mlsanityData to 7 Oct 2026

Birth of a Transformer: A Memory Viewpoint

arXiv 2306.00802 ↗First posted 1 Jun 2023stat.ML cs.CL cs.LGcited by 154

Not on the map. Why?

Abstract (arXiv)

Large language models based on transformers have achieved great empirical successes. However, as they are deployed more widely, there is a growing need to better understand their internal mechanisms in order to make them more reliable. These models appear to store vast amounts of knowledge from their training data, and to adapt quickly to new information provided in their context or prompt. We study how transformers balance these two types of knowledge by considering a synthetic setup where tokens are generated from either global or context-specific bigram distributions. By a careful empirical analysis of the training process on a simplified two-layer transformer, we illustrate the fast learning of global bigrams and the slower development of an "induction head" mechanism for the in-context bigrams. We highlight the role of weight matrices as associative memories, provide theoretical insights on how gradients enable their learning during training, and study the role of data-distributional properties.

Key facts. Cited by 154 papers in the map (data through 7 Oct 2026). Builds on 46 of its 61 references that are in the map. First posted on arXiv 1 Jun 2023.

Builds on

46 of 61 references.

References · 46

  1. 2017Attention Is All You Needcited by 59,932
  2. 2020Language Models are Few-Shot Learnerscited by 32,199
  3. 2016Layer Normalizationcited by 5,579
  4. 2022PaLM: Scaling Language Modeling with Pathwayscited by 5,304
  5. 2013Efficient Estimation of Word Representations in Vector Spacecited by 4,346
  6. 2022Locating and Editing Factual Associations in GPTcited by 2,609
  7. 2018Neural Tangent Kernel: Convergence and Generalization in Neural Networkscited by 2,597
  8. 2019Root Mean Square Layer Normalizationcited by 1,425
  9. 2022Rethinking the Role of Demonstrations: What Makes In-Context Learning Work?cited by 1,413
  10. 2020Transformer Feed-Forward Layers Are Key-Value Memoriescited by 1,307
  11. 2021Scaling Language Models: Methods, Analysis & Insights from Training Gophercited by 1,199
  12. 2022Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 smallcited by 1,130
34 more
  1. 2022In-context Learning and Induction Headscited by 930
  2. 2022Toy Models of Superpositioncited by 928
  3. 2021An Explanation of In-context Learning as Implicit Bayesian Inferencecited by 915
  4. 2023Progress measures for grokking via mechanistic interpretabilitycited by 822
  5. 2013Exact solutions to the nonlinear dynamics of learning in deep linear neural networkscited by 748
  6. 2018On Lazy Training in Differentiable Programmingcited by 671
  7. 2014Neural Turing Machinescited by 630
  8. 2023Dissecting Recall of Factual Associations in Auto-Regressive Language Modelscited by 512
  9. 2021Linear Transformers Are Secretly Fast Weight Programmerscited by 471
  10. 2014Memory Networkscited by 354
  11. 2017Implicit Regularization in Matrix Factorizationcited by 318
  12. 2022Data Distributional Properties Drive Emergent In-Context Learning in Transformerscited by 303
  13. 2020Towards Understanding Ensemble, Knowledge Distillation and Self-Distillation in Deep Learningcited by 292
  14. 2022Tensor Programs V: Tuning Large Neural Networks via Zero-Shot Hyperparameter Transfercited by 272
  15. 2022Transformers Learn Shortcuts to Automatacited by 260
  16. 2016Dense Associative Memory for Pattern Recognitioncited by 230
  17. 2022Towards Understanding Grokking: An Effective Theory of Representation Learningcited by 229
  18. 2022Neural Networks can Learn Representations with Gradient Descentcited by 161
  19. 2022High-dimensional Asymptotics of Feature Learning: How One Gradient Step Improves the Representationcited by 146
  20. 2021Inductive Biases and Variable Creation in Self-Attention Mechanismscited by 144
  21. 2022Impact of Pretraining Term Frequencies on Few-Shot Reasoningcited by 142
  22. 2021Saturated Transformers are Constant-Depth Threshold Circuitscited by 137
  23. 2022Analyzing Transformers in Embedding Spacecited by 131
  24. 2020Infinite attention: NNGP and NTK for deep attention networkscited by 117
  25. 2022Unveiling Transformers with LEGO: a synthetic reasoning taskcited by 105
  26. 2023Scan and Snap: Understanding Training Dynamics and Token Composition in 1-layer Transformercited by 101
  27. 2022Vision Transformers provably learn spatial structurecited by 92
  28. 2023How Do Transformers Learn Topic Structure: Towards a Mechanistic Understandingcited by 85
  29. 2022On the Effect of Pretraining Corpora on In-context Learning by a Large-scale Language Modelcited by 78
  30. 2015Inferring Algorithmic Patterns with Stack-Augmented Recurrent Netscited by 76
  31. 2021Approximating How Single Head Attention Learnscited by 38
  32. 2021Attention Approximates Sparse Distributed Memorycited by 29
  33. 2022What is my math transformer doing? -- Three results on interpretability and generalizationcited by 19
  34. 2020Associative Memory in Iterated Overparameterized Sigmoid Autoencoderscited by 9

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2023: 8; 2024: 58; 2025: 53; 2026: 35.

