Transformers have been established as the most popular backbones in sequence modeling, mainly due to their effectiveness in in-context retrieval tasks and the ability to learn at scale. Their quadratic memory and time complexity, however, bound their applicability in longer sequences and so has motivated researchers to explore effective alternative architectures such as modern recurrent neural networks (a.k.a long-term recurrent memory module). Despite their recent success in diverse downstream tasks, they struggle in tasks that requires long context understanding and extrapolation to longer sequences. We observe that these shortcomings come from three disjoint aspects in their design: (1) limited memory capacity that is bounded by the architecture of memory and feature mapping of the input; (2) online nature of update, i.e., optimizing the memory only with respect to the last input; and (3) less expressive management of their fixed-size memory. To enhance all these three aspects, we present Atlas, a long-term memory module with high capacity that learns to memorize the context by optimizing the memory based on the current and past tokens, overcoming the online nature of long-term memory models. Our experimental results on language modeling, common-sense reasoning, recall-intensive, and long-context understanding tasks support the effectiveness of Atlas compared to other modern recurrent neural networks.
论文检索
输入标题、作者或关键词，从 13,033 篇学术成果中精准定位
Deep Learning · Generative Models and Autoencoders
The two-sample testing problem, a fundamental task in statistics and machine learning, seeks to determine whether two sets of samples, drawn from underlying distributions $p$ and $q$, are in fact identically distributed (i.e.~whether $p=q$). A popular and intuitive approach is the classifier two-sample test (C2ST), where a classifier is trained to distinguish between samples from $p$ and $q$. Yet despite simplicity of the C2ST, its reliability hinges on access to a near-Bayes-optimal classifier, a requirement that is rarely met and difficult to verify. This raises a major open question: can a weak classifier still be useful for two-sample testing? We show that the answer is a definitive yes. Building on the work of Hu & Lei (2024), we analyze two conformal variants of the C2ST that convert the scores from any trained classifier---even if weak, biased, or overfit---into exact, finite-sample p-values. We establish two key theoretical properties of the conformal C2ST: (i) finite-sample Type-I error control, and (ii) non-trivial power that degrades gently in tandem with the error of the trained classifier. The upshot is that even poorly performing classifiers can yield powerful and reliable two-sample tests. This general framework finds a powerful application in Bayesian inference, particularly for validating Neural Posterior Estimation (NPE) models, where the task of comparing a learned posterior approximation $q(\theta \mid y)$ to the true posterior $p(\theta \mid y)$ can be framed as a two-sample test. Empirically, the Conformal C2ST outperforms classical discriminative tests across a wide range of benchmarks for this task. Our results establish the conformal C2ST as a practical, theoretically grounded diagnostic tool.
Applications · Chemistry, Physics, and Earth Sciences
The high cost and data scarcity in scientific exploration have motivated the use of large language models (LLMs) as knowledge-driven components in Bayesian optimization (BO). However, existing approaches typically embed LLMs directly into the sampling or surrogate modeling pipeline, without fully leveraging their significantly lower evaluation cost compared to real-world experiments. To address this limitation, we propose LLM-Accelerated Bayesian Optimization (LABO), a framework that combines LLM predictions with experimental observations within a single BO loop. LABO employs a gating mechanism to dynamically balance reliance on LLM predictions versus actual experiments. By leveraging inexpensive LLM evaluations to broadly explore the search space and reserving costly real experiments only for regions with high uncertainty, LABO achieves more sample-efficient optimization. We provide a theoretical analysis with a cumulative regret bound that formalizes this efficiency gain. Empirical results across diverse scientific tasks demonstrate that LABO consistently outperforms existing methods under identical experimental budgets. Our results suggest that LABO offers a practical and theoretically grounded approach for integrating LLMs into scientific discovery workflows.
