mlsanityData to 7 Oct 2026

Layer by Layer: Uncovering Hidden Representations in Language Models

arXiv 2502.02013 ↗First posted 4 Feb 2025cs.LG cs.AI cs.CLcited by 293

Not on the map. Why?

Cited by 39 in LLM reasoning, but placed by its own abstract, which didn’t fit.

Abstract (arXiv)

From extracting features to generating text, the outputs of large language models (LLMs) typically rely on the final layers, following the conventional wisdom that earlier layers capture only low-level cues. However, our analysis shows that intermediate layers can encode even richer representations, often improving performance on a range of downstream tasks. To explain and quantify these hidden-layer properties, we propose a unified framework of representation quality metrics based on information theory, geometry, and invariance to input perturbations. Our framework highlights how each layer balances information compression and signal preservation, revealing why mid-depth embeddings can exceed the last layer's performance. Through extensive experiments on 32 text-embedding tasks across various architectures (transformers, state-space models) and domains (language, vision), we demonstrate that intermediate layers consistently provide stronger features, challenging the standard view on final-layer embeddings and opening new directions on using mid-layer representations for more robust and accurate representations.

Key facts. Cited by 293 papers in the map (data through 7 Oct 2026). Builds on 53 of its 71 references that are in the map. First posted on arXiv 4 Feb 2025.

Builds on

53 of 71 references.

References · 53

  1. 2017Attention Is All You Needcited by 59,932
  2. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,456
  3. 2020Language Models are Few-Shot Learnerscited by 32,199
  4. 2020An Image is Worth 16x16 Words: Transformers for Image Recognition at Scalecited by 29,915
  5. 2024The Llama 3 Herd of Modelscited by 14,192
  6. 2023Llama 2: Open Foundation and Fine-Tuned Chat Modelscited by 12,865
  7. 2023DINOv2: Learning Robust Visual Features without Supervisioncited by 7,764
  8. 2018Representation Learning with Contrastive Predictive Codingcited by 7,662
  9. 2021Masked Autoencoders Are Scalable Vision Learnerscited by 7,339
  10. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,146
  11. 2023Mamba: Linear-Time Sequence Modeling with Selective State Spacescited by 4,616
  12. 2016Pointer Sentinel Mixture Modelscited by 2,992
38 more
  1. 2021BEiT: BERT Pre-Training of Image Transformerscited by 2,360
  2. 2023Efficient Streaming Language Models with Attention Sinkscited by 2,115
  3. 2023Pythia: A Suite for Analyzing Large Language Models Across Training and Scalingcited by 1,734
  4. 2022Competition-Level Code Generation with AlphaCodecited by 1,427
  5. 2019Contrastive Multiview Codingcited by 1,297
  6. 2016Understanding intermediate layers using linear classifier probescited by 1,279
  7. 2019BERT Rediscovers the Classical NLP Pipelinecited by 1,046
  8. 2024Training Large Language Models to Reason in a Continuous Latent Spacecited by 763
  9. 2022Discovering Latent Knowledge in Language Models Without Supervisioncited by 733
  10. 2023The Linear Representation Hypothesis and the Geometry of Large Language Modelscited by 679
  11. 2022MTEB: Massive Text Embedding Benchmarkcited by 623
  12. 2017Opening the Black Box of Deep Neural Networks via Informationcited by 494
  13. 2024LLM2Vec: Large Language Models Are Secretly Powerful Text Encoderscited by 396
  14. 2019Linguistic Knowledge and Transferability of Contextual Representationscited by 382
  15. 2023Language Models Represent Space and Timecited by 306
  16. 2023Language Modeling Is Compressioncited by 225
  17. 2024When Attention Sink Emerges in Language Models: An Empirical Viewcited by 200
  18. 2024The Geometry of Categorical and Hierarchical Concepts in Large Language Modelscited by 151
  19. 2022RankMe: Assessing the downstream performance of pretrained self-supervised representations by their rankcited by 145
  20. 2019The Bottom-up Evolution of Representations in the Transformer: A Study with Machine Translation and Language Modeling Objectivescited by 139
  21. 2024The Remarkable Robustness of LLMs: Stages of Inference?cited by 138
  22. 2019On Identifiability in Transformerscited by 115
  23. 2024Scalable Pre-training of Large Autoregressive Image Modelscited by 115
  24. 2023The geometry of hidden representations of large transformer modelscited by 108
  25. 2024Multimodal Autoregressive Pre-training of Large Vision Encoderscited by 108
  26. 2024Not All Layers of LLMs Are Necessary During Inferencecited by 95
  27. 2023Eliciting Latent Knowledge from Quirky Language Modelscited by 56
  28. 2024Emergence of a High-Dimensional Abstraction Phase in Language Transformerscited by 56
  29. 2020Emergence of Separable Manifolds in Deep Language Representationscited by 39
  30. 2022Information Theory with Kernel Methodscited by 37
  31. 2023The Shape of Learning: Anisotropy and Intrinsic Dimensions in Transformer-Based Modelscited by 37
  32. 2024Diff-eRank: A Novel Rank-Based Metric for Evaluating Large Language Modelscited by 35
  33. 2023Large language models implicitly learn to straighten neural sentence trajectories to construct a predictive representation of natural languagecited by 32
  34. 2023LiDAR: Sensing Linear Probing Performance in Joint Embedding SSL Architecturescited by 27
  35. 2022Guillotine Regularization: Why removing layers is needed to improve generalization in Self-Supervised Learningcited by 25
  36. 2022Information Flow in Deep Neural Networkscited by 22
  37. 2023DiME: Maximizing Mutual Information by a Difference of Matrix-Based Entropiescited by 20
  38. 2023Implicit regularization of deep residual networks towards neural ODEscited by 20

