OpenReview · 实时热榜

期刊：Transactions on Machine Learning Research · 摘要：Reinforcement Learning (RL) training for Large Language Models (LLMs) often suffers from instability due to the discrepancy between training and inference. This training-inference discrepancy stems from two primary factors: an architectural separation between training and inference engines, and the use of low-precision quantization in inference versus higher-precision computation in training. To address training instability issues caused by high training-inference discrepancy, we present the principles and methods for its adaptive control. We propose Adaptive Control Reinforcement Learning (A… · 篇幅：Long submission (more than 12 pages of main content) · OpenReview ID：8sBZSgAkAH

期刊：Transactions on Machine Learning Research · 摘要：For robots to operate reliably in real-world environments, they need to perceive their surroundings, act, and reason about the consequences of those actions. Rapid progress in the domains of representation learning, vision-language-action (VLA) models, and world models has significantly enhanced the capabilities of robot learning systems, enabling robots to work in increasingly complex environments. However, these paradigms are typically developed in isolation, resulting in fragmented systems that struggle with generalization, long-horizon temporal reasoning and planning, and deployment in un… · 篇幅：Long submission (more than 12 pages of main content) · OpenReview ID：AXv7OjfuiT

期刊：Transactions on Machine Learning Research · 摘要：Robot foundation models (RFMs), including vision-language-action (VLA) policies, are often discussed through a scaling view: more data, larger models, and broader benchmarks should improve generalization. In robotics, however, a model can generalize while work still remains before it can run on a robot with a particular body. The work required differs across methods and target robots, and those differences affect practical deployment. We call the gap between reusable models, representations, or data and their use in execution on the target robot the embodiment gap. This survey examines what can… · 篇幅：Long submission (more than 12 pages of main content) · OpenReview ID：D0XcH9Cso4

期刊：Transactions on Machine Learning Research · 摘要：Recently reconstruction-based deep models have been widely used for time series anomaly detection, but as their capacity and generalization capability increase, these models tend to over-generalize, often reconstructing unseen anomalies accurately. Prior works have attempted to mitigate this by incorporating a memory architecture that stores prototypes of normal patterns. Nevertheless, these approaches suffer from high training costs and have yet to be effectively integrated with time series foundation models (TSFMs). To address these challenges, we propose MOMEMTO, an improved TSFM variant f… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：Fm2ddpR0aw

期刊：Transactions on Machine Learning Research · 摘要：Path-based link prediction methods reconstruct missing links between two vertices of a knowledge graph. They reconstruct a missing link by finding a path through the knowledge graph connecting both vertices. The path is the reasoning of the link prediction method. However, path-based link prediction methods are vulnerable to Clever Hans biases. They learn invalid reasoning patterns if these patterns are dominant and generalize well to the training and validation set. As a result, performance drops when evaluated on the real-world distribution. Whether reasoning is valid, that is, semantically… · 篇幅：Long submission (more than 12 pages of main content) · OpenReview ID：FrMBxE7J0v

期刊：Transactions on Machine Learning Research · 摘要：CLIP-style image encoders are trained to be discriminative for every category set a user might supply, since the category set is unknown at training time. This makes the encoder's job harder than the job any single deployment actually requires, and is part of why small image encoders underperform large ones on zero-shot classification. In CLIP, the class prompts available at inference are used only to define the classifier head; we argue they carry more task structure than this role exposes, enough to also modulate the image encoder's feature extraction through a small channel (BatchNorm scal… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/locuslab/HyperCLIP · OpenReview ID：kX7iwF5s3v

期刊：Transactions on Machine Learning Research · 摘要：We formalize online learning-unlearning (OLU) in the Online Convex Optimization (OCO) setting, where a learner updates a model sequentially on a stream of convex losses while accommodating occasional unlearning requests between updates. We require that after a deletion, the distribution of all future outputs is statistically indistinguishable from that of a learner trained on the same stream without the deleted data. We propose two OLU algorithms based on Online Gradient Descent (OGD). Passive OLU leverages the contractive dynamics of OGD and injects calibrated noise, incurring no additional… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：lyvCVfTdiY

期刊：Transactions on Machine Learning Research · 摘要：The dominant paradigm in machine intelligence typically defines learning as the minimization of an empirical loss function. While highly successful, this approach can result in systems prone to catastrophic forgetting and reward hacking, contrasting with the homeostatic stability observed in biological organisms. We propose Constraint-First Machine Learning (CFML), a framework that reformulates learning as the maintenance of feasibility under an expanding set of structural constraints rather than the pursuit of a global optimum. Utilizing Viability Projection Updates (VPU) based on reflected… · 篇幅：Long submission (more than 12 pages of main content) · OpenReview ID：sgvPGWZ1uv

期刊：Transactions on Machine Learning Research · 摘要：Large language models (LLMs) excel at program synthesis, yet their ability to produce symbolic graphics programs (SGPs) that render into precise visual content remains underexplored. We study symbolic graphics programming, where the goal is to generate an SGP from a natural-language description. This task also serves as a lens into how LLMs understand the visual world by prompting them to generate images rendered from SGPs. Among various SGPs, we focus on scalable vector graphics (SVGs), as they are widely used and can be easily rendered into images. We begin by examining the extent to which… · 篇幅：Long submission (more than 12 pages of main content) · 代码：https://github.com/Sphere-AI-Lab/SGP-RL · OpenReview ID：JfK8EHJAQn

期刊：Transactions on Machine Learning Research · 摘要：We introduce a failure-aware adversarial retrieval-augmented framework for improving robustness in natural language understanding. Rather than selecting synthetic examples with a fixed reward threshold, our method formulates adversarial data curation as a failure-mode contextual bandit problem. Candidate examples are generated with retrieval-augmented prompting, filtered by the current target model, automatically validated by an LLM judge ensemble, and clustered into recurring failure modes. A stochastic policy then selects which failure modes to sample for retraining, and is updated using va… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：DSpKdN6whZ

期刊：Transactions on Machine Learning Research · 摘要：Text-to-image models have transformed visual content creation, but their reliance on large uncurated web data can encode and amplify societal biases. We present \emph{FairT2I}, a training-free, inference-time framework that leverages large language models to detect implicit bias dimensions in prompts and mitigate them during generation. FairT2I has three components. First, LLM-based bias detection identifies bias-relevant attributes implied by the prompt and makes them explicit for control. Second, attribute resampling generates bias-aware prompts by sampling these attributes from a user-spec… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/jinya1013/fairt2i · OpenReview ID：WHYE1WjMYf

期刊：Transactions on Machine Learning Research · 摘要：Neural visual decoding is a central problem in brain–computer interface research, aiming to reconstruct human visual perception and to elucidate the structure of neural representations. Recent contrastive neural visual decoding methods commonly align neural signals with the final embeddings of pretrained vision encoders. However, such representations are optimized for high-level semantic invariance, whereas EEG/MEG signals contain information spanning multiple levels of visual abstraction, potentially creating a representational granularity mismatch. Motivated by prior evidence that brain rep… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：SzDHkICZpm

期刊：Transactions