Applications · Time Series
Effective time series forecasting enables various real-world applications, benefiting from the proliferation of mobile devices. However, the volume of time series data may vary significantly across domains due to low sampling rates and data regulations. To maximally create value from sparse data, this study focuses on a new problem of source-free time series forecasting, aiming to adapt a pretrained model from sufficient source time series to the sparse target time series without access to the source data, enabling data protection. To achieve this, we propose TimeID, a novel source-free time series forecasting framework with a large language model (LLM) centric proxy denoising inspired by the powerful generalization capabilities of LLMs. Specifically, TimeID consists of three key components: (1) dual-branch invariant disentangled feature learning that enforces representation- and gradient-wise invariance by means of season-trend decomposition; (2) lightweight, parameter-free proxy denoising that dynamically calibrates systematic biases of LLMs; and (3) knowledge distillation that bidirectionally aligns the denoised prediction and the original target prediction. Extensive experiments on real-world datasets demonstrate that TimeID outperforms state-of-the-art baselines, improving MSE and MAE by 10.7\% and 9.3\% on average. The code is publicly available at https://anonymous.4open.science/r/TimeID-6D1D/.
Applications · Chemistry, Physics, and Earth Sciences
Inferring cellular trajectories from destructive snapshots is complicated by the challenges of stochasticity and non-conservative mass dynamics such as cell proliferation and apoptosis. Existing unbalanced Optimal Transport (OT) methods treat mass as a continuous fluid, performing inference at the population level. However, this macroscopic view often fails to capture the discrete, jump-like nature of birth-death events at single-cell resolution, which is essential for understanding lineage branching and fate decisions. We present **Unbalanced Schrödinger Bridge (USB)**, a simulation-free framework for learning underlying dynamics that effectively integrates both stochastic and unbalanced effects which also models the discrete, jump-like birth–death dynamics at single-cell resolution. Theoretically, USB provides a tractable solution to the Branching Schrödinger Bridge (BSB) problem, offering a rigorous microscopic interpretation where individual cells undergo both Brownian motion and discrete birth-death jumps. Technically, the method implements an efficient solver by introducing a simulation-free training objective that effectively scales to high-dimensional omics data. Empirically, we demonstrate on both simulated and real-world datasets that USB not only achieves trajectory reconstruction performance better than or comparable to deterministic baselines but also uniquely enables realistic discrete simulation of birth-death dynamics at single-cell resolution.
Deep Learning · Large Language Models
Large language models (LLMs) are increasingly deployed as automated evaluators that assign numeric scores to model outputs, a paradigm known as LLM-as-a-Judge. However, standard Reinforcement Learning (RL) methods typically rely on binary rewards (e.g., 0-1 accuracy), thereby ignoring the ordinal structure inherent in regression tasks; for instance, they fail to recognize that predicting 4 is significantly better than predicting 1 when the ground truth is 5. Conversely, existing regression-aware approaches are often confined to Supervised Fine-Tuning (SFT), limiting their ability to explore optimal reasoning paths. To bridge this gap, we propose REAL (Regression-Aware Reinforcement Learning), a principled RL framework designed to optimize regression rewards, and also proven to be optimal for correlation metrics. A key technical challenge is that the regression objective is explicitly policy-dependent, thus invalidating standard policy gradient methods. To address this, we employ the generalized policy gradient estimator, which naturally decomposes optimization into two complementary components: (1) exploration over Chain-of-Thought (CoT) trajectory, and (2) regression-aware prediction refinement of the final score. Extensive experiments across model scales (8B to 32B) demonstrate that REAL consistently outperforms both regression-aware SFT baselines and standard RL methods, exhibiting significantly better generalization on out-of-domain benchmarks. On Qwen3-32B specifically, we achieve gains of +8.40 Pearson and +7.20 Spearman correlation over the SFT baseline, and +18.30/+11.20 over the base model. These findings highlight the critical value of integrating regression objectives into RL exploration for accurate LLM evaluation.