50 most cited of 53 All 53 (JSON)

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 108; 2026: 185.

293 papers.

Sort:

By area:

  1. 2025Towards Reasoning Era: A Survey of Long Chain-of-Thought for Reasoning Large Language Modelscited by 381
  2. 2026LTX-2: Efficient Joint Audio-Visual Foundation Modelcited by 213
  3. 2025Why do LLMs attend to the first token?cited by 122
  4. 2025A Survey on Latent Reasoningcited by 71
  5. 2025Do Language Models Use Their Depth Efficiently?cited by 48
  6. 2026LoopFormer: Elastic-Depth Looped Transformers for Latent Reasoning via Shortcut Modulationcited by 48
  7. 2025Attention Sinks and Compression Valleys in LLMs are Two Sides of the Same Coincited by 47
  8. 2025Unlocking General Long Chain-of-Thought Reasoning Capabilities of Large Language Models via Representation Engineeringcited by 34
  9. 2025Overcoming Sparsity Artifacts in Crosscoders to Interpret Chat-Tuningcited by 31
  10. 2025Encode, Think, Decode: Scaling test-time reasoning with recursive latent thoughtscited by 23
  11. 2026The Spike, the Sparse and the Sink: Anatomy of Massive Activations and Attention Sinkscited by 23
  12. 2025Mantis: Lightweight Foundation Model for Time Series Classificationcited by 21
38 more
  1. 2025BiggerGait: Unlocking Gait Recognition with Layer-wise Representations from Large Vision Modelscited by 18
  2. 2025Latent Thinking Optimization: Your Latent Reasoning Language Model Secretly Encodes Reward Signals in Its Latent Thoughtscited by 18
  3. 2025Adaptive Multi-Subspace Representation Steering for Attribute Alignment in Large Language Modelscited by 16
  4. 2025Exploring the Deep Fusion of Large Language Models and Diffusion Transformers for Text-to-Image Synthesiscited by 15
  5. 2025False Sense of Security: Why Probing-based Malicious Input Detection Fails to Generalizecited by 15
  6. 2025Demystifying the Roles of LLM Layers in Retrieval, Knowledge, and Reasoningcited by 15
  7. 2025Large Language Models Encode Semantics and Alignment in Linearly Separable Representationscited by 14
  8. 2025DREAM: Drafting with Refined Target Features and Entropy-Adaptive Cross-Attention Fusion for Multimodal Speculative Decodingcited by 13
  9. 2025Visual Multi-Agent System: Mitigating Hallucination Snowballing via Visual Flowcited by 13
  10. 2025Understanding Language Prior of LVLMs by Contrasting Chain-of-Embeddingcited by 13
  11. 2025LayerIF: Estimating Layer Quality for Large Language Models using Influence Functionscited by 12
  12. 2025GLSim: Detecting Object Hallucinations in LVLMs via Global-Local Similaritycited by 11
  13. 2026ROCKET: Residual-Oriented Multi-Layer Alignment for Spatially-Aware Vision-Language-Action Modelscited by 11
  14. 2026Prompt Injection as Role Confusioncited by 11
  15. 2025MindGuard: Intrinsic Decision Inspection for Securing LLM Agents Against Metadata Poisoningcited by 10
  16. 2025EvolKV: Evolutionary KV Cache Compression for LLM Inferencecited by 10