154 papers.

Sort:

By area:

  1. 2024Titans: Learning to Memorize at Test Timecited by 303
  2. 2025TTT3R: 3D Reconstruction as Test-Time Trainingcited by 122
  3. 2024The Evolution of Statistical Induction Heads: In-Context Learning Markov Chainscited by 115
  4. 2024How Transformers Learn Causal Structure with Gradient Descentcited by 114
  5. 2024A Practical Review of Mechanistic Interpretability for Transformer-Based Language Modelscited by 113
  6. 2024Secret Collusion among AI Agents: Multi-Agent Deception via Steganographycited by 107
  7. 2024A Primer on the Inner Workings of Transformer-based Language Modelscited by 83
  8. 2024Heavy-Tailed Class Imbalance and Why Adam Outperforms Gradient Descent on Language Modelscited by 78
  9. 2024Transformers, parallel computation, and logarithmic depthcited by 75
  10. 2023How Do Transformers Learn In-Context Beyond Simple Functions? A Case Study on Learning with Representationscited by 72
  11. 2024Attention Heads of Large Language Models: A Surveycited by 65
  12. 2025Test-time regression: a unifying framework for designing sequence models with associative memorycited by 63
38 more
  1. 2023Towards Vision-Language Mechanistic Interpretability: A Causal Tracing Tool for BLIPcited by 62
  2. 2025It's All Connected: A Journey Through Test-Time Memorization, Attentional Bias, Retention, and Online Optimizationcited by 60
  3. 2025ATLAS: Learning to Optimally Memorize the Context at Test Timecited by 57
  4. 2024Asymptotic theory of in-context learning by linear attentioncited by 54
  5. 2024Active-Dormant Attention Heads: Mechanistically Demystifying Extreme-Token Phenomena in LLMscited by 54
  6. 2023JoMA: Demystifying Multilayer Transformers via JOint Dynamics of MLP and Attentioncited by 52
  7. 2024Why Larger Language Models Do In-context Learning Differently?cited by 44
  8. 2024Attention with Markov: A Framework for Principled Analysis of Transformers via Markov Chainscited by 39
  9. 2024Outlier-Efficient Hopfield Layers for Large Transformer-Based Modelscited by 38
  10. 2025Muon Outperforms Adam in Tail-End Associative Memory Learningcited by 38
  11. 2023On the Optimization and Generalization of Multi-head Attentioncited by 37
  12. 2025Beyond Semantics: Rediscovering Spatial Awareness in Vision-Language Modelscited by 35
  13. 2025Emergent Abilities in Large Language Models: A Surveycited by 34
  14. 2024Large Language Models as Markov Chainscited by 33
  15. 2024Implicit Bias and Fast Convergence Rates for Self-attentioncited by 31
  16. 2023Scaling Laws for Associative Memoriescited by 30
  17. 2024A phase transition between positional and semantic learning in a solvable model of dot-product attentioncited by 30
  18. 2024One-layer transformers fail to solve the induction heads taskcited by 30
  19. 2024Towards a Theoretical Understanding of the 'Reversal Curse' via Training Dynamicscited by 27
  20. 2025Taming Knowledge Conflicts in Language Modelscited by 27
  21. 2024Transformers on Markov Data: Constant Depth Sufficescited by 26
  22. 2025The emergence of sparse attention: impact of data distribution and benefits of repetitioncited by 25
  23. 2024BiSHop: Bi-Directional Cellular Learning for Tabular Data with Generalized Sparse Modern Hopfield Modelcited by 21
  24. 2024Bypassing the Exponential Dependency: Looped Transformers Efficiently Learn In-context by Multi-step Gradient Descentcited by 21
  25. 2026There Will Be a Scientific Theory of Deep Learningcited by 21
  26. 2025How Do LLMs Perform Two-Hop Reasoning in Context?cited by 20
  27. 2024Self-attention Networks Localize When QK-eigenspectrum Concentratescited by 19
  28. 2024Provably Optimal Memory Capacity for Modern Hopfield Models: Transformer-Compatible Dense Associative Memories as Spherical Codescited by 19
  29. 2024Differential learning kinetics govern the transition from memorization to generalization during in-context learningcited by 19
  30. 2024Memory Mosaicscited by 16
  31. 2024Understanding the Language Model to Solve the Symbolic Multi-Step Reasoning Problem from the Perspective of Buffer Mechanismcited by 15
  32. 2024Attention layers provably solve single-location regressioncited by 15
  33. 2025Understanding the Skill Gap in Recurrent Language Models: The Role of the Gather-and-Aggregate Mechanismcited by 15
  34. 2025Learning In-context n-grams with Transformers: Sub-n-grams Are Near-stationary Pointscited by 15
  35. 2024Learning Associative Memories with Gradient Descentcited by 14
  36. 2024Do LLMs dream of elephants (when told not to)? Latent concept association and associative memory in transformerscited by 14
  37. 2023In-Context Learning Dynamics with Random Binary Sequencescited by 13
  38. 2025Where to show Demos in Your Prompt: A Positional Bias of In-Context Learningcited by 13

50 most cited of 154 All 154 (JSON)