Transformers have been established as the de-facto backbones for most recent advances in sequence modeling, mainly due to their growing memory capacity that scales with the context length. While plausible for retrieval tasks, it causes quadratic complexity and so has motivated recent studies to explore viable subquadratic recurrent alternatives. Despite showing promising preliminary results in diverse tasks, such recurrent architectures underperform Transformers in recall-intensive tasks, {often attributed to their fixed-size memory. In this paper, we introduce Memory Caching (MC), a simple yet effective technique that enhances recurrent models by caching checkpoints of their memory states (a.k.a. hidden states). Memory Caching allows the effective memory capacity of RNNs to grow with sequence length, offering a flexible trade-off that interpolates between the fixed memory ( $O(L)$ complexity) of RNNs and the growing memory ( $O(L^2)$ complexity) of Transformers. We propose four variants of MC, including gated aggregation and sparse selective mechanisms, and discuss their implications on both linear and deep memory modules.} Our experimental results on language modeling, and long-context understanding tasks show that MC enhances the performance of recurrent models, supporting its effectiveness. In in-context recall tasks, our results indicate that while Transformers still achieve the best performance, our MC variants show competitive performance, close the gap with Transformers, and performs better than state-of-the-art recurrent models.
Social Aspects · Privacy
Large language models (LLMs) are trained on vast datasets that may contain sensitive information. Differential privacy (DP), the de facto standard for formal privacy guarantees, provides a principled framework for training LLMs with provable privacy protection. However, state-of-the-art DP training implementations rely on *fast gradient clipping* techniques with memory overhead $O(B\min(T^2, d^2))$, where $B$ is the batch size, $T$ is the sequence length, and $d$ is the layer width. This becomes prohibitive as both model width and context length grow. We propose DP-SGD-RC, a novel variant of DP-SGD with *randomized clipping* that reduces memory and compute overhead. DP-SGD-RC leverages *stochastic trace estimation* methods, specifically *Hutchinson's estimator* and its improved variant, Hutch$^{++}$, to reduce the memory footprint of per-sample gradient norm estimation. We provide a tight privacy analysis showing that DP-SGD-RC achieves noise multipliers competitive with deterministic clipping. Experiments fine-tuning Llama 3.2 1B on long-context benchmarks spanning classification, question answering, and summarization tasks demonstrate that DP-SGD-RC matches baseline utility while significantly reducing memory and compute.
Applications · Chemistry, Physics, and Earth Sciences
Regional climate prediction presents unique challenges for time series foundation models, which typically process temporal patterns through a single-pass inference. Expert climatologists, in contrast, employ multi-scale temporal analysis and iterative refinement based on systematic error diagnosis. We present RGMR (Residual-Guided Multi-Resolution Refinement), an inference-time framework that adapts pre-trained foundation models to perform structured multi-scale reasoning for climate forecasting without parameter modification. Our approach combines hierarchical coarse-to-fine prediction refinement with residual-guided error correction; together, they systematically address prediction failures at each resolution level. Applied to drought forecasting using the Standardized Precipitation Evapotranspiration Index (SPEI), RGMR consistently enhances foundation model performance across diverse climate regions within an Australian regional area. Experimental results demonstrate substantial improvements over direct foundation model application, achieving up to 18.9\% reduction in mean squared error, 10.2\% reduction in root mean squared error, and 21.1\% relative gain in $R^2$ when applied to TimesFM, with the largest benefits observed in climatologically complex regions where multi-scale temporal dynamics are most pronounced. The framework's inference-time operation enables immediate deployment on existing operational climate prediction systems without model retraining, offering a practical solution for enhancing foundation model capabilities in specialized forecasting domains.
Deep Learning · Large Language Models
Activation steering promises to be an extremely parameter-efficient form of adaptation, but its effectiveness depends on critical design choices---such as intervention location and parameterization---that currently rely on empirical heuristics rather than a principled foundation. We establish a first-order equivalence between activation-space interventions and weight-space updates, deriving the conditions under which activation steering can replicate fine-tuning behavior. This equivalence yields a principled framework for steering design and identifies the post-block output as a theoretically-backed and highly expressive intervention site. We further explain why certain intervention locations outperform others and show that weight updates and activation updates play distinct, complementary functional roles. This analysis motivates a new approach---joint adaptation---that trains in both spaces simultaneously. Our post-block steering method achieves accuracy within 0.4%-1.5% of full-parameter tuning while training only $0.04% of model parameters, consistently outperforming prior activation steering methods such as ReFT and PEFT approaches including LoRA while using significantly fewer parameters. Finally, we show that joint adaptation often surpasses the performance ceilings of weight and activation updates in isolation, introducing a new paradigm for efficient model adaptation.
