Self Identity Mapping

2025-09-17

AI Summary

Traditional deep learning regularization methods often rely on heuristic designs, limiting their generalizability and reliability. To address this, we propose Self-Identity Mapping (SIM), a data-intrinsic regularization framework that reconstructs inputs via invertible mapping to mitigate information loss during forward propagation and enhance representation robustness. SIM is a plug-and-play, task- and model-agnostic module; its lightweight variant, ρSIM, significantly reduces computational overhead while remaining orthogonal and composable with existing regularization techniques. Technically, SIM employs patch-wise feature sampling and projection-driven latent-space reconstruction, optimized via a backward reconstruction loss. Extensive experiments across six diverse tasks—image classification, few-shot learning, domain generalization, semantic segmentation, audio classification, and time-series anomaly detection—demonstrate consistent and significant improvements over strong baselines. SIM effectively strengthens semantic fidelity and gradient flow without architectural modifications.

Technology Category

• Machine Learning: Unsupervised & Self-Supervised Learning
• Search and Optimization: Sampling/Simulation-based Search
• Computer Vision: Representation Learning for Vision

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems
• Web Mining and Content Analysis: Web data generation and simulation

Abstract

Regularization is essential in deep learning to enhance generalization and mitigate overfitting. However, conventional techniques often rely on heuristics, making them less reliable or effective across diverse settings. We propose Self Identity Mapping (SIM), a simple yet effective, data-intrinsic regularization framework that leverages an inverse mapping mechanism to enhance representation learning. By reconstructing the input from its transformed output, SIM reduces information loss during forward propagation and facilitates smoother gradient flow. To address computational inefficiencies, We instantiate SIM as $ ρ ext{SIM} $ by incorporating patch-level feature sampling and projection-based method to reconstruct latent features, effectively lowering complexity. As a model-agnostic, task-agnostic regularizer, SIM can be seamlessly integrated as a plug-and-play module, making it applicable to different network architectures and tasks. We extensively evaluate $ρ ext{SIM}$ across three tasks: image classification, few-shot prompt learning, and domain generalization. Experimental results show consistent improvements over baseline methods, highlighting $ρ ext{SIM}$'s ability to enhance representation learning across various tasks. We also demonstrate that $ρ ext{SIM}$ is orthogonal to existing regularization methods, boosting their effectiveness. Moreover, our results confirm that $ρ ext{SIM}$ effectively preserves semantic information and enhances performance in dense-to-dense tasks, such as semantic segmentation and image translation, as well as in non-visual domains including audio classification and time series anomaly detection. The code is publicly available at https://github.com/XiudingCai/SIM-pytorch.

Problem

• Enhancing generalization in deep learning with data-intrinsic regularization
• Reducing information loss during forward propagation via inverse mapping
• Creating model-agnostic regularization applicable to diverse network architectures

Innovation

• Self Identity Mapping uses inverse mapping for regularization
• Patch-level sampling reduces computational complexity
• Model-agnostic plug-and-play module for various tasks