跳到正文
研踪PaperTrail

论文库

检索计算机领域论文,按研究方向、作者或关键词筛选。

部分采集异常
正在读取账号与研究记录…

全站更新概况

+479新增论文
+361其中 AI 相关
+64已有论文修订
+95新增 AI 导读

arXiv 最近完成扫描:2026-11-16 01:34 · 全站最近入库:2026-11-16 21:56各信源进度 →

统计口径与数据新鲜度

统计 2026-11-16 02:00 — 2026-11-17 02:00(北京时间) · 点击数字查看对应简报;AI 相关包含在新增论文内。

最近完成检查:2026-11-16 01:34最近发现新增版本:2026-11-16 21:56库内论文最新更新时间:2026-11-14 02:59

每分钟刷新已保存的数据。空检查不代表采集异常;新收录可能包含历史论文。AI 导读按新增版本计数。

23,827 已收录论文19,941 AI 相关6,922 有修订112 HF 社区发现

浏览论文库

时间范围跟随排序:版本更新、本站收录或首次发表。最新收录可能包含历史论文。部分信源异常,查看状态 →

找到 291 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-16 01:33(北京时间)
新论文 · v1arXiv:2611.18553

WeightBench:评估语言模型作为权重空间程序员的能力

WeightBench: Evaluating Language Models as Weight-Space Programmers

Aayush Marasini, Barsat Khadka, Nick Rahimi

AI 速读 基于摘要

作者提出 WEIGHTBENCH 基准,用 42 个任务、五个方向考察大模型能否直接读写和修改其他神经网络的权重,而不依赖梯度下降。

cs.LG · cs.AI

查看原文摘要

Recent work shows that large language models (LLMs) can operate directly on the parameters of other neural networks. They can read what a network was trained to do, fix a network that fails, and write working networks from scratch, all without gradient descent. This capability has been termed Weight-Space Programming (WSP). Exploring this new paradigm systematically requires a robust benchmark. We introduce WEIGHTBENCH, a benchmark that evaluates LLMs of increasing capability on 42 WSP tasks across five tracks: reading, fixing and writing weights, recognizing edits that are impossible, and data-free compression. The networks whose weights the LLMs read and edit range from 2K-parameter control policies to 7B-parameter language models. Experimental benchmark results show that frontier agents can already read small networks and write working control policies without ever running them, but cannot yet patch language models, compress them without data, or tell an impossible edit from a hard one. On all 42 tasks, Claude Opus 5.5 reaches a macro score of 0.535 (Sonnet 5.5 0.466, Haiku 5.5 0.402, DeepSeek v4-flash 0.215); on the 38 tasks it was run on, GPT-6.1-sol reaches 0.568 against Opus's 0.546. Code, the task registry, our OS-level sandbox harness, and the live leaderboard are publicly available at https://github.com/Aayush-Marasini/WeightBench.

新论文 · v1arXiv:2611.18542

椭圆采样:高斯及更一般分布的最优插值调度

Elliptic Sampling: Optimal Interpolant Schedules for Gaussians and Beyond

Cole Becker, Taylan Kargin, Kimia Hassibi, Asuman Ozdaglar, Pablo A. Parrilo

AI 速读 基于摘要

基于复化ODE轨迹解析性,提出椭圆调度与Zoloflow采样算法,在高斯目标下达到极小极大最优,并在对数凹情形给出复杂度保证。

stat.ML · cs.AI · cs.LG

查看原文摘要

Generative modeling based on stochastic interpolants requires specifying two key choices: an interpolant schedule bridging the initial and target distributions, and an efficient discretization of the resulting probability-flow ODE. In this work, we introduce a new framework for joint schedule/discretization design based on the analyticity properties of the (complexified) ODE trajectories, which in turn ensure good numerical behavior. Our formulation can be solved exactly via conformal mapping, yielding a natural family of variance-preserving elliptic schedules, parametrized by Jacobi elliptic functions. Pairing an elliptic schedule with a tailored discretization, we obtain Zoloflow, a new sampling algorithm closely connected to Zolotarev's optimal rational function approximants. Crucially, both the elliptic schedule and Zoloflow explicitly incorporate the condition number $κ$ of the target distribution and reduce to the classical trig schedule in the limit $κ\rightarrow 1$. For Gaussian targets, Zoloflow achieves minimax optimality among all oblivious score-span algorithms, attaining total-variation error $\varepsilon$ using $Θ(\logκ\,\log(\sqrt d/\varepsilon))$ (exact) score queries. For the log-concave case, the elliptic schedule allows sampling with at most $\varepsilon$ 2-Wasserstein-error by requiring merely $\widetilde{O}(\sqrt{d} \, \mathrm{polylog}(κ, 1/\varepsilon))$ score evaluations. Complementing these theoretical results, we empirically evaluate the elliptic schedule on general (non-log-concave) target distributions, including detailed comparisons with alternative schedule proposals on both synthetic benchmarks and existing image generation datasets.

新论文 · v1arXiv:2611.18480

正则指数族中的Bregman几何:对偶表示与相对光滑性

Bregman geometry in regular exponential families: dual representations and relative smoothness

Mame Diarra Fall, Éric Barat, Nicolas Dobigeon

AI 速读 基于摘要

该论文在正则指数族中构建等价的Bregman对偶表示,并利用这些表示建立相对光滑性的参考几何与显式界。

stat.ML · cs.LG · math.OC

查看原文摘要

Regular exponential-family distributions are ubiquitous in statistics due to their appealing and convenient properties. Notably, they can be expressed in terms of Bregman divergences, revealing fundamental connections between statistics, convex analysis, and information geometry. However, this geometric structure has not been systematically exploited to characterize these distributions with respect to their expectation parametrization. Building on a projective characterization, we establish a general construction of equivalent dual Bregman representations, together with normalization and uniqueness results. These representations express log-densities as ordinary Bregman divergences in dual coordinate systems, providing geometric characterizations of the parameter space. So-called $β$- and $α$-scaled dual representations emerge as particular instances of this framework. We further derive dual Bregman representations of conjugate priors and establish that they preserve conjugacy. Applications to classical distributions illustrate the geometric structures underlying seemingly different statistical models. Moreover, we show that these representations provide natural reference geometries for establishing relative smoothness of the primal Bregman representation of the distribution, with explicit bounds that are sharp in the scalar and separable settings. An illustrative application to the Poisson distribution sheds new light on several works in the literature, providing complementary geometric justifications for reference geometries while deriving existing algorithms.

