function tailored to this framework to improve accuracy by focusing on precise prediction error minimization. Experimental results demonstrate the superiority of our model over other approaches, showcasing faster convergence and higher predictive accuracy in CIFAR-10, CIFAR-100, MNIST, and FashionMNIST datasets.

Problem

• Improves error feedback mechanism in predictive coding networks
• Enhances dynamic adjustment to input data complexity
• Develops tailored loss function for better prediction accuracy

Innovation

• Hybrid feedback mechanism combining global and local details
• Dynamic modulation adjusting to input complexity
• Tailored loss function for precise error minimization