  17. 2026MantisV2: Closing the Zero-Shot Gap in Time Series Classification with Synthetic Data and Test-Time Strategiescited by 10
  18. 2026LLM Safety From Within: Detecting Harmful Content with Internal Representationscited by 10
  19. 2025Learning to Interpret Weight Differences in Language Modelscited by 9
  20. 2026From LLMs to LRMs: Rethinking Pruning for Reasoning-Centric Modelscited by 9
  21. 2026The Myth of Expert Specialization in MoEs: Why Routing Reflects Geometry, Not Necessarily Domain Expertisecited by 9
  22. 2025Probing Latent Subspaces in LLM for AI Security: Identifying and Manipulating Adversarial Statescited by 8
  23. 2025HiProbe-VAD: Video Anomaly Detection via Hidden States Probing in Tuning-Free Multimodal LLMscited by 8
  24. 2025Spectral Scaling Laws in Language Models: How Effectively Do Feed-Forward Networks Use Their Latent Space?cited by 8
  25. 2025Geometry of Decision Making in Language Modelscited by 8
  26. 2025UPCORE: Utility-Preserving Coreset Selection for Balanced Unlearningcited by 7
  27. 2025A quantitative analysis of semantic information in deep representations of text and imagescited by 7
  28. 2025GRACE: Generative Representation Learning via Contrastive Policy Optimizationcited by 7
  29. 2025LaRe: Latent Refocusing for Multimodal Reasoningcited by 7
  30. 2025Entropy-Lens: Uncovering Decision Strategies in LLMscited by 6
  31. 2025Enhancing Cross-task Transfer of Large Language Models via Activation Steeringcited by 6
  32. 2025Layerwise Importance Analysis of Feed-Forward Networks in Transformer-based Language Modelscited by 6
  33. 2025Revealing Perception and Generation Dynamics in LVLMs: Mitigating Hallucinations via Validated Dominance Correctioncited by 6
  34. 2026Differential syntactic and semantic encoding in LLMscited by 6
  35. 2026Context Structure Reshapes the Representational Geometry of Language Modelscited by 6
  36. 2026Knowledge Localization in Mixture-of-Experts LLMs Using Cross-Lingual Inconsistencycited by 6
  37. 2026UniVLR: Unifying Text and Vision in Visual Latent Reasoning for Multimodal LLMscited by 6
  38. 2026Hide-and-Seek in Trajectories: Discovering Failure Signals for VLA Runtime Monitoringcited by 6

50 most cited of 293 All 293 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2024Does Representation Matter? Exploring Intermediate Layers in Large Language Models
  2. 2026Rethinking Layer Relevance in Large Language Models Beyond Cosine Similarity
  3. 2023Jump to Conclusions: Short-Cutting Transformers With Linear Transformations
  4. 2026Hidden Decoding at Scale: Latent Computation Scaling for Large Language Models
  5. 2026On the Geometric Structure of Layer Updates in Deep Language Models
  6. 2026Layer Collapse in Diffusion Language Models
  7. 2026SPIDER: Multi-Layer Semantic Token Pruning and Adaptive Sub-Layer Skipping in Multimodal Large Language Models
  8. 2025How Language Directions Align with Token Geometry in Multilingual LLMs