版本更新 · v2arXiv:2609.35348

从数据到程序:从经验数据快速直接生成程序推断

From Data to Program: Fast & Direct Generative Program Inference from Empirical Data

Simon Klüttermann, Xueying Ding, Leman Akoglu

AI 速读 基于摘要

PRODiGI 是一个预训练的数据到程序模型,单次前向传播即可从经验数据推断出显式可执行的生成程序,并支持程序空间微调。

cs.LG · cs.AI

查看原文摘要

Estimating probability densities from a finite set of samples typically requires dataset-specific model fitting. We introduce PRODiGI, a pretrained data-to-program model that infers an explicit, executable generative program in a single forward pass. Pretrained on synthetic datasets paired with their ground-truth programs, PRODiGI accommodates diverse generative families and data dimensionalities through template prediction and non-autoregressive program parameter decoding. Its inferred programs support direct sampling, density and score evaluation, and inspection independently of the pretrained model. We further introduce program-space fine-tuning, which refines differentiable program parameters by maximizing the likelihood of empirical data while keeping model parameters intact. Experiments show that PRODiGI-FT achieves better density and score metrics than existing pretrained models, while offering multi-fold speedups over its closest competitors. Program-space fine-tuning further reduces generation MMD by 90%. By turning empirical data into explicit, reusable programs, PRODiGI introduces a new direction for fast, interpretable tabular generative modeling.

版本更新 · v2arXiv:2611.01929

Watch Your Step: Adam's Margin Maximization Bias Depends on the Learning Rate Schedule

Eitan Gronich, Gal Vardi

原文摘要

We study the implicit bias of Adam and its similarity to sign gradient descent on homogeneous models in binary classification tasks. Perhaps surprisingly, we uncover that these depend on the learning rate schedule. While previous works studied strongly decaying learning rate schedules $η(t)=\mathrm o \left(t^{1/L-1}\right)$, where $L$ is the homogeneity degree, and showed $\ell_\infty$-margin-maximization biases, we focus on the more common complementary regime, including constant and weakly decaying schedules. For a polynomial schedule $t^α$ and homogeneity degree $L$, we show that when $α> \frac 1L - \frac12$, Adam has no trajectory-independent margin-maximization bias for deep linear networks (in the KKT sense): first, there is no non-trivial, locally Lipschitz, convex implicit regularizer in this setting, including any squared norm. Second, on width-$1$ networks, the result holds without requiring convexity, implying that any implicit regularizer for wide networks must be trivial when restricted to a width-$1$ network. Nevertheless, we positively characterize the implicit bias in this regime as maximizing a trajectory-dependent weighted-$\ell_2$ margin for any $L$-homogeneous model. The fundamental mechanism behind our results is convergence of Adam's preconditioner to a trajectory-dependent vector, which also allows showing that the limiting parameter direction depends on initialization. In the intermediate regime $\frac 1L - 1 \leq α\leq \frac1L - \frac12$ we prove an $\ell_\infty$ bias, conditional on gradient directional convergence to a full-support vector. Of independent interest, we establish a loss decay rate for full batch Adam on general homogeneous models, whose leading order depends only on the second-moment parameter.

cs.LG · stat.ML

新论文 · v1arXiv:2611.18163

On the Limits of Sufficient Explanations

Klaus Weinbauer, Peter F. Stadler, Thomas Gärtner, Sagar Malhotra

原文摘要

We formally characterize the conditions under which sufficient explanations exist for classifiers on structured data such as graphs and strings. We focus on post-hoc explainers that select parts of the input as sufficient explanations. An explanation is sufficient if the selected part alone ensures the predicted label. Sufficient explanations have been widely adopted across data modalities, including graphs, strings, and images. However, a unified formal analysis of sufficient explanations remains missing. We bridge this gap with four main contributions: (i) We define a general explanation formalism consisting of an instance space, a classifier, and a select relation. The select relation specifies which potential sub-instances or feature subsets can be selected as explanations from each instance; (ii) We show that the existence of sufficient explainers is fully characterized by the interplay between the classifier and the select relation; (iii) We extend this analysis to practically relevant variants of sufficiency, including probabilistic sufficiency and deletion sufficiency; (iv) Based on the formal analysis, we devise statistics that bound the maximum achievable sufficiency for a given task. We evaluate these statistics on language, protein, and molecular classification datasets and find that full sufficiency is unattainable on most of them. Since these bounds hold for every explainer, they provide a task-level reference against which any explainer's sufficiency can be judged in absolute terms.

cs.LG

版本更新 · v2arXiv:2603.09527

Efficiently Aligning Draft Models via Parameter- and Data-Efficient Adaptation

Luxi Lin, Zhihang Lin, Zhanpeng Zeng, Yuhao Chen, Qingyu Zhang, Jixiang Luo, Xuelong Li, Rongrong Ji

原文摘要

Speculative decoding accelerates LLM inference but suffers from performance degradation when target models are fine-tuned for specific domains. A naive solution is to retrain draft models for every target model, which is costly and inefficient. To address this, we introduce a practical parameter- and data-efficient framework named Efficient Draft Adaptation, abbreviated as EDA, for efficiently adapting draft models. EDA introduces three innovations: (1) a decoupled architecture that utilizes shared and private components to model the shared and target-specific output distributions separately, enabling parameter-efficient adaptation by updating only the lightweight private component; (2) a data regeneration strategy that utilizes the fine-tuned target model to regenerate training data, thereby improving the alignment between training and speculative decoding, leading to higher average acceptance length; (3) a sample selection mechanism that prioritizes high-value data for efficient adaptation. Our experiments show that EDA effectively restores speculative performance on fine-tuned models, achieving superior average acceptance lengths with significantly reduced training costs compared to full retraining. Code is available at https://github.com/Lyn-Lucy/Efficient-Draft-Adaptation.

