Multi-Task Learning with Additive U-Net for Image Denoising and Classification

2026-02-13 · Source: Takara TLDR - Daily AI Papers · Field: Technology & Digital — Artificial Intelligence & Machine Learning · Depth: Advanced, medium

Summary

Researchers Vikram Lakkavalli and Neelam Sinha introduce the Additive U-Net (AddUNet), an architectural modification to the standard U-Net that replaces concatenative skip connections with gated additive fusion. This change constrains shortcut capacity and maintains fixed feature dimensionality, acting as a structural regularizer to control encoder-decoder information flow and stabilize joint optimization in multi-task learning (MTL). Evaluated on single-task image denoising and joint denoising-classification, AddUNet achieves competitive reconstruction performance with enhanced training stability. In MTL scenarios, the learned skip weights dynamically redistribute, with shallow skips prioritizing reconstruction and deeper features supporting classification. The model demonstrates robust reconstruction even with limited classification capacity, suggesting implicit task decoupling through its additive fusion mechanism.

Key takeaway

For Computer Vision Engineers developing multi-task learning models, consider integrating additive skip fusion into your U-Net architectures. This approach, exemplified by AddUNet, can enhance training stability and improve performance in tasks like image denoising and classification by implicitly decoupling task information flow. You should experiment with gated additive fusion to achieve more robust and scalable multi-task learning without increasing model complexity.

Key insights

Additive skip fusion in U-Nets improves multi-task learning stability and performance by regularizing information flow.

Principles

• Constraining skip connection capacity stabilizes MTL.
• Additive fusion enables implicit task decoupling.
• Learned skip weights adapt to task requirements.

Method

AddUNet replaces concatenative U-Net skips with gated additive fusion, regularizing information flow and stabilizing joint optimization for image denoising and classification tasks.

In practice

• Implement gated additive fusion in U-Net architectures.
• Apply AddUNet for stable image denoising.
• Use AddUNet for joint denoising-classification tasks.

Topics

• Additive U-Net
• Multi-Task Learning
• Image Denoising
• Image Classification
• U-Net Architecture

Best for: Computer Vision Engineer, Research Scientist, AI Researcher, AI Scientist, Deep Learning Engineer

Related on AIssential

• Asymptotic Behavior of Multi--Task Learning: Implicit Regularization and Double Descent Effects — Multi-task learning improves generalization by implicitly adding regularization through shared task information.
• U-Net Explained: The CNN That Revolutionized Image Segmentation — U-Net's U-shaped encoder-decoder with skip connections enables precise pixel-level segmentation from limited data.
• SURGELLM: Rethinking Multi-Task Evaluation through Task-Aware Feature Gating with Class-Balanced Normalization — SURGELLM enhances multi-task NLP performance by integrating task-aware feature gating and class-balanced normalization within a unified transformer framework.
• Post-Hoc Merging is Not Enough: Many-Shot Model Merging with Loss-Gap Balancing — Iterative, loss-aware many-shot model merging effectively mitigates task interference and information erasure in multi-task LLMs.
• Multiple Additive Neural Networks for Structured and Unstructured Data — MANN enhances gradient boosting with shallow neural networks for diverse data types, improving accuracy and robustness.
• The Information-Theoretic Benefit of Shared Representations under Orthogonality Constraints — Joint approximation significantly reduces description-length in multi-task deep learning when tasks share a latent hard feature.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Takara TLDR - Daily AI Papers.