Beyond Distribution Estimation: Simplex Anchored Structural Inference Towards Universal Semi-Supervised Learning

Yaxin Hou, Jun Ma, Hanyang Li, Bo Han, Jie Yu, Yuheng Jia, Beyond Distribution Estimation: Simplex Anchored Structural Inference Towards Universal Semi-Supervised Learning, International Conference on Machine Learning, 6th-11th July, Seoul, 2026.

This is an official PyTorch implementation for Beyond Distribution Estimation: Simplex Anchored Structural Inference Towards Universal Semi-Supervised Learning.

Introduction

This code is based on the public and widely-used codebase USB.

What I've done is just adding our SAGE algorithm in semilearn/imb_algorithms/sage.

Also, I've made corresponding modifications to semilearn/nets/ and several __init__.py.

How to run

For example, on CIFAR-10 with arbitrary unlabeled data($\gamma_u=100$)

CUDA_VISIBLE_DEVICES=0 python train.py --c "/config/config-2/sage/003-fixmatch_sage_cifar10_lb40_1_ulb4996_100_random_0.0_2.yaml"

(Note: I know that USB supports multi-GPUs, but I still recommend you to run on single GPU, as some weird problems may occur.)

The model will be automatically evaluated every 1024 iterations during training. After training, the last two lines in /saved_models/sage/003-fixmatch_sage_cifar10_lb40_1_ulb4996_100_random_0.0_2/log.txt will tell you the best accuracy.

For example,

[2025-12-21 07:10:45,980 INFO] model saved: ./saved_models/sage/003-fixmatch_sage_cifar10_lb40_1_ulb4996_100_random_0.0_2/latest_model.pth
[2025-12-21 07:10:45,990 INFO] Model result - eval/best_acc : 0.6683
[2025-12-21 07:10:45,990 INFO] Model result - eval/best_it : 133119

Results

The reported accuracies in Table 1, 2, and 3 in our paper are the average over three different runs (random seeds are 0/1/2).

Citation

If you find our method useful, please consider citing our paper:

@inproceedings{SAGE,
  author     = {Yaxin Hou and
                Jun Ma and
                Hanyang Li and
                Bo Han and
                Jie Yu and
                Yuheng Jia},
  title      = {Beyond Distribution Estimation: Simplex Anchored Structural Inference
                Towards Universal Semi-Supervised Learning},
  booktitle  = {International Conference on Machine Learning (ICML)},
  volume     = {},
  pages      = {},
  year       = {2026}
}