cs.LG · cs.AI

新论文 · v1arXiv:2611.17886

Quantum Graph Neural Networks: Architectures, Trainability, and Applications

Zeinab Sharifi, Hossein Shirazi

原文摘要

Quantum graph neural networks (QGNNs) apply parametrized quantum circuits to graph-structured data, with the aim of overcoming limits of classical graph neural networks in scalability, expressivity, and long-range information flow. Since the first Hamiltonian-based model in 2019, the field has produced many architectures, applications, and claims, but the evidence behind them has not been assessed systematically. This survey introduces a unified formalism that describes every QGNN as an encoder, a graph-conditioned circuit, observables, and an optional classical head, together with a circuit-level condition for permutation equivariance and a resource model for comparing costs. We organize the literature into five architecture families (convolutional, equivariant, message-passing and recurrent, Hamiltonian, and hybrid) and compare them by qubit cost, symmetry, scalability, and evidence. We then review symmetry and geometric inductive biases (permutation, rotation, and Lorentz), the results on expressivity, trainability, noise, and scalability, and applications in high-energy physics, chemistry, finance, healthcare, earth observation, and transportation. An audit of datasets and evaluation practices leads to a minimum reporting protocol. Our main finding is that the design space is rich and symmetry is the best-supported inductive bias, whereas the evidence remains thin: results are mostly simulated, on small graphs, at parity with classical models or with fewer parameters, and no advantage has yet been established on tasks over classical graph data. We identify five open challenges and a research agenda covering provable advantage, equivariant and geometric designs, fault-tolerant implementations, and shared benchmarks. The coverage extends through 2026 and emphasizes recent work.

cs.LG

新论文 · v1arXiv:2611.17246

用大语言模型编程量子电路参数

Programming Quantum Circuit Parameters with Large Language Models

Tiago de Souza Farias, Caio Moraes Porto, Matheus da Silva Fonseca, Nicolas Armando Cabrera Carpi, Anderson de Araujo Oliveira, Celso Jorge Villas-Boas

AI 速读 基于摘要

研究冻结的通用大语言模型能否从符号化电路与哈密顿量描述直接生成变分量子电路参数,并考察问题结构、迭代反馈与可验证性的影响。

quant-ph · cs.LG

查看原文摘要

Large language models (LLMs) have recently demonstrated the ability to generate parameters for variational quantum circuits without conventional optimization. However, the conditions under which this approach is effective remain unclear. Here, we investigate a frozen, general-purpose LLM that generates complete circuit parameter vectors from symbolic circuit and Hamiltonian descriptions, with optional iterative energy feedback. We evaluate the approach on general random Hamiltonians, structured parent Hamiltonians, and molecular systems. For general Hamiltonians, LLM-generated parameters exhibit larger energy errors than conventional Adam optimization, while additional feedback generally provides limited improvement. In contrast, for structured Hamiltonians, the LLM can generate promising low-energy parameter configurations, suggesting that identifiable physical and mathematical structure facilitates parameter inference. We also show that direct LLM predictions of minimum energies can violate the Hamiltonian spectral lower bound, whereas generated circuit parameters allow independent evaluation of physically realizable energy expectation values. Finally, we investigate blind circuit interpretation and find that the LLM can predict measurement outcomes even when it fails to identify the underlying quantum algorithm. These findings highlight the role of problem structure, the limitations of iterative LLM optimization, and the importance of independently verifiable predictions in quantum parameter programming.

新论文 · v1arXiv:2611.17191

最大化LoRA适配器容量利用率以缓解学习-遗忘权衡

Maximizing Capacity Utilization of LoRA Adapters for Mitigating the Learning-Forgetting Trade-off

Tomohiro Sawada, Kartik Goyal

AI 速读 基于摘要

作者提出一种无需SVD的Rényi-2有效秩正则项,通过提升LoRA适配器的有效秩来改善学习与遗忘之间的权衡。

cs.LG · cs.CL

查看原文摘要

Low-Rank Adaptation (LoRA) often underperforms full fine-tuning, especially in heterogeneous and continual-learning settings in which substantial forgetting occurs. In general, it suffers from an undesirable learning-forgetting trade-off across diverse fine-tuning settings. We identify a potential contributor: LoRA adapters, especially high-rank ones, frequently fail to realize the available capacity afforded by the configured rank. Specifically, we find that standard LoRA updates have effective rank far below their configured rank, limiting how broadly the update energy is distributed. Therefore, we propose training mechanisms that optimize the original objective while maximizing adapter effective rank. Our main method adds a simple, stable, differentiable, and SVD-free Rényi-2 effective-rank regularizer to the language-modeling loss. We also connect capacity maximization to norm-bounded updates. In mixed-domain rank sweeps on three models evaluated with multiple random seeds, a penalty weight of $λ=0.01$ lowers mean Wiki perplexity at every tested rank of at least $16$ relative to unpenalized LoRA. Mean training perplexity improves or increases by at most $0.02$ in all but one model-rank comparison. Single-task and scheduling experiments further show that the balance between learning and retention depends on the configuration. We also observe that several LoRA variants in the literature also suffer from the identified capacity issue, and our approach provides a complementary axis of improvement to LoRA-based parameter-efficient fine-tuning. These results suggest that effective rank is not only a post-hoc diagnostic of adapter quality, but a trainable objective for improving the learning-forgetting frontier. Our code is available at https://github.com/tomohiro-sawada/lora-capacity-utilization