Multi-modal prompt learning is a parameter-efficient approach to adapt large vision--language models to downstream classification tasks. However, prompts can inadvertently evolve into a high-capacity pathway encoding environment-dependent spurious correlations that are only predictive in the source domain, thereby undermining transferability. To address this issue, this paper introduces \textbf{Do-Prompt}, a \emph{compress-and-intervene} framework that brings together variational bottlenecks and causal interventions for robust prompt tuning. We model prompts as stochastic latent variables and impose a \emph{variational prompt bottleneck} to explicitly regulate the information transmitted through prompts, effectively mitigating their propensity to memorize spurious nuisance cues. Building on this capacity constraint, we propose lightweight \emph{prompt-level interventions} by perturbing the environment-related prompt components and enforcing prediction consistency under these \textit{do}-style perturbations. This synergistic integration encourages reliance on task-stable, invariant semantics rather than spurious prompt content. Notably, Do-Prompt is plug-and-play compatible with existing multi-modal prompt tuning pipelines with negligible computation overhead. Extensive experiments on base-to-novel generalization, cross-dataset transfer, and ImageNet distribution shifts demonstrate consistent performance gains, with particularly notable improvements on datasets exhibiting pronounced domain or texture biases.
Deep Learning · Large Language Models
Large language models (LLMs) exhibit exceptional general language processing capabilities, but their memory and compute costs hinder deployment. Ternarization has emerged as a promising compression technique, offering significant reductions in model size and inference complexity. However, existing methods struggle with heavy-tailed activation distributions and therefore keep activations in high precision, fundamentally limiting end-to-end inference acceleration. To overcome this limitation, we propose **TWLA** (**T**ernarized **W**eights and **L**ow-bit **A**ctivations), a post-training quantization (PTQ) framework that achieves 1.58-bit weight compression and 4-bit activation quantization while maintaining high accuracy. TWLA comprises three components: (1) Euclidean-to-Manifold Asymmetric Ternary Quantizer (E2M-ATQ) minimizes layer-output error under weight ternarization via a two-stage optimization from Euclidean initialization to manifold relocation; (2) Kronecker Orthogonal Tri-Modal Shaping (KOTMS) applies a Kronecker-structured orthogonal rotation to reshape weights into ternary-friendly tri-modal distributions, while the shared rotation statistically suppresses activation outliers; and (3) Inter-Layer Aware Activation Mixed Precision (ILA-AMP) explicitly introduces adjacent-layer second-order interaction costs in bit allocation and jointly optimizes for the layer-wise disparity of activation quantization gains induced by the shared orthogonal transform, preventing cascades triggered by a few weak layers. Extensive experiments demonstrate that TWLA is a PTQ method that maintains high accuracy under the **W1.58A4** configuration, while delivering significant inference acceleration. The code is available at [TWLA](https://anonymous.4open.science/r/TWLA-0212/).
Applications · Chemistry, Physics, and Earth Sciences
Accurately modeling the macroscopic dynamics of high-dimensional microscopic systems is of broad interest across the sciences. Many data-driven approaches learn a low-dimensional latent state through an autoencoder trained for pointwise input reconstruction. These methods typically assume a fixed ordering of microscopic degrees of freedom in the input. However, in many settings such as particle systems the microscopic state is inherently unordered. This motivates an autoencoder framework that learns permutation-invariant latent representations. To this end, we adopt a permutation-invariant encoder and design the decoder to reconstruct the mass distribution centered at the observed points rather than per-sample reconstruction. We then jointly learn the macroscopic dynamics of the observables together with the latent states. We demonstrate the effectiveness and robustness of the proposed method across a range of microscopic settings, including learning the energy in interacting particle systems, predicting mixing dynamics in Lennard–Jones fluids, and modeling the stretching dynamics from videos of a more realistic polymer system.
