Out-of-distribution generalization via composition: a lens through induction heads in Transformers

Proceedings of the National Academy of Sciences of the United States of America (PNAS), 2024

31 December 2024

Papers citing "Out-of-distribution generalization via composition: a lens through induction heads in Transformers"

50 / 99 papers shown

Title
Understanding the Staged Dynamics of Transformers in Learning Latent Structure Rohan Saha Farzane Aminmansour Alona Fyshe 28 0 0 24 Nov 2025
Can Language Models Compose Skills In-Context? Zidong Liu Zhuoyan Xu Zhenmei Shi Yingyu Liang ReLM CoGe LRM 207 0 0 27 Oct 2025
Can GRPO Help LLMs Transcend Their Pretraining Origin? Kangqi Ni Zhen Tan Zijie Liu Pingzhi Li Tianlong Chen OffRL LRM 78 0 0 14 Oct 2025
Localizing Task Recognition and Task Learning in In-Context Learning via Attention Head Analysis Haolin Yang Hakaze Cho Naoya Inoue 76 0 0 29 Sep 2025
Task Vectors, Learned Not Extracted: Performance Gains and Mechanistic Insight Haolin Yang Hakaze Cho Kaize Ding Naoya Inoue 120 0 0 29 Sep 2025
Is Chain-of-Thought Reasoning of LLMs a Mirage? A Data Distribution Lens Chengshuai Zhao Zhen Tan Pingchuan Ma Dawei Li Bohan Jiang Yancheng Wang Yingzhen Yang Huan Liu LRM 249 27 0 02 Aug 2025
Born a Transformer -- Always a Transformer? On the Effect of Pretraining on Architectural Abilities Yana Veitsman Mayank Jobanputra Yash Sarrof Aleksandra Bakalova Vera Demberg Ellie Pavlick Michael Hahn 390 2 0 27 May 2025
Decomposing Complex Visual Comprehension into Atomic Visual Skills for Vision Language Models Hyunsik Chae Seungwoo Yoon J. Park Chloe Yewon Chun Yongin Cho Mu Cai Yong Jae Lee Ernest K. Ryu CoGe VLM 242 3 0 26 May 2025
Characterizing Pattern Matching and Its Limits on Compositional Task Structures Hoyeon Chang Jinho Park Hanseul Cho Sohee Yang Miyoung Ko Hyeonbin Hwang Seungpil Won Dohaeng Lee Youbin Ahn Minjoon Seo 230 1 0 26 May 2025
Unifying Attention Heads and Task Vectors via Hidden State Geometry in In-Context Learning Haolin Yang Hakaze Cho Yiqiao Zhong Naoya Inoue 125 2 0 24 May 2025
ICL CIPHERS: Quantifying "Learning" in In-Context Learning via Substitution Ciphers Zhouxiang Fang Aayush Mishra Muhan Gao Anqi Liu Daniel Khashabi 353 2 0 28 Apr 2025
Learning to Inference Adaptively for Multimodal Large Language Models Zhuoyan Xu Khoi Duc Nguyen Preeti Mukherjee Saurabh Bagchi Somali Chaterji Yingyu Liang Yin Li LRM 315 4 0 13 Mar 2025
Can In-context Learning Really Generalize to Out-of-distribution Tasks?International Conference on Learning Representations (ICLR), 2024 Qixun Wang Yifei Wang Yisen Wang Xianghua Ying OOD 179 15 0 13 Oct 2024
GSM-Symbolic: Understanding the Limitations of Mathematical Reasoning in Large Language ModelsInternational Conference on Learning Representations (ICLR), 2024 Iman Mirzadeh Keivan Alizadeh Hooman Shahrokhi Oncel Tuzel Samy Bengio Mehrdad Farajtabar AIMat LRM 433 386 0 07 Oct 2024
Task Diversity Shortens the ICL Plateau Jaeyeon Kim Sehyun Kwon Joo Young Choi Jongho Park Jaewoong Cho Jason D. Lee Ernest K. Ryu MoMe 318 3 0 07 Oct 2024
Generalization vs. Specialization under Concept Shift Alex Nguyen David J. Schwab Vudtiwat Ngampruetikorn OOD 196 0 0 23 Sep 2024
Gemma 2: Improving Open Language Models at a Practical Size Gemma Team Gemma Team Morgane Riviere Shreya Pathak Pier Giuseppe Sessa Cassidy Hardin ... Noah Fiedel Armand Joulin Kathleen Kenealy Robert Dadashi Alek Andreev VLM MoE OSLM 549 1,485 0 31 Jul 2024
