SAGE

KO | EN

gitlite — search

#react #ai #go #machine-learning #rust #deep-learning #docker #nodejs #java #windows #vue #pytorch #openai #open-source

(★ 23)

This is the repository of the CVPR 2025 paper: "Mind the Gap: Confidence Discrepancy Can Guide Federated Semi-Supervised Learning Across Pseudo-Mismatch".

SAGE Latest Version Download

Download Latest Version (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# SAGE

<div id="toc">
	<ul align="center" style="list-style: none">
		<summary>
			<h2> Mind the Gap: Confidence Discrepancy Can Guide Federated Semi-Supervised Learning
            Across Pseudo-Mismatch </h2>
			<h3> —— CVPR 2025 —— </h3>
		</summary>
	</ul>
</div>

[![arXiv](https://img.shields.io/badge/arXiv-2503.13227-b31b1b.svg)](https://arxiv.org/abs/2503.13227)
[![cvpr](https://img.shields.io/badge/CVPR-HomePage-blue)](https://cvpr.thecvf.com/virtual/2025/poster/33062)

*Original Title: Federated Semi-Supervised Learning via Pseudo-Correction utilizing Confidence Discrepancy.*

![pipeline.png](https://raw.githubusercontent.com/Jay-Codeman/SAGE/master/pipeline.png)

</div>

### Abstract
Federated Semi-Supervised Learning (FSSL) aims to leverage unlabeled data across clients with limited labeled data to train a global model with strong generalization ability. Most FSSL methods rely on consistency regularization with pseudo-labels, converting predictions from local or global models into hard pseudo-labels as supervisory signals. However, we discover that the quality of pseudo-label is largely deteriorated by data heterogeneity, an intrinsic facet of federated learning. In this paper, we study the problem of FSSL in-depth and show that (1) heterogeneity exacerbates pseudo-label mismatches, further degrading model performance and convergence, and (2) local and global models' predictive tendencies diverge as heterogeneity increases. Motivated by these findings, we propose a simple and effective method called **S**emi-supervised **A**ggregation for **G**lobally-Enhanced **E**nsemble (SAGE), that can flexibly correct pseudo-labels based on confidence discrepancies. This strategy effectively mitigates performance degradation caused by incorrect pseudo-labels and enhances consensus between local and global models. Experimental results demonstrate that SAGE outperforms existing FSSL methods in both performance and convergence.

### Setup

1. Create a new Python environment:

```bash
   conda create --name sage python=3.8.18 
   conda activate sage
   ```

2. Install the required dependencies:

```bash
   pip install -r requirements.txt
   ```

### Dataset

Supported datasets:

* CIFAR-10
* CIFAR-100
* CINIC-10
* SVHN

Before running, please ensure the dataset paths are correctly set in `options.py`.

### Usage

Here is an example shell script to run SAGE on CIFAR-100 :

```bash
python SAGE.py --dataset='CIFAR100' --alpha=0.1 --gpu_id=0
```

Please replace `--dataset`, `--alpha`, and `--gpu_id` with appropriate values to customize the training configuration.

SAGE

SAGE Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

SAGE

SAGE Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

Download README