# Analyzing and Improving Representations with the Soft Nearest Neighbor Loss

arXiv: https://arxiv.org/abs/1902.01889 · HTML: https://mlsanity.com/paper/1902.01889/

## Key facts

- Cited by 95 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 5 Feb 2019.

## Built on by (most cited first, 30 of 95)

- [With a Little Help from My Friends: Nearest-Neighbor Contrastive Learning of Visual Representations](https://mlsanity.com/paper/2104.14548/) (2021)
- [Dissecting Supervised Contrastive Learning](https://mlsanity.com/paper/2102.08817/) (2021)
- [Contrastive Boundary Learning for Point Cloud Segmentation](https://mlsanity.com/paper/2203.05272/) (2022)
- [Neural Collapse with Normalized Features: A Geometric Analysis over the Riemannian Manifold](https://mlsanity.com/paper/2209.09211/) (2022)
- [Mitigating Data Sparsity for Short Text Topic Modeling by Topic-Semantic Contrastive Learning](https://mlsanity.com/paper/2211.12878/) (2022)
- [Revisiting Nearest Neighbor for Tabular Data: A Deep Tabular Baseline Two Decades Later](https://mlsanity.com/paper/2407.03257/) (2024)
- [Beyond Just Vision: A Review on Self-Supervised Representation Learning on Multimodal and Temporal Data](https://mlsanity.com/paper/2206.02353/) (2022)
- [It Takes Two: Your GRPO Is Secretly DPO](https://mlsanity.com/paper/2510.00977/) (2025)
- [Adapting to Distribution Shift by Visual Domain Prompt Generation](https://mlsanity.com/paper/2405.02797/) (2024)
- [On the Difficulty of Defending Self-Supervised Learning against Model Extraction](https://mlsanity.com/paper/2205.07890/) (2022)
- [ReMeDi: Resources for Multi-domain, Multi-service, Medical Dialogues](https://mlsanity.com/paper/2109.00430/) (2021)
- [Separating the "Chirp" from the "Chat": Self-supervised Visual Grounding of Sound and Language](https://mlsanity.com/paper/2406.05629/) (2024)
- [A Closer Look at Few-Shot Video Classification: A New Baseline and Benchmark](https://mlsanity.com/paper/2110.12358/) (2021)
- [Learning Debiased and Disentangled Representations for Semantic Segmentation](https://mlsanity.com/paper/2111.00531/) (2021)
- [Modeling Dynamic Topics in Chain-Free Fashion by Evolution-Tracking Contrastive Learning and Unassociated Word Exclusion](https://mlsanity.com/paper/2405.17957/) (2024)
- [Neighbor GRPO: Contrastive ODE Policy Optimization Aligns Flow Models](https://mlsanity.com/paper/2511.16955/) (2025)
- [Quality-aware Pre-trained Models for Blind Image Quality Assessment](https://mlsanity.com/paper/2303.00521/) (2023)
- [Contrastive Learning for Enhancing Robust Scene Transfer in Vision-based Agile Flight](https://mlsanity.com/paper/2309.09865/) (2023)
- [Toward Fairness Through Fair Multi-Exit Framework for Dermatological Disease Diagnosis](https://mlsanity.com/paper/2306.14518/) (2023)
- [Uncovering Intermediate Variables in Transformers using Circuit Probing](https://mlsanity.com/paper/2311.04354/) (2023)
- [Symbol as Points: Panoptic Symbol Spotting via Point-based Representation](https://mlsanity.com/paper/2401.10556/) (2024)
- [Syntriever: How to Train Your Retriever with Synthetic Data from LLMs](https://mlsanity.com/paper/2502.03824/) (2025)
- [Achieve Fairness without Demographics for Dermatological Disease Diagnosis](https://mlsanity.com/paper/2401.08066/) (2024)
- [Fool Me Once: Robust Selective Segmentation via Out-of-Distribution Detection with Contrastive Learning](https://mlsanity.com/paper/2103.00869/) (2021)
- [Leveraging Multi-lingual Positive Instances in Contrastive Learning to Improve Sentence Embedding](https://mlsanity.com/paper/2309.08929/) (2023)
- [Achieving Fairness Through Channel Pruning for Dermatological Disease Diagnosis](https://mlsanity.com/paper/2405.08681/) (2024)
- [Logit Normalization for Long-tail Object Detection](https://mlsanity.com/paper/2203.17020/) (2022)
- [Identity Encoder for Personalized Diffusion](https://mlsanity.com/paper/2304.07429/) (2023)
- [Contrastive Initial State Buffer for Reinforcement Learning](https://mlsanity.com/paper/2309.09752/) (2023)
- [Bucks for Buckets (B4B): Active Defenses Against Stealing Encoders](https://mlsanity.com/paper/2310.08571/) (2023)

## Abstract (arXiv)

We explore and expand the textitSoft Nearest Neighbor Loss to measure the textitentanglement of class manifolds in representation space: i.e., how close pairs of points from the same class are relative to pairs of points from different classes. We demonstrate several use cases of the loss. As an analytical tool, it provides insights into the evolution of class similarity structures during learning. Surprisingly, we find that textitmaximizing the entanglement of representations of different classes in the hidden layers is beneficial for discrimination in the final layer, possibly because it encourages representations to identify class-independent similarity structures. Maximizing the soft nearest neighbor loss in the hidden layers leads not only to improved generalization but also to better-calibrated estimates of uncertainty on outlier data. Data that is not from the training distribution can be recognized by observing that in the hidden layers, it has fewer than the normal number of neighbors from the predicted class.