新论文 · v1arXiv:2611.17162

时间中的对称性:线性循环记忆的表示理论

Symmetry in Time: A Representation Theory of Linear Recurrent Memory

Anif N. Shikder, Ramit Dey, Jan Minac, Lyle E. Muller

AI 速读 基于摘要

作者用表示理论刻画线性循环状态空间模型的状态追踪能力,并提出引入四元数通道的 HyperSSM 以突破对角实/复 SSM 只能追踪交换群的限制。

cs.LG · cs.NE · math.RT

查看原文摘要

State space models (SSMs) are an emerging alternative to Transformers. SSMs offer linear-time inference and parallel training, but most implementations make the parallel scan efficient by making the recurrence diagonal: each channel of the hidden state is multiplied by an input-dependent real or complex number. These numbers commute, so the state records which inputs occurred but not their order. We formalize this with representation theory, which explains and fixes the state-tracking failures previously reported for SSMs. The state update defines a linear representation of the monoid of input sequences, and we show that an SSM tracks a group at every sequence length exactly when that group is a quotient of the group generated by its transition matrices. Two consequences follow. Diagonal real- and complex-valued SSMs can track only commutative groups, while the transition matrices of any SSM that can track a group must split into independent blocks over the real, complex, and quaternion numbers. We then introduce HyperSSM, which adds quaternion channels while keeping linear-time inference and parallel training. Trained only on final answers, HyperSSM solves the group tasks of recent state-tracking benchmarks exactly with at most five state numbers. This includes $S_5$ with 32 generators, where previous SSMs are at chance, and HyperSSM generalizes to sequences 25 times longer than those seen in training. Because the theory says what the transitions must encode, the learned weights of the linear recurrence can be interpreted directly. We demonstrate this by recovering the calibration of a real gyroscope, the rotations of imperfect qubit gates and the geometry of protein backbones.

新论文 · v1arXiv:2611.17121

用能量像素方法估计第二代DEL的规模以训练高分辨率结合模型

Using an energy pixels approach to estimate the size a 2nd Generation DEL must be to train a high-resolution binding model

Boris P. Belotserkovskii, R. Edward Watts

AI 速读 基于摘要

作者提出用能量像素模型,从朴素DEL的亲和力分布估计围绕命中物构建第二代DNA编码库所需复杂度,以支撑高分辨率结合预测模型的训练。

cs.LG · q-bio.BM

查看原文摘要

DNA-encoded libraries (DELs) are commonly used to find small-molecule binders to a target protein. But the hard part is no longer finding a hit. It is building a machine-learning model good enough to guide the medicinal chemistry that follows, one that can resolve the shift in affinity when a chemist changes a single atom. Such a model needs two things from its training data: dense sampling of the small patch of chemical space around the hit, and an adequate number of genuinely tight binders within it. A diverse naive DEL gives neither. It samples an enormous space, so any one neighborhood is covered far too thinly, and the tightest binders are rare. A model trained on that naive DEL data must extrapolate to binders tighter than any it has seen, and that is where models can fail. One way to get training data with both properties is a second-generation DEL built around the hit. We ask how complex that library must be. We treat binding as a sum of small, independent interactions, called pixels on the target and antipixels on the ligand. From the spread of affinities measured in the naive library, with no structural information and nothing yet synthesized, we estimate the complexity such a library must have to supply that dense sampling and those tight binders. The method does two useful things. It turns the number of tight binders a model needs into the number of compounds a library must contain. And it estimates the expected yield of a randomly designed second-generation DEL, so the gain from designing the library with machine learning or active learning can be measured against a baseline, without the expense of a control library as large as the experiment itself.

版本更新 · v3arXiv:2602.15919

无需重训模型评估逐样本成员推断脆弱性

Assessing Per-Sample Membership Inference Vulnerability without Retraining

Valentin Dorseuil, Jamal Atif, Olivier Capp{é}

AI 速读 基于摘要

作者提出一种仅需单个已训练模型、无需影子模型的逐样本成员推断攻击(MIA)脆弱性替代评分,并在实验中报告其优于损失与梯度范数基线。

stat.ML · cs.AI · cs.LG

查看原文摘要

Recent work in the privacy literature shows that sample-targeted membership inference attacks (MIAs) significantly outperform untargeted approaches by a wide margin. Motivated by this observation, we address the following question: can the privacy vulnerability of individual training points be assessed without training shadow models? We show that per-sample exposure to MIA is governed not only by a point's loss, but also by a data-dependent geometric measure. In the linear setting, we derive a closed-form decomposition of individual black-box MIA vulnerability into a population leverage score and a residual loss term, making explicit how sample-dependent geometry translates into privacy exposure. Since the final layer of most modern architectures is linear, we extend this framework to deep networks and propose a surrogate score operating on last-layer representations that requires only a single trained model and no shadow models. Empirical evaluations across diverse datasets and architectures show that our score outperforms loss and gradient-norm baselines at identifying the highest-risk points under state-of-the-art attacks, providing a computationally efficient and theoretically grounded tool for per-sample privacy risk assessment.

新论文 · v1arXiv:2611.17042

可验证人工智能的基础设施:证明成本去向与硬件如何助力

Infrastructure for Verifiable AI: Where Proof Costs Go and How Hardware Can Help

Yihan Xia, Wenxin Xu, Taotao Wang

AI 速读 基于摘要

作者提出四层分析框架并补充六种证明系统与七类内核的实测,指出可验证AI的证明成本瓶颈随证明系统与输入长度变化,硬件加速需围绕目标工作负载的成本画像来设计。

cs.AR · cs.CR · cs.LG

查看原文摘要

Users of artificial intelligence services, including those powered by large language models and other transformer-based models, need assurance that inference and training follow the declared computation without exposing private models or data. Zero-knowledge proofs can provide this assurance, but proof generation remains expensive. Hardware acceleration research often reports gains for individual cryptographic operations, leaving their effect on the full cost of proving neural-network workloads, particularly large transformer models, unclear. We introduce a four-layer framework linking verifiable AI workloads, proof systems, cryptographic kernels and hardware platforms, use it to review the literature, and add original measurements of six provers and seven kernels to identify which costs matter in practice. Our measurements show that the dominant cost changes with the proof system and input length. In low-rank fine-tuning, the cost of opening commitments to fixed model weights remains nearly constant across input lengths from 64 to 2048, while the work associated with intermediate values grows. A 13-billion-parameter fine-tuning configuration exceeds the capacity of a 40-gigabyte graphics processor but completes on an 80-gigabyte device, showing why capacity must be assessed alongside execution time. Our kernel profiling further identifies instruction issue and synchronization limits that a bandwidth-only model misses. We use these findings to assess graphics processors, reconfigurable logic, custom chips and existing artificial intelligence accelerators. The resulting research agenda connects proof-friendly numeric formats, reuse of fixed-weight commitments, heterogeneous execution and shared benchmarks to the measured bottlenecks. Effective acceleration requires selecting hardware and proof-system optimizations around the cost profile of the target workload.

