Recirculation

• URL: http://arxiv.org/abs/2608.17981v1
• Date: Tue, 18 Aug 2026 16:30:21 GMT
• Title: Recirculation
• Abstract summary: We describe an inference-time architectural enhancement for off-the-shelf foundation models.<n>Our approach incurs essentially no additional latency during generation, though it requires serial processing in the prefill phase.<n>Relative to the off-the-shelf baseline, adaptive recirculation achieves remarkable gains.
• Score: 10.742684685694869
• License: http://creativecommons.org/licenses/by-nc-sa/4.0/
• Abstract: We describe an inference-time architectural enhancement for off-the-shelf foundation models that markedly reduces perplexity and boosts accuracy across generation and reasoning tasks. Our approach incurs essentially no additional latency during generation, though it requires serial processing in the prefill phase. Motivated by the fundamental limitation that state updates in feedforward transformers are bounded by model depth, our technique, recirculation, introduces a specific form of recurrence that allows the model to act as a dynamical system and track belief states. We distinguish this technique from chain-of-thought computation---which is better reserved for complex inferences rather than basic state tracking---as well as from popular depth-recurrence techniques (looping) and the costly training of recurrent transformers. We also propose and evaluate an adaptive variant of recirculation which requires only light tuning of hyperparameters while freezing the original model weights. Relative to the off-the-shelf baseline, adaptive recirculation achieves remarkable gains on the Gemma3 family, including a 23% reduction in perplexity on a suite of datasets, a 21% increase in accuracy on GSM8k, and reliable improvements in accuracy on other downstream tasks. Our training-free approach succeeds by leveraging the model itself to inform architectural modifications, suggesting a route to architectural evolution guided by a trained network's properties rather than forced, arbitrary design choices.

Related papers

• Resource-Efficient Pruning for Transformer via Low-Rank Importance Estimation [20.409069904706346]
  REP-LIE is a new approach to enable resource-efficient pruning during the process of finetuning.<n>To address the inherent randomness in importance estimation, a stability score is introduced.<n>Experiments on both medium-scale encoder models and large-scale generative models (LLaMA-7B and Mistral-7B) demonstrate that REP-LIE still achieves competitive performance compared to existing approaches.
  arXiv  Detail & Related papers  (2026-08-25T11:36:01Z)
• Adaptive Recurrent Message Passing for Test Time Computing on Graphs [80.56829418752513]
  We show that recurrent graph models can overcome the intrinsic mismatch between graph data and architectural designs.<n>We propose AdaR, an Adaptive Recurrent graph model, empowering flexible test-time computing on various downstream tasks without changing model parameters.
  arXiv  Detail & Related papers  (2026-06-21T12:14:11Z)
• Unlocking Feature Learning in Gated Delta Networks at Scale [52.705562336559446]
  Training and scaling Large Language Models demand enormous computational resources.<n>We derive the scaling rules for Gated Delta Network.<n> Experiments on language-model pre-training confirm that our configurations enable stable learning-rate transfer.
  arXiv  Detail & Related papers  (2026-06-02T08:45:24Z)
• Neural Network Optimization Reimagined: Decoupled Techniques for Scratch and Fine-Tuning [49.751529745537546]
  We propose DualOpt, a novel approach that decouples optimization techniques specifically tailored for training from scratch.<n>For training from scratch, we introduce real-time layer-wise weight decay, designed to enhance both convergence and generalization.<n>We extend the layer-wise weight decay to dynamically adjust the rollback levels across layers, adapting to the varying demands of different downstream tasks.
  arXiv  Detail & Related papers  (2026-04-21T06:27:18Z)
• Adaptive Domain Models: Bayesian Evolution, Warm Rotation, and Principled Training for Geometric and Neuromorphic AI [0.0]
  This paper develops an alternative training architecture grounded in three prior results.<n>We introduce Bayesian distillation, a mechanism by which the latent prior structure of a general-purpose model is extracted through the ADM training regime.<n>For deployment, we introduce warm rotation, an operational pattern in which an updated model transitions into an active inference pathway without service interruption.
  arXiv  Detail & Related papers  (2026-03-18T12:36:19Z)
• Efficient Real-Time Adaptation of ROMs for Unsteady Flows Using Data Assimilation [7.958594167693376]
  We propose an efficient retraining strategy for a parameterized Reduced Order Model (ROM)<n>The strategy attains accuracy comparable to full retraining while requiring only a fraction of the computational time.<n>We show that, for the dynamical system considered, the dominant source of error in out-of-sample forecasts stems from distortions of the latent manifold.
  arXiv  Detail & Related papers  (2026-02-26T16:43:28Z)
• SaRA: High-Efficient Diffusion Model Fine-tuning with Progressive Sparse Low-Rank Adaptation [52.6922833948127]
  In this work, we investigate the importance of parameters in pre-trained diffusion models.<n>We propose a novel model fine-tuning method to make full use of these ineffective parameters.<n>Our method enhances the generative capabilities of pre-trained models in downstream applications.
  arXiv  Detail & Related papers  (2024-09-10T16:44:47Z)
• Solving Inverse Problems with Model Mismatch using Untrained Neural Networks within Model-based Architectures [14.551812310439004]
  We introduce an untrained forward model residual block within the model-based architecture to match the data consistency in the measurement domain for each instance. Our approach offers a unified solution that is less parameter-sensitive, requires no additional data, and enables simultaneous fitting of the forward model and reconstruction in a single pass.
  arXiv  Detail & Related papers  (2024-03-07T19:02:13Z)
• Symplectic Autoencoders for Model Reduction of Hamiltonian Systems [0.0]
  It is crucial to preserve the symplectic structure associated with the system in order to ensure long-term numerical stability. We propose a new neural network architecture in the spirit of autoencoders, which are established tools for dimension reduction. In order to train the network, a non-standard gradient descent approach is applied.
  arXiv  Detail & Related papers  (2023-12-15T18:20:25Z)
• Precision-Recall Divergence Optimization for Generative Modeling with GANs and Normalizing Flows [54.050498411883495]
  We develop a novel training method for generative models, such as Generative Adversarial Networks and Normalizing Flows. We show that achieving a specified precision-recall trade-off corresponds to minimizing a unique $f$-divergence from a family we call the textitPR-divergences. Our approach improves the performance of existing state-of-the-art models like BigGAN in terms of either precision or recall when tested on datasets such as ImageNet.
  arXiv  Detail & Related papers  (2023-05-30T10:07:17Z)
• Extrapolation for Large-batch Training in Deep Learning [72.61259487233214]
  We show that a host of variations can be covered in a unified framework that we propose. We prove the convergence of this novel scheme and rigorously evaluate its empirical performance on ResNet, LSTM, and Transformer.
  arXiv  Detail & Related papers  (2020-06-10T08:22:41Z)

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.