Deep Learning · Large Language Models
Reinforcement Learning from Human Feedback (RLFH) or Verifiable Rewards (RLVR) are two key steps in the post-training of modern Language Models (LMs). A common problem is reward hacking, where the policy may exploit inaccuracies of the reward and learn an unintended behavior. Most previous works address this by limiting the policy update with a Kullback-Leibler (KL) penalty towards a reference model. We propose a different framing: Train the LM in a way that biases policy updates towards regions in which the reward is more accurate. First, we derive a theoretical connection between the accuracy of a reward model and the flatness of an optimum at convergence. Gradient regularization (GR) can then be used to bias training to flatter regions and thereby maintain reward model accuracy. We confirm these results by showing that the gradient norm and reward accuracy are empirically correlated in RLHF. We then show that Reference Resets of the KL penalty implicitly use GR to find flatter regions with higher reward accuracy. We further improve on this by proposing to use explicit GR with an efficient finite-difference estimate. Empirically, GR performs better than a KL penalty across a diverse set of RL experiments with LMs. GR achieves a higher GPT-judged win-rate in RLHF, avoids overly focusing on the format in rule-based math rewards, and prevents hacking the judge in LLM-as-a-Judge math tasks.
Deep Learning · Algorithms
Post-training quantization (PTQ) is a promising approach to reducing the storage and computational requirements of large language models (LLMs) without additional training cost. Recent PTQ studies have primarily focused on quantizing only weights to sub-$8$-bits while maintaining activations at $8$-bits or higher. Accurate sub-8-bit quantization for both weights and activations without relying on quantization-aware training remains a significant challenge. We propose a novel quantization method called block clustered quantization (BCQ) wherein each operand tensor is decomposed into blocks (a block is a group of contiguous scalars), blocks are clustered based on their statistics, and a dedicated optimal quantization codebook is designed for each cluster. As a specific embodiment of this approach, we propose a PTQ algorithm called Locally-Optimal BCQ (LO-BCQ) that iterates between the steps of block clustering and codebook design to greedily minimize the quantization mean squared error. When weight and activation scalars are encoded to W4A4 format (with $0.5$-bits of overhead for storing scaling factors and codebook selectors), we advance the current state-of-the-art by demonstrating $<1$\% loss in inference accuracy across several LLMs and downstream tasks.
Deep Learning · Large Language Models
Despite the growing use of large language models (LLMs) in subjective tasks such as role-playing, humor, emotional intelligence, and dialogue quality, their evaluation faces a pressing reproducibility crisis: even the same evaluator may contradict itself when re-judging the exact same sample. We attribute this instability to dimension drift, where free-form evaluation protocols (e.g., Chain-of-Thought reasoning) unpredictably shift their implicit criteria, undermining reliability. To address this fundamental challenge, we reformulate subjective evaluation as an information-theoretic optimization problem. Specifically, we propose an **Expected Information Gain (EIG)-based framework** that constructs a stable yet adaptive personalized rubric to eliminate dimension drift. Our two-stage “generate–then–score” design first produces a diverse pool of candidate evaluation questions and then selects the most informative subset via EIG, yielding explicit and repeatable criteria. Experiments on six benchmarks, including CharacterEval, The rJokes, and MT_bench, demonstrate that our approach substantially improves both evaluation consistency and alignment with human judgments, outperforming CoT-based and fixed-questionnaire baselines. These results highlight that information-theoretic questionnaire construction offers a principled and reliable path toward reproducible evaluation of subjective tasks.