版本更新 · v3arXiv:2509.03244

FoMEMO: Towards Foundation Models for Expensive Multi-objective Optimization

Yiming Yao, Fei Liu, Liang Zhao, Xi Lin, Yilu Liu, Qingfu Zhang

原文摘要

Expensive multi-objective optimization is a prevalent and crucial concern in many real-world scenarios, where sample-efficiency is vital due to the limited evaluations to recover the true Pareto front for decision making. Existing works either involve repeatedly fitting Gaussian process surrogates from scratch for each newly encountered problem, or rely on computationally demanding hypervolume-oriented policy learning for amortized optimization, making it challenging to achieve scalable pre-training and efficient yet robust adaptation for diverse emerging real-world applications. To address these challenges, we propose FoMEMO (Foundation Models for Expensive Multi-objective Optimization), which adopts a decomposition-based training paradigm that converts multi-objective optimization into preference-wise aggregated posterior learning tasks, enabling scalable pre-training on hundreds of millions of diverse synthetic datasets without relying on extensive real-world domain experiments. At test time, given observed trajectories from unseen problems and user preferences, the foundation model performs in-context posterior prediction and optimizes the derived acquisition functions for efficient candidate generation, achieving strong generalization and optimization performance without any subsequent model training or updates.

cs.LG · cs.AI

版本更新 · v4arXiv:2609.39329

PatchKV: Weight-Space Compensation of KV Cache

Chanryeol Lee, Chanhyuk Lee, Yeonwoo Choi, Donggyun Kim, Seunghoon Hong

原文摘要

Long-context inference with Large Language Models (LLMs) is bottlenecked by the linearly growing memory of the key-value (KV) cache. Existing compression methods reduce the cache through token eviction or approximation, but degrade sharply at aggressive compression budgets. We propose PatchKV, a training-free framework that compensates KV cache compression methods by carrying part of the context in the model's weights. PatchKV pairs an off-the-shelf compressed KV cache with a context-specific weight patch, which is computed once at context-loading time and served for downstream queries for the context. The weight patch is derived in closed form via ridge regression, by aligning the block-wise activations of context-derived reference query tokens under the full cache and the compressed cache. Once merged into the model, the patch leaves the forward graph and per-query inference cost unchanged in the single-context, multi-query setting. Across long-context QA (SCBench with up to 170K tokens, SQuAD, NIAH) and math (GSM8K) benchmarks on three model architectures, PatchKV consistently improves cache compression methods, suggesting an alternative direction to compensate them at aggressive budgets. Code is available at https://github.com/cusasak/PatchKV.

cs.LG

新论文 · v1arXiv:2611.16726

From Data to Wisdom: Rethinking AI Through an Epistemic Lens

Afonso Eduardo

原文摘要

This paper proposes the Data-to-Wisdom Epistemic Architecture for AI (EAAI), a conceptual framework treating Data, Information, Knowledge, Understanding and Wisdom as five distinct but recursively interacting epistemic functions: observation, contextual interpretation, modeling and prediction, critical evaluation and informed judgment. Two observations motivate it: sequential readings of the Knowledge Pyramid obscure how these functions interact, and a targeted review of ten Bayesian reference books finds that prior distributions are defined mathematically while "prior information" and "prior knowledge" are not consistently distinguished as epistemic categories. A Bayesian interpretation relates the likelihood, prior and posterior distributions, posterior prediction, model checking and decision analysis to the five functions, without limiting EAAI to Bayesian systems. Extending "prior" in an epistemic sense, the paper distinguishes five kinds of prior context that can shape the model, the likelihood, the prior, evaluation and action. Three worked applications -- quantitative financial risk, actuarial modeling in health insurance and agentic AI for medical reimbursement claims -- show how EAAI structures the analysis of model assumptions, evidence interpretation, validation, dependencies and decision consequences, separating the prediction of a risk, the appraisal of its reliability and the justification of action. It draws implications for AI and machine learning, including evaluation by function, treating intermediate model outputs as evidence open to challenge and separating generation from appraisal, and compares EAAI with work on DIKW, Bayesian workflow, AI evaluation and risk governance. Its contributions are an epistemic vocabulary with recursive relationships, a more accessible reading of Bayesian methodology, a typology of prior context and clearer methodological terminology.

cs.LG · cs.AI

版本更新 · v2arXiv:2611.13607

One Byte, One Token: Byteifying a 27B Hybrid-Attention LLM with Self-Generated Data

Liwei Zhang, ZhiMing Tong, Mochen Guan, Linzhe Li

原文摘要

Byte-level BPE fragments Chinese text (for Qwen3.8 on Chinese Wikipedia, 3.84 bytes per token on average; 64% of tokens are single characters) and hides spelling from the model. We ask whether an already-trained 27B subword model can be converted directly into a pure byte-level model (one byte, one token), without a patching structure or a local encoder, while keeping most of its ability. Starting from Qwen3.8-27B (48 Gated DeltaNet layers + 16 full-attention layers), we shrink the vocabulary from 248,320 to 384 entries (the 256 byte symbols already in the BPE vocabulary plus special ids), leave every other weight untouched, and continue training in a two-stage curriculum (encoding first, then reading) on 4.77 GB of supervised bytes: all dialogue/behavioral supervision is generated on-policy by the original model itself (character dictionary, word dictionary, general dialogue, coding sessions), and knowledge reading uses encyclopedia text (54% of supervised bytes). Training takes 178.6 h on 8xH200. On the same GPU and serving stack, the byte model's gap to the original falls within the paired bootstrap interval on MATH-500, HumanEval+, MMLU-Pro and non-thinking GSM8K, and it loses 4-6 points on C-Eval, thinking-mode GSM8K and non-thinking IFEval. Without thinking it leads on every character-level task (CUTE 75.2 -> 84.0, EXECUTE-zho 81.6 -> 93.9, our Chinese character suite 54.4 -> 64.0), but lookup knowledge about Chinese characters such as pronunciation and stroke count drops, and thinking does not recover it. The cost is throughput: per-step speed is identical, but each step yields 3-5x less output; Chinese characters per second is about 1/4 of the original, and prefill at 64K bytes is 6.5x slower. We also document four byte-model failure modes (greedy runaway, sampling breaking UTF-8, ineffective penalty terms on a small vocabulary, repetition) and their fixes.

