# Twin Neural Network Regression is a Semi-Supervised Regression Algorithm

arXiv: https://arxiv.org/abs/2106.06124 · HTML: https://mlsanity.com/paper/2106.06124/

## Key facts

- Cited by 6 papers in the map (data through 14 Oct 2026).
- Builds on 3 of its 25 references that are in the map.
- First posted on arXiv 11 Jun 2021.

## Builds on (most cited first, 3 of 3)

- [Learning with Pseudo-Ensembles](https://mlsanity.com/paper/1412.4864/) (2014)
- [Semi-supervised Deep Kernel Learning: Regression with Unlabeled Data by Minimizing Predictive Variance](https://mlsanity.com/paper/1805.10407/) (2018)
- [Twin Neural Network Regression](https://mlsanity.com/paper/2012.14873/) (2020)

## Built on by (most cited first, 6 of 6)

- [Semi-Supervised Deep Regression with Uncertainty Consistency and Variational Model Ensembling via Bayesian Neural Networks](https://mlsanity.com/paper/2302.07579/) (2023)
- [Twin Neural Network Improved k-Nearest Neighbor Regression](https://mlsanity.com/paper/2310.00664/) (2023)
- [Pairwise Difference Learning for Classification](https://mlsanity.com/paper/2406.20031/) (2024)
- [Bridged Clustering: Semi-Supervised Sparse Bridging](https://mlsanity.com/paper/2510.07182/) (2025)
- [Contrastive Learning for Semi-Supervised Deep Regression with Generalized Ordinal Rankings from Spectral Seriation](https://mlsanity.com/paper/2512.09267/) (2025)
- [Inverting Non-Injective Functions with Twin Neural Network Regression](https://mlsanity.com/paper/2601.05378/) (2026)

## Abstract (arXiv)

Twin neural network regression (TNNR) is a semi-supervised regression algorithm, it can be trained on unlabelled data points as long as other, labelled anchor data points, are present. TNNR is trained to predict differences between the target values of two different data points rather than the targets themselves. By ensembling predicted differences between the targets of an unseen data point and all training data points, it is possible to obtain a very accurate prediction for the original regression problem. Since any loop of predicted differences should sum to zero, loops can be supplied to the training data, even if the data points themselves within loops are unlabelled. Semi-supervised training improves TNNR performance, which is already state of the art, significantly.