Emergence in non-neural models: grokking modular arithmetic via average gradient outer product Neil Rohit Mallinar Daniel Beaglehole Libin Zhu Adityanarayanan Radhakrishnan Parthe Pandit Misha Belkin 295 14 0 29 Jul 2024
Do Large Language Models Have Compositional Ability? An Investigation into Limitations and Scalability Zhuoyan Xu Zhenmei Shi Yingyu Liang CoGe LRM 290 50 0 22 Jul 2024
Scaling and evaluating sparse autoencoders Leo Gao Tom Dupré la Tour Henk Tillman Gabriel Goh Rajan Troll Alec Radford Ilya Sutskever Jan Leike Jeffrey Wu 232 283 0 06 Jun 2024
Why Larger Language Models Do In-context Learning Differently? Zhenmei Shi Junyi Wei Zhuoyan Xu Yingyu Liang 199 43 0 30 May 2024
Grokked Transformers are Implicit Reasoners: A Mechanistic Journey to the Edge of Generalization Boshi Wang Xiang Yue Yu-Chuan Su Huan Sun LRM 284 71 0 23 May 2024
Position: Understanding LLMs Requires More Than Statistical GeneralizationInternational Conference on Machine Learning (ICML), 2024 Patrik Reizinger Szilvia Ujváry Anna Mészáros A. Kerekes Wieland Brendel Ferenc Huszár 282 21 0 03 May 2024
Many-Shot In-Context Learning Rishabh Agarwal Avi Singh Lei M. Zhang Bernd Bohnet Luis Rosias ... John D. Co-Reyes Eric Chu Feryal M. P. Behbahani Aleksandra Faust Hugo Larochelle ReLM OffRL BDL 366 172 0 17 Apr 2024
What needs to go right for an induction head? A mechanistic study of in-context learning circuits and their formationInternational Conference on Machine Learning (ICML), 2024 Aaditya K. Singh Ted Moskovitz Felix Hill Stephanie C. Y. Chan Andrew M. Saxe AI4CE 238 53 0 10 Apr 2024
What Can Transformer Learn with Varying Depth? Case Studies on Sequence Learning TasksInternational Conference on Machine Learning (ICML), 2024 Xingwu Chen Difan Zou ViT 247 20 0 02 Apr 2024
Gemma: Open Models Based on Gemini Research and Technology Gemma Team Gemma Team Thomas Mesnard Cassidy Hardin Robert Dadashi Surya Bhupatiraju ... Armand Joulin Noah Fiedel Evan Senter Alek Andreev Kathleen Kenealy VLM LLMAG 505 805 0 13 Mar 2024
Machine learning and information theory concepts towards an AI MathematicianBulletin of the American Mathematical Society (BAMS), 2024 Yoshua Bengio Nikolay Malkin 161 14 0 07 Mar 2024
Towards Few-Shot Adaptation of Foundation Models via Multitask Finetuning Zhuoyan Xu Zhenmei Shi Junyi Wei Fangzhou Mu Yin Li Yingyu Liang 92 33 0 22 Feb 2024
OLMo: Accelerating the Science of Language Models Dirk Groeneveld Iz Beltagy Pete Walsh Akshita Bhagia Rodney Michael Kinney ... Jesse Dodge Kyle Lo Luca Soldaini Noah A. Smith Hanna Hajishirzi OSLM 585 538 0 01 Feb 2024
In-Context Language Learning: Architectures and AlgorithmsInternational Conference on Machine Learning (ICML), 2024 Ekin Akyürek Bailin Wang Yoon Kim Jacob Andreas LRM ReLM 273 74 0 23 Jan 2024
Successor Heads: Recurring, Interpretable Attention Heads In The WildInternational Conference on Learning Representations (ICLR), 2023 Rhys Gould Euan Ong George Ogden Arthur Conmy LRM 214 63 0 14 Dec 2023
The mechanistic basis of data dependence and abrupt learning in an in-context classification taskInternational Conference on Learning Representations (ICLR), 2023 Gautam Reddy 256 87 0 03 Dec 2023
Dichotomy of Early and Late Phase Implicit Biases Can Provably Induce GrokkingInternational Conference on Learning Representations (ICLR), 2023 Kaifeng Lyu Jikai Jin Zhiyuan Li Simon S. Du Jason D. Lee Wei Hu AI4CE 265 52 0 30 Nov 2023