cs.CL · cs.LG

新论文 · v1arXiv:2611.16426

SoScale: Second-Order Scale Search for FP4 Quantization

Pengxiang Zhao, Xing Li, Hui-Ling Zhen, Weizhe Lin, Xianzhi Yu, Zhenhua Dong

原文摘要

Low-bit quantization reduces the memory footprint and inference cost of large language models (LLMs), yet accurate FP4 quantization remains challenging. Practical formats such as MXFP4, NVFP4, and HiF4 pair a 4-bit floating-point payload with block-wise or hierarchical scale codes, making accuracy highly sensitive to how each local distribution is scaled. Default maximum-absolute-value scaling preserves dynamic range, but can spend scarce FP4 levels on rarely occupied outlier regions and leave dense tensor values poorly resolved. Existing scale-search methods minimize local element-wise reconstruction error, optimizing tensor fidelity rather than layer-output fidelity. We propose SoScale, a second-order weight scale-code search method for hierarchical FP4 quantization. SoScale searches hardware-valid scale codes under a Hessian-aware layer reconstruction objective and couples this search with GPTQ-style error recovery, selecting scales by their impact on layer outputs. Across Qwen3 models, SoScale consistently reduces downstream quantization degradation under MXFP4, NVFP4, and HiF4.

cs.LG · cs.AI

版本更新 · v3arXiv:2601.19070

Critical Organization of p-Adic Deep Neural Networks

W. A. Zúñiga-Galindo

原文摘要

We study continuous neural networks whose neurons are the points of the ring of p-adic integers, a fractal tree-like space that encodes hierarchical architectures. The hidden states solve a nonlinear integral equation, which describes the equilibria of recurrent networks and deep equilibrium models; feedforward architectures appear as the special case of kernels without feedback. We prove that the state is unique and depends Lipschitz-continuously on the parameters whenever the product of the Lipschitz constant of the activation function and the operator norm of the connectivity is less than one, and that, when the parameters are locally constant, every state is exactly the state of a finite network with a tree-like topology. We introduce p-adic discrete deep networks, whose layers are the levels of the tree and whose weights are coarse-grainings of a single kernel, and show that, in the stable region, they converge to the continuous network as the depth tends to infinity. Feedforward networks always have a unique state; hence a multiplicity of states, and therefore critical organization in our sense, requires recurrence. For recurrent networks with radial kernels we prove a sharp transition: the zero state is the only state if and only if the total coupling does not exceed one. For hierarchical kernels, above the threshold, the states are indexed by sign patterns on the leaves of the tree, their number is computed exactly, and they form a tree-indexed partially ordered set graded by instability. Numerical experiments illustrate these results and test conjectures on general radial kernels and on random recurrent networks on the tree. Finally, we interpret the critical organization hypothesis for deep neural networks, originally formulated for brain organization, through the lens of multistability, and we discuss the continuous limits of families of finite networks.

cs.LG

新论文 · v1arXiv:2611.15698

Learning Mixtures of Exponential Family

Mansi Sood, Devavrat Shah

原文摘要

Motivated by causal inference and physical computing systems, we consider the problem of learning mixture of two bounded-support exponential-family distributions with a shared interaction parameter matrix and mixture component-specific linear parameter vectors. The unknown sequence of component labels belongs to a known structured class. We propose an estimator based on a convex loss that avoids the log-partition function; for each fixed label assignment, estimation is a convex optimization problem. We show that the conditional population loss uniquely identifies the interaction matrix and the linear parameter attached to each sample. For a global empirical minimizer, we establish finite-sample bounds on the interaction error in Frobenius norm and the average squared linear-parameter error. The bounds depend on the logarithm of the number of admissible label assignments, and, for fixed dimension and model bounds, imply consistency when this complexity grows sublinearly with sample size. Additional occupancy and separation conditions yield component and label recovery.

stat.ML · cs.LG · eess.SP

新论文 · v1arXiv:2611.15671

The Noise Floor of a Quantization Comparison: Engine, restart, batch and kernel drift, measured against 8-bit and 4-bit effects in two decision models

Aleks Blagojevich, Trevor Sunderland

原文摘要

A quantized model is usually judged by comparing it with its BF16 original, but that comparison changes more than the weights. It often changes the serving engine, it always runs on a different server instance, it may change how requests are batched, and on some stacks the library selects a different kernel depending only on which packages import. We measure each of these sources on the same 2,400 public decisions for two of Cloudflare's decision models, Clef-flash (9B) and Clef (27B), on an NVIDIA GB10, and state every quantization effect as the excess over them. Sources that are not quantization move between 0 and 0.92% of decisions. llama.cpp restarts are bit-identical on every record for both models. vLLM restarts flip 0, 5 and 20 decisions at BF16, FP8 and NVFP4 on Clef-flash and 4, 0 and 22 on Clef; a third instance of Clef's quantized builds (exploratory) leaves FP8 at 0 and puts NVFP4 at 4 to 22, depending on which two instances are compared. An engine change at BF16 flips 6 to 16, a batch-size change 8 to 15, and the Gated DeltaNet kernel transformers selects unasked 10 to 14. Quantization inside one engine moves more: for Clef-flash and Clef, 0.88% and 1.04% at Q8_0, 2.12% and 1.62% at FP8, 5.79% and 3.50% at Q4_K_M, and 5.79% and 5.67% at NVFP4. On both models, every 4-bit interval excludes every non-quantization source. Two of our five pre-registered predictions failed, and the same two fail when the tests are rerun on the 27B: FP8 is not inside restart noise, and changing engine moves fewer decisions than 8-bit quantization does. Accuracy hides most of this. Over the full 31,275 public questions, Clef's Q4_K_M build moves accuracy by a net six questions (-0.02 points [-0.20, +0.16]) while 436 of its decisions get worse (exploratory), and NVFP4 costs Clef-flash 0.95 points and Clef 0.65. We release the code, quantization recipes, per-record outputs and the pre-registration.

