mlsanityData to 14 Oct 2026

Stolen Probability: A Structural Weakness of Neural Language Models

arXiv 2005.02433 ↗First posted 5 May 2020cs.LG stat.MLcited by 22

Not on the map. Why?

Abstract (arXiv)

Neural Network Language Models (NNLMs) generate probability distributions by applying a softmax function to a distance metric formed by taking the dot product of a prediction vector with all word vectors in a high-dimensional embedding space. The dot-product distance metric forms part of the inductive bias of NNLMs. Although NNLMs optimize well with this inductive bias, we show that this results in a sub-optimal ordering of the embedding space that structurally impoverishes some words at the expense of others when assigning probability. We present numerical, theoretical and empirical analyses showing that words on the interior of the convex hull in the embedding space have their probability bounded by the probabilities of the words on the hull.

Key facts. Cited by 22 papers in the map (data through 14 Oct 2026). First posted on arXiv 5 May 2020.

Builds on

No reference list yet.

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2021: 2; 2022: 2; 2023: 9; 2024: 6; 2025: 1; 2026: 2.

22 papers.

Sort:

By area:

  1. 2023Enhancing Uncertainty-Based Hallucination Detection with Stronger Focuscited by 99
  2. 2024Logits of API-Protected LLMs Leak Proprietary Informationcited by 40
  3. 2023On the Expressivity Role of LayerNorm in Transformers' Attentioncited by 33
  4. 2023Closing the Curious Case of Neural Text Degenerationcited by 32
  5. 2023Why do Nearest Neighbor Language Models Work?cited by 29
  6. 2023CodeCoT: Tackling Code Syntax Errors in CoT Reasoning for Code Generationcited by 29
  7. 2021Rare Tokens Degenerate All Tokens: Improving Neural Text Generation via Adaptive Gradient Gating for Rare Token Embeddingscited by 27
  8. 2024Prompting a Pretrained Transformer Can Be a Universal Approximatorcited by 25
  9. 2023Task-Adaptive Tokenization: Enhancing Long-Form Text Generation Efficacy in Mental Health and Beyondcited by 23
  10. 2022Low-Rank Softmax Can Have Unargmaxable Classes in Theory but Rarely in Practicecited by 15
  11. 2023Taming the Sigmoid Bottleneck: Provably Argmaxable Sparse Multi-Label Classificationcited by 11
  12. 2024Continual Learning Optimizations for Auto-regressive Decoder of Multilingual ASR systemscited by 7
10 more
  1. 2024Tokenization for Molecular Foundation Modelscited by 5
  2. 2022Switching to Discriminative Image Captioning by Relieving a Bottleneck of Reinforcement Learningcited by 4
  3. 2023AxomiyaBERTa: A Phonologically-aware Transformer Model for Assamesecited by 4
  4. 2025Compression Hacking: A Supplementary Perspective on Informatics Properties of Language Models from Geometric Distortioncited by 2
  5. 2024Norm of Mean Contextualized Embeddings Determines their Variancecited by 1
  6. 2021Shape of Elephant: Study of Macro Properties of Word Embeddings Spacescited by 0
  7. 2023LABO: Towards Learning Optimal Label Regularization via Bi-level Optimizationcited by 0
  8. 2024Reconsidering Degeneration of Token Embeddings with Definitions for Encoder-based Pre-trained Language Modelscited by 0
  9. 2026On the Tip of the Tongue: Why LLMs Hallucinate Answers They Can Decodecited by 0
  10. 2026The Geometry of Language Model Logits: Generic Structure and Learned Departurescited by 0