Theory · Learning Theory
Neural collapse (NC) describes the structured geometry that emerges in the features and weights of trained classifiers. Recent theory suggests NC can be suboptimal in deep architectures, attributing this to an explicit low-rank bias from L2 regularization. We study the deep unconstrained feature model (UFM)—equivalent to a deep linear network with orthogonal inputs—trained without regularization, to isolate how gradient descent and depth alone shape NC. We show that depth induces an implicit low-rank bias: low-rank matrices propagate norm more efficiently through successive multiplications, promoting low-rank alternatives to NC. These alternatives, we argue, correspond to softmax codes: max-margin solutions previously found in width-bottlenecked networks. Analyzing training dynamics under spectral initialization, we identify an early-time repulsion among singular values that drives low-rank emergence, and characterize how depth shrinks NC's basin of attraction. Finally, we show that some effects act in the opposite direction: for randomly initialized networks, increasing width biases training toward higher-rank solutions. Our results provide the first asymptotic and dynamic characterization of implicit bias in deep UFMs trained with unregularized multiclass cross-entropy.
Probabilistic Methods · Gaussian Processes
Nonlinear conservation laws are at the heart of many of the most important dynamical systems in science and engineering. In practical applications, such systems are often subject to various sources of uncertainty, e.g. due to sparse or noisy measurements. Inferring physical quantities and fields of interest then becomes an ill-posed problem which both classical numerical methods and modern deep learning-based methods struggle to treat appropriately. Recent work has framed classical numerical methods as Bayesian inference under Gaussian process priors, resulting in a physics-aware treatment of uncertainties. Following this line of work, we develop a novel numerically conservative method for uncertainty-aware simulations of nonlinear conservation laws. Our method uses recent sparse approximation techniques to scale up to large-scale forward and inverse problems. For forward simulation, we match the accuracy of classical solvers while providing structurally meaningful uncertainty. On inverse problems, we recover posteriors over nonparametric source fields in seconds --- outperforming neural baselines that take minutes to produce a less accurate point estimate.
Optimization · Large Scale, Parallel and Distributed
Quantum computation offers the potential for a significant constant-factor speedup for the Ordered Search Problem (OSP). A classical construction is the $k$-query quantum ordered search algorithm, which can exactly search an $N$-element ordered list and achieves a query complexity improvement of a factor of $\frac{k}{\log_2 N}$. For larger $k$, stronger constant-factor improvements could be obtained by finding the largest admissible list size $N^\star$, a task that can be formulated as a structured semidefinite program (SDP). However, solving this SDP becomes computationally intractable beyond $k=6$, as existing CPU and GPU solvers rely on explicit construction of prohibitively large constraint matrices. In this paper, we introduce a matrix-free GPU SDP framework that evaluates the highly structured constraints in OSP on-the-fly using custom CUDA kernels, reducing memory complexity from quadratic to linear and shifting the bottleneck from memory to computation. Using this approach, we tightly bracket the optimal list size for $k=6$ as $90,000 \le N^\star < 94,000$, improving the best known upper bound on the query coefficient from $0.390$ to $0.365$. We further certify these results by constructing rigorous dual infeasibility certificates via matrix-free minimum-eigenvalue estimation.
Hybrid sequence models—combining Transformer and state-space model layers—seek to gain the expressive versatility of attention as well as the computational efficiency of state-space model layers. Despite burgeoning interest in hybrid models, we lack a basic understanding of the settings where—and underlying mechanisms through which—they offer benefits over their constituent models. In this paper, we study this question, focusing on a broad family of core synthetic tasks. For this family of tasks, we prove the existence of fundamental limitations for non-hybrid models. Specifically, any Transformer or state-space model that solves the underlying task requires either a large number of parameters or a large working memory. On the other hand, for two prototypical tasks within this family—namely selective copying and associative recall—we construct hybrid models of small size and working memory that provably solve these tasks, thus achieving the best of both worlds. Our experimental evaluation empirically validates our theoretical findings. Importantly, going beyond the settings in our theoretical analysis, we empirically show that learned—rather than constructed—hybrids outperform non-hybrid models with up to $6 \times$ as many parameters. We additionally demonstrate that hybrid models exhibit stronger length generalization and out-of-distribution robustness than non-hybrids.