RowNet: A Memory Transformer for Tabular Regression
2026-06-03 • Machine Learning
Machine LearningArtificial Intelligence
AI summaryⓘ
The authors address the challenge of predicting real estate prices, which involves complex features and regional effects. They propose RowNet, a new neural network model that compares a property to similar ones in a memory bank to help make predictions. RowNet uses layers that first look at feature similarities, then refine comparisons using actual price information, and finally combines different prediction methods to improve accuracy. This approach aims to better mimic how real estate values are influenced by comparable properties.
real estate valuationregressionneural networkretrieval-based modelattention headsmixture-of-expertsfeature similaritygradient-boosted decision treestabular dataresidual correction
Authors
Askat Rakhymbekov, Gulshat Muhametjanova
Abstract
Real estate valuation is a structured regression problem in which prices are governed by heterogeneous feature types, sparse regional effects, nonlinear interactions, and the practical logic of comparable properties. Standard multilayer perceptrons treat each row as an isolated vector and must learn locality, scale sensitivity, and categorical matching from supervision alone. Gradient-boosted decision trees provide strong tabular baselines, but their feature-centric splitting mechanism does not explicitly model the retrieval of similar historical observations. This paper presents RowNet, a retrieval-based neural architecture for real estate price-per-square-meter prediction. RowNet represents a query property through pairwise similarity features against a memory bank of labeled properties. A first retrieval layer estimates a coarse target from feature-only similarities. A second layer augments the memory comparison with target-consistency features and uses multiple learned attention heads to retrieve complementary comparable sets. A final mixture-of-experts module combines learned gating, residual correction, entropy regularization, and head-diversity regularization to produce the prediction.