Learning through Internalization

• URL: http://arxiv.org/abs/2606.20937v1
• Date: Thu, 18 Jun 2026 21:01:42 GMT
• Title: Learning through Internalization
• Abstract summary: We study internalization processes, by which neural-network-based systems absorb an explicit computational procedure into their own weights.<n>We investigate how transformers internalize the simulation of semiautomata by internalizing chain-of-thought (CoT) tokens.<n>We show that a simplified one-layer transformer provably first learns the target with explicit CoT supervision and then internalizes the autoregressive generation as CoT tokens are progressively removed.
• Score: 34.00603055182392
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: We study internalization processes, by which neural-network-based systems absorb an explicit computational procedure into their own weights, and how they facilitate learning. We investigate how transformers internalize the simulation of semiautomata by internalizing chain-of-thought (CoT) tokens, which classes of semiautomata are harder to internalize, and expose the flip side of internalization, that is, a progressive degradation of out-of-distribution performance. We then provide the first provable analysis of successful internalization: for the task of learning parities, we show that a simplified one-layer transformer provably first learns the target with explicit CoT supervision and then internalizes the autoregressive generation as CoT tokens are progressively removed, learning to directly compute the parity. This task is computationally hard to learn from data without CoT supervision. Finally, we discuss how learning through internalization relates to the \textit{Positive Distribution Shift} phenomenon recently introduced by~\citet{Med+26}.

Related papers

• A train--prune--readout--rewrite workflow for interpretable quantum learning [5.641464469724376]
  AI for Science aims not only to predict complex physical systems from data, but also to extract mathematical structure and physically testable representations from learned models.<n>Here, a train--prune--readout--rewrite workflow is developed that separates physical-domain grounding from three increasingly stringent analysis claims.<n>The workflow is implemented with complex-valued networks, whose explicit edge functions enable post-pruning analytic readout of the retained computation.
  arXiv  Detail & Related papers  (2026-09-14T07:12:21Z)
• Tracing the Thought of a Grandmaster-level Chess-Playing Transformer [62.403498221631715]
  We introduce a sparse decomposition framework to interpret Leela Chess Zero's internal computation.<n>We conduct a detailed case study showing that these pathways expose rich, interpretable considerations.<n>To the best of our knowledge, this is the first work to decompose the internal computation of a transformer on both computation and attention modules.
  arXiv  Detail & Related papers  (2026-04-11T11:11:06Z)
• Provable Benefit of Curriculum in Transformer Tree-Reasoning Post-Training [76.12556589212666]
  We show that curriculum post-training avoids the exponential complexity bottleneck.<n>Under outcome-only reward signals, reinforcement learning finetuning achieves high accuracy with sample complexity.<n>We establish guarantees for test-time scaling, where curriculum-aware querying reduces both reward oracle calls and sampling cost from exponential to order.
  arXiv  Detail & Related papers  (2025-11-10T18:29:54Z)
• The Kinetics of Reasoning: How Chain-of-Thought Shapes Learning in Transformers? [25.29458951592086]
  Chain-of-thought (CoT) supervision can substantially improve transformer performance.<n>We investigate these learning dynamics through the lens of grokking by pretraining transformers on symbolic reasoning tasks.
  arXiv  Detail & Related papers  (2025-10-28T20:14:26Z)
• (How) Do Language Models Track State? [52.03388573068501]
  Transformer language models (LMs) exhibit behaviors that seem to require tracking the unobserved state of an evolving world.<n>We study state tracking in LMs trained or fine-tuned to compose permutations.<n>We show that LMs consistently learn one of two state tracking mechanisms for this task.
  arXiv  Detail & Related papers  (2025-03-04T18:31:02Z)
• A distributional simplicity bias in the learning dynamics of transformers [50.91742043564049]
  We show that transformers, trained on natural language data, also display a simplicity bias.<n>Specifically, they sequentially learn many-body interactions among input tokens, reaching a saturation point in the prediction error for low-degree interactions.<n>This approach opens up the possibilities of studying how interactions of different orders in the data affect learning, in natural language processing and beyond.
  arXiv  Detail & Related papers  (2024-10-25T15:39:34Z)
• Interpreting Affine Recurrence Learning in GPT-style Transformers [54.01174470722201]
  In-context learning allows GPT-style transformers to generalize during inference without modifying their weights. This paper focuses specifically on their ability to learn and predict affine recurrences as an ICL task. We analyze the model's internal operations using both empirical and theoretical approaches.
  arXiv  Detail & Related papers  (2024-10-22T21:30:01Z)
• Progress measures for grokking via mechanistic interpretability [27.35925102247588]
  We study the recently-discovered phenomenon of grokking'' exhibited by small transformers trained on modular addition tasks. Our results show that grokking, rather than being a sudden shift, arises from the gradual amplification of structured mechanisms encoded in the weights.
  arXiv  Detail & Related papers  (2023-01-12T18:56:49Z)
• Understanding Learning Dynamics for Neural Machine Translation [53.23463279153577]
  We propose to understand learning dynamics of NMT by using Loss Change Allocation (LCA)citeplan 2019-loss-change-allocation. As LCA requires calculating the gradient on an entire dataset for each update, we instead present an approximate to put it into practice in NMT scenario. Our simulated experiment shows that such approximate calculation is efficient and is empirically proved to deliver consistent results.
  arXiv  Detail & Related papers  (2020-04-05T13:32:58Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.