cs.LG

版本更新 · v3arXiv:2606.11270

Quantifying Subliminal Behavioral Transfer Ratios in Language Model Distillation

Uwe König, Hamza Kazmi, Ruizhe Li, Maheep Chaudhary

原文摘要

Distillation of a language model intended to transfer benign behavior to a student model may also transfer undesirable characteristics, if they are present in the teacher model, a phenomenon known as subliminal learning. While qualitative evidence supports the existence of this effect, its magnitude has not been systematically characterized. This study quantifies subliminal behavioral transfer ratios by steering two teacher models (Llama-2-7B-Chat and Qwen2.5-7B-Instruct) at varying steering strengths and distilling student models using only benign data. Evaluation on 100 JailbreakBench prompts with GPT-4.1, serving as the evaluator, indicates that transfer is robust but exhibits distinct scaling behaviors. Llama-2 demonstrates a sharp threshold ($τ= {0.25,0.32} \ \text{beyond} \ α= -0.15$), whereas Qwen2.5 displays continuous and higher levels of transfer ($τ$ up to $0.61$).

cs.LG · cs.AI · cs.CL

新论文 · v1arXiv:2611.15607

FADO: Learning Causal Explanations with Prior-Fitted Networks

Seth Flaxman, Jess Carr, Annabel Jakob, Marek Masiak, Dino Sejdinovic

原文摘要

In regression and classification tasks with tabular datasets, most of the literature on explanation in ML (e.g. SHAP, permutation importance) has focused on predictive attribution: does feature $X_i$ help predict outcome $Y$? At its heart, predictive attribution captures correlation, not causation. We instead ask a causal question: would manipulating feature $X_i$ change outcome $Y$? We define a natural quantity to answer this question, the feature intervention effect: $Δ_i = E[Y\mid do(X_i=+1)]-E[Y\mid do(X_i=-1)]$. In general, estimating this quantity from observational data alone would entail strong causal assumptions. We propose FADO (Feature-wise Amortized Do-Operators), an amortized inference method using prior-fitted networks trained on synthetic tabular datasets generated by structural causal models. During training, the synthetic data-generating process is known, so we can compute ground-truth intervention effects $Δ_i$, and train a model to predict them at test time from a purely observational tabular dataset. As we demonstrate through extensive experiments, our new causal explainability method is easy to train and provides better explanations under a range of settings for which standard predictive attribution fails, including proxy variables, confounders, colliders, and placebos. We evaluate it on real, synthetic and semi-synthetic datasets, with and without hidden confounders, and compare it to existing predictive attribution methods and state-of-the-art causal inference methods.

cs.LG · stat.ML

新论文 · v1arXiv:2611.15465

On-Edge Acoustic Continual Anomaly Detection: A Cross-Platform Comparative Study

Rami Zewail, Yehia Ehab, Karim Ali, Ahmed Mahrous, Mohamed Awad, Bassem Mokhtar

原文摘要

Acoustic Anomaly Detection (AAD) serves as a non-contact approach for machine condition monitoring, yet much of the existing research is evaluated offline and rarely addresses embedded hardware constraints or microphone variations under operational drift. This paper presents a comparative evaluation of on-edge continual AAD across two embedded platforms the ESP32-S3 (with an external I2S digital microphone) and the Arduino Nicla Vision (with an onboard Pulse Density Modulation (PDM) microphone) using a shared detection model. The model utilizes a lightweight ShallowNet encoder and contrastive prototype learning to map spectrograms into a 32-dimensional embedding space. Both boards run an identical TensorFlow Lite for Microcontrollers (TFLM) encoder, performing on-device prototype learning updated via an exponential moving average. A cloud pipeline orchestrated through Firebase and MQTT enables remote data collection and retraining. Evaluated via a controlled speaker-replay protocol using standardized Detection and Classification of Acoustic Scenes and Events (DCASE) Challenge Task 2 machine sounds, the ESP32-S3 achieved 82.35% to 95.00% accuracy across five machine types, while the Nicla Vision achieved 79.41% to 95.00%. The two platforms demonstrated comparable performance, confirming that the lightweight prototypical architecture transfers reliably across distinct hardware and sensor tiers.

cs.SD · cs.LG

版本更新 · v2arXiv:2611.09203

Copies Are Not Confirmations: How News Duplication Fools LLM Judges and Continual Learners

Abdelkader Metales, Miriam Grabenheinrich, Aicha Metales

原文摘要

A language model that keeps learning from the news sees the same claim many times, but many reports are copies of one source: wire stories, content shared within a media group, syndication and press releases. We ask whether LLM judges and continual learners treat such copies as independent confirmations. In 200,000 English articles from CC-News, a near-duplicate detector identifies copy links to another outlet for 18.9% of the articles, a detector-based lower bound. In a controlled experiment with claims about invented people, three identical reprints instead of one report raised the lies accepted by Llama 3.2 from 0.06 to 0.71 and by phi4 from 0.21 to 0.90; Gemini 2.5 Flash already believed a single report, and Gemini 2.5 Pro followed provenance information most consistently. In a simulated continual-learning stream, showing the judge one report per detected copy chain cut the copied lies accepted by 0.10-0.24 across four judges, at a mean cost of at most 4.2 points of accepted truths. Text similarity links none of the copies reworded by three LLMs. An entity-novelty test, the share of a report's named entities absent from an earlier look-alike, separates these rewrites from 100 author-labelled independent reports of the same event (AUC 0.98-0.99), statistically indistinguishable from TF-IDF and the best tested embedding model, but recognises only 17% of natural same-source derivatives. With both detectors run on 15,084 real articles carrying inserted claims, one report per chain cut the lies accepted by phi4 from 0.57-0.67 to 0.09 (copies), 0.14 (rewrites) and 0.34 (rewrites with invented details), and by Llama 3.2 from 0.56-0.62 to 0.07, 0.09 and 0.25, at a cost of 2.4 and 10 points of accepted truths. Counting reports or chains without a judge is unreliable, and colluding sources that do not copy each other remain an open problem.

cs.LG · cs.CL

新论文 · v1arXiv:2611.15311

Arbor: When Labels Are Not Coarsened at Random

Faisal Mohammad, Mohamed Younis

原文摘要

Merging datasets labelled at different levels of a taxonomy leaves some images with only a coarse label, and some classes with no fine-grained label at all. Standard training scores a coarse label by the total probability of the classes under it. Coarse labels name superclasses, and the standard objective is exact only when every class within a superclass is equally likely to receive a coarse label. We call a class novel if it is never labelled finely, and known otherwise. A novel class violates the assumption, and the classifier assigns its images to a known sibling. We propose Arbor, which models how labels become coarse. In each superclass with a novel class, Arbor adds a virtual novel class, and a logistic gate divides the superclass's probability between it and the known classes. The gate is trained with the likelihood of the observed labels under this model, the coarsening likelihood. We prove that at its population optimum the gate equals the Elkan-Noto correction of positive-unlabelled learning, and show that training the correction into the classifier improves on applying it after training. Across Cars, CUB and Aircraft with 90% coarse labels, Arbor correctly assigns 76-85% of novel-class images to their superclass, more than hierarchical classifiers, hierarchical novelty detection and the Elkan-Noto correction. Its top-1 accuracy over the leaf classes on known-class images is within 1.6 points of the best baseline, although its final predictions return the superclass more often. On the more challenging iNaturalist-2019 benchmark, Arbor improves novel-class accuracy from 46% (uniform target, the strongest baseline) to 68%.

cs.CV · cs.LG

新论文 · v1arXiv:2611.15234

Schedule-Calibrated Commitment Order in Blockwise Masked Diffusion Language Models

Gaurav Sarkar, Umang Pandey, Pratik Prabhanjan Brahma, Zicheng Liu, Emad Barsoum

原文摘要

Sequential block schedules can make token commitment in masked diffusion language models appear strongly left-to-right even when within-block order is random or reversed. For equal contiguous blocks processed without interleaving in positional order, with untied irreversible commitments, we use a Spearman-rank identity to express global position-commitment correlation as a deterministic schedule reference plus a weighted mean within-block correlation. In a reported paired study of LLaDA-8B-Instruct on 500 math and code prompts with 256 generated positions and updates, increasing block size from 64 to 128 lowers mean global correlation from 0.9845 to 0.9149. Uniform-random within-block references are already 0.9375 and 0.7500. After calibration, mean within-block correlations are 0.751 and 0.660, so the larger raw residual at block size 128 does not indicate stronger local positional ordering. We give sharp bounds and derive calibrated intervals from reported summaries, while documenting the limitations of the aggregate-only empirical release. The analysis supports schedule-aware interpretation of commitment order; its empirical scope is one checkpoint and two schedules, without evaluation of output quality or inference speed.

cs.LG · cs.CL

版本更新 · v2arXiv:2605.05659

Structural Correspondence and Universal Approximation in Diagonal plus Low-Rank Neural Networks

Ying Chen, Aoxi Li, Jihun Kim, Javad Lavaei

原文摘要

Structured low-rank and sparse-plus-low-rank parameterizations, motivated by methods such as LoRA and low-rank factorization, impose strong structural constraints on the linear transformations within neural networks. We study the expressive consequences of these constrained structures from the perspective of universal approximation. We first characterize that purely low-rank neural networks, despite being able to exactly interpolate arbitrary finite scalar datasets, can fail to approximate continuous functions that vary along directions ignored by the low-rank hidden mapping. We show that augmenting the low-rank structure with a one-parameter diagonal component suffices to overcome this geometric bottleneck, yielding a Diagonal-plus-Low-Rank (DLoR) structure. We show that a full-rank linear transformation can be reconstructed using DLoR components either by expanding network width through additive decomposition or by extending network depth through multiplicative decomposition. We term this correspondence between additive width expansion and multiplicative depth extension the Structural Correspondence framework. By combining these constructions with local linearization of the activation function, we prove that DLoR neural networks restore universal approximation for general activation functions. These results characterize how architectural width and depth can compensate for severe rank constraints without requiring unrestricted dense hidden transformations. Diagonal terms are required for multiplicative construction, while additive expansion works with purely low-rank terms. We further determine its exact cost: both the minimal number of rank-$r$ DLoR factors representing $W$ and the minimal network depth for that layer equal $\lceilκ(W)/r\rceil$, where $κ(W)=\min_{γ\ne0}rank(W-γI)$, achieving a lower cost than its dense representation when $W$ is identity-centered.

cs.LG

新论文 · v1arXiv:2611.15073

When Does Causality Help in Multivariate Time-Series Anomaly Detection?

Long Zhang, Wei Huang, Aishui Rao, Jian Hu, Binxiang Wu, Jie Li, Chen Gong, Linxi Feng

原文摘要

Causality-aware scoring is common in multivariate time-series anomaly detection, yet the literature rarely states when it helps. Holding one causality-guided autoencoder (DeCauAE) fixed, we vary graph quality, anomaly type and protocol against ground truth. Graph quality fixes the payoff: on a twenty-variable DAG the gain over a graph-free autoencoder decays as the graph is corrupted. Anomaly type fixes the target: the structural factor lifts window F1 from 0.800 to 0.976 where anomalies break conditional structure, and its advantage vanishes below a break severity s 0.3. Protocol fixes the ranking: point-adjusted F1 lets a random score match the full model, while range-aware VUS separates the autoencoder family from the classical baselines. On real benchmarks the causal module pays only where density makes a system eligible and cross-subsample reproducibility selects entities inside it; enabling it everywhere costs 0.012 VUS-ROC. The structural factor also transfers to classical detectors .

cs.LG