Continual Self-Improvement with Lightweight Experiential Latent Memories

• URL: http://arxiv.org/abs/2606.17803v1
• Date: Tue, 16 Jun 2026 11:27:28 GMT
• Title: Continual Self-Improvement with Lightweight Experiential Latent Memories
• Abstract summary: Large language models achieve strong reasoning performance by scaling inference-time compute.<n>We show that In-Context Learning (ICL) over raw reasoning traces fails to generalize.<n>We propose an online method that distills inference-time compute spent on encountered problems into compact modular latent memories.
• Score: 5.476321641759679
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large language models achieve strong reasoning performance by scaling inference-time compute, yet remain fundamentally stateless, discarding the rich, self-produced reasoning traces generated during this process. We investigate whether models can instead learn online from this experience, converting transient computation (reasoning traces) into persistent reusable knowledge, and without external supervision or access to future data. We show that In-Context Learning (ICL) over raw reasoning traces fails to generalize, reflecting a fundamental limitation of token-level reuse: individual traces lack the abstraction needed for transfer, even after refinement (e.g. self-reflection). In contrast, drawing inspiration from recent works on unsupervised reinforcement learning, we find that lightweight per-instance training with self-generated test-time signals (majority voting) as rewards yields substantial gains, often surpassing full-dataset offline training, motivating a shift from raw traces to learned latent representations. Building on this insight, we propose an online method that distills inference-time compute spent on encountered problems into compact modular latent memories capturing the underlying reasoning structure. These memories are stored and retrieved for future inputs, enabling continual improvement while avoiding catastrophic forgetting through modular design. Importantly, our method is highly efficient, parametrized as extremely lightweight soft prompt memories (~0.001% of model parameters) and trained with only a few gradient steps, yet achieving performance competitive with full parametric updates and offline training. Across challenging mathematical reasoning benchmarks, our approach significantly outperforms zero-shot and raw data ICL baselines, while transferring effectively across datasets.

Related papers

• Train Smarter, Not Longer: Memorization-Guided Data Reuse for Efficient LLM Training [11.074386515074446]
  "Memorization-guided Data Reuse" is a training paradigm that adaptively determines when and how data should be reused.<n>Preliminary experiments reveal a consistent memorization-driven regime.
  arXiv  Detail & Related papers  (2026-07-06T11:57:32Z)
• A Progressive Training Strategy for Vision-Language Models to Counteract Spatio-Temporal Hallucinations in Embodied Reasoning [49.61652671596548]
  "Multi-image hallucination reasoning", where a massive performance drop between forward and temporal queries reveals a dependence on superficial shortcuts instead of genuine understanding.<n>To mitigate this, we develop a new dataset: Chain-of-temporal building on detailed reasoning into Chain steps and definitive judgments.<n>Our experiments demonstrate that this approach not only improves accuracy but also the forward-backward performance gap from over 70% to only 6.53%.
  arXiv  Detail & Related papers  (2026-04-12T07:48:44Z)
• Forget Less, Retain More: A Lightweight Regularizer for Rehearsal-Based Continual Learning [51.07663354001582]
  Deep neural networks suffer from catastrophic forgetting, where performance on previous tasks degrades after training on a new task.<n>We present a novel approach to address this challenge, focusing on the intersection of memory-based methods and regularization approaches.<n>We formulate a regularization strategy, termed Information Maximization (IM) regularizer, for memory-based continual learning methods.
  arXiv  Detail & Related papers  (2025-12-01T15:56:00Z)
• Efficient Machine Unlearning via Influence Approximation [75.31015485113993]
  Influence-based unlearning has emerged as a prominent approach to estimate the impact of individual training samples on model parameters without retraining.<n>This paper establishes a theoretical link between memorizing (incremental learning) and forgetting (unlearning)<n>We introduce the Influence Approximation Unlearning algorithm for efficient machine unlearning from the incremental perspective.
  arXiv  Detail & Related papers  (2025-07-31T05:34:27Z)
• Temporal Sampling for Forgotten Reasoning in LLMs [14.895026967556088]
  We show that models often forget how to solve problems they previously answered correctly during training.<n>We introduce Temporal Sampling, a simple decoding strategy that draws outputs from multiple checkpoints along the training trajectory.<n>This approach recovers forgotten solutions without retraining or ensembling, and leads to substantial improvements in reasoning performance.
  arXiv  Detail & Related papers  (2025-05-26T16:39:52Z)
• Remember and Recall: Associative-Memory-based Trajectory Prediction [25.349986959111757]
  We propose the Fragmented-Memory-based Trajectory Prediction (FMTP) model, inspired by the remarkable learning capabilities of humans. The FMTP model employs discrete representations to enhance computational efficiency by reducing information redundancy. We develop an advanced reasoning engine based on language models to deeply learn the associative rules among these discrete representations.
  arXiv  Detail & Related papers  (2024-10-03T04:32:21Z)
• Towards Robust Continual Learning with Bayesian Adaptive Moment Regularization [51.34904967046097]
  Continual learning seeks to overcome the challenge of catastrophic forgetting, where a model forgets previously learnt information. We introduce a novel prior-based method that better constrains parameter growth, reducing catastrophic forgetting. Results show that BAdam achieves state-of-the-art performance for prior-based methods on challenging single-headed class-incremental experiments.
  arXiv  Detail & Related papers  (2023-09-15T17:10:51Z)
• Fast Machine Unlearning Without Retraining Through Selective Synaptic Dampening [51.34904967046097]
  Selective Synaptic Dampening (SSD) is a fast, performant, and does not require long-term storage of the training data. We present a novel two-step, post hoc, retrain-free approach to machine unlearning which is fast, performant, and does not require long-term storage of the training data.
  arXiv  Detail & Related papers  (2023-08-15T11:30:45Z)
• Dissecting Continual Learning a Structural and Data Analysis [0.0]
  Continual Learning is a field dedicated to devise algorithms able to achieve lifelong learning. Deep learning methods can attain impressive results when the data modeled does not undergo a considerable distributional shift in subsequent learning sessions. When we expose such systems to this incremental setting, performance drop very quickly.
  arXiv  Detail & Related papers  (2023-01-03T10:37:11Z)
• Automatic Recall Machines: Internal Replay, Continual Learning and the Brain [104.38824285741248]
  Replay in neural networks involves training on sequential data with memorized samples, which counteracts forgetting of previous behavior caused by non-stationarity. We present a method where these auxiliary samples are generated on the fly, given only the model that is being trained for the assessed objective. Instead the implicit memory of learned samples within the assessed model itself is exploited.
  arXiv  Detail & Related papers  (2020-06-22T15:07:06Z)

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.