The Falcon Series of Open Language Models Ebtesam Almazrouei Hamza Alobeidli Abdulaziz Alshamsi Alessandro Cappelli Ruxandra-Aimée Cojocaru ... Quentin Malartic Daniele Mazzotta Badreddine Noune B. Pannier Guilherme Penedo AI4TS ALM 408 589 0 28 Nov 2023
The Linear Representation Hypothesis and the Geometry of Large Language ModelsInternational Conference on Machine Learning (ICML), 2023 Kiho Park Yo Joong Choe Victor Veitch LLMSV MILM 409 307 0 07 Nov 2023
What Algorithms can Transformers Learn? A Study in Length GeneralizationInternational Conference on Learning Representations (ICLR), 2023 Hattie Zhou Arwen Bradley Etai Littwin Noam Razin Omid Saremi Josh Susskind Samy Bengio Preetum Nakkiran 219 157 0 24 Oct 2023
When can transformers reason with abstract symbols? Enric Boix-Adserà Omid Saremi Emmanuel Abbe Samy Bengio Etai Littwin Josh Susskind LRM NAI 220 19 0 15 Oct 2023
Circuit Component Reuse Across Tasks in Transformer Language ModelsInternational Conference on Learning Representations (ICLR), 2023 Jack Merullo Carsten Eickhoff Ellie Pavlick 282 94 0 12 Oct 2023
Mistral 7B Albert Q. Jiang Alexandre Sablayrolles A. Mensch Chris Bamford Devendra Singh Chaplot ... Teven Le Scao Thibaut Lavril Thomas Wang Timothée Lacroix William El Sayed MoE LRM 338 2,871 0 10 Oct 2023
Uncovering hidden geometry in Transformers via disentangling position and context Jiajun Song Yiqiao Zhong 198 12 0 07 Oct 2023
A Theory for Emergence of Complex Skills in Language Models Sanjeev Arora Anirudh Goyal LRM 282 98 0 29 Jul 2023
Overthinking the Truth: Understanding how Language Models Process False DemonstrationsInternational Conference on Learning Representations (ICLR), 2023 Danny Halawi Jean-Stanislas Denain Jacob Steinhardt 247 72 0 18 Jul 2023
Llama 2: Open Foundation and Fine-Tuned Chat Models Hugo Touvron Louis Martin Kevin R. Stone Peter Albert Amjad Almahairi ... Sharan Narang Aurelien Rodriguez Robert Stojnic Sergey Edunov Thomas Scialom AI4MH ALM 5.4K 14,923 0 18 Jul 2023
One Step of Gradient Descent is Provably the Optimal In-Context Learner with One Layer of Linear Self-AttentionInternational Conference on Learning Representations (ICLR), 2023 Arvind V. Mahankali Tatsunori B. Hashimoto Tengyu Ma MLT 152 139 0 07 Jul 2023
Teaching Arithmetic to Small Transformers Nayoung Lee Kartik K. Sreenivasan Jason D. Lee Kangwook Lee Dimitris Papailiopoulos LRM 235 113 0 07 Jul 2023
Reasoning or Reciting? Exploring the Capabilities and Limitations of Language Models Through Counterfactual TasksNorth American Chapter of the Association for Computational Linguistics (NAACL), 2023 Zhaofeng Wu Linlu Qiu Alexis Ross Ekin Akyürek Boyuan Chen Bailin Wang Najoung Kim Jacob Andreas Yoon Kim LRM ReLM 358 290 0 05 Jul 2023
The Clock and the Pizza: Two Stories in Mechanistic Explanation of Neural NetworksNeural Information Processing Systems (NeurIPS), 2023 Ziqian Zhong Ziming Liu Max Tegmark Jacob Andreas 238 132 0 30 Jun 2023
Trained Transformers Learn Linear Models In-ContextJournal of machine learning research (JMLR), 2023 Ruiqi Zhang Spencer Frei Peter L. Bartlett 341 273 0 16 Jun 2023
Transformers as Statisticians: Provable In-Context Learning with In-Context Algorithm SelectionNeural Information Processing Systems (NeurIPS), 2023 Yu Bai Fan Chen Haiquan Wang Caiming Xiong Song Mei 211 256 0 07 Jun 2023