Learnijoy
ResearchAI ResearchAI Engineering & DevTools

Curvature-Aware Method Boosts Nearest Neighbor Classification

Alexandre L. M. Levada· August 31, 2026 View original

Key takeaways

  • Standard k-NN struggles with varying local data geometry.
  • CARSANN adapts neighborhood radii based on local manifold curvature.
  • Highly curved regions get smaller radii; flatter regions get larger.
  • It significantly improves balanced accuracy over traditional k-NN.

Who benefits

HealthcareFinanceE-commerceManufacturingCybersecurity

Summary

CARSANN (Curvature-Aware Radius Shrinkage for Adaptive Nearest Neighbor Classification) is a new framework that improves k-NN by adapting neighborhood spatial support based on local manifold curvature. It shrinks radii in highly curved regions and expands them in flatter areas, consistently outperforming standard k-NN and other adaptive methods on numerous datasets.

This research introduces CARSANN, a novel framework designed to enhance nearest neighbor classification by dynamically adjusting the spatial extent of neighborhoods based on the local geometry of the data manifold. Traditional k-NN methods assume a uniform neighborhood cardinality across the entire feature space, which can be suboptimal when data geometry varies significantly. CARSANN addresses this by first estimating the intrinsic dimensionality and constructing a local intrinsic representation. It then estimates local mean curvature using a shape-operator-based formulation. This curvature estimate directly controls the neighborhood scale: regions with high curvature experience stronger radius shrinkage, while flatter regions retain broader spatial support. This explicit adaptation of spatial evidence, rather than just the number of neighbors or the metric, leads to significant improvements. Experiments across over 70 real-world datasets demonstrate that CARSANN consistently outperforms standard k-NN and is competitive with other adaptive nearest-neighbor methods, showing statistically significant gains in balanced accuracy.

Why it matters

Professionals relying on nearest neighbor algorithms for classification, anomaly detection, or recommendation systems can achieve higher accuracy and more robust performance by incorporating geometric awareness into their models.

How to implement this in your domain

  1. 1Integrate CARSANN's curvature-aware radius shrinkage into existing nearest neighbor classification pipelines.
  2. 2Experiment with adaptive neighborhood definitions in machine learning models, especially for datasets with complex, varying local geometries.
  3. 3Develop tools to visualize and analyze the local manifold curvature of data to inform model design.
  4. 4Apply CARSANN to improve the performance of similarity-based algorithms in various applications.

Original post by Alexandre L. M. Levada

"arXiv:2608.27634v1 Announce Type: new Abstract: Nearest neighbor classification relies fundamentally on how locality is defined, yet conventional $k$-NN imposes the same neighborhood cardinality throughout the feature space. This assumption can be inadequate for data whose local…"

View on X

Originally posted by Alexandre L. M. Levada on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses

More in AI Research

AI ResearchAI Engineering & DevTools

New Optimizer Accelerates LLM Pretraining with Curvature-Conditioned Momentum

This research proposes a curvature-conditioned multiscale momentum method with sphere constraints to accelerate large language model pretraining. It addresses challenges from noise-dominant gradients and ill-conditioned loss landscapes by enhancing progress along flat directions, significantly improving upon existing adaptive optimizers like AdamW and Muon.

Shuchen Zhu, Yuxin Fang, Mingze Wang, Kun YuanAug 31, 2026
AI ResearchAI Engineering & DevTools

Euclidean Fourier Neural Operators Enhance Domain Transferability

This paper introduces Euclidean Fourier Neural Operators (EFNOs) as a domain-independent alternative to traditional FNOs, addressing their limitation in transferring across different periodic domains. EFNOs achieve this by parameterizing the spectral kernel as a continuous function of the physical wavevector, enabling consistent operator learning across varying domain shapes and sizes.

Nathanael Bosch, Niklas Frederik Schmitz, Michael F. HerbstAug 31, 2026
AI Engineering & DevToolsAI Research

SymboLLM-FE Boosts Feature Engineering with LLMs and Symbolic Regression

This paper introduces SymboLLM-FE, a novel approach combining symbolic regression and large language models for automated feature engineering on tabular data. It aims to generate highly interpretable and performant features while overcoming the limitations of traditional AutoFE and LLM-based methods.

Zi-Jian Cheng, Zi-Yi Jia, Zhi Zhou, Yu-Feng Li, Lan-Zhe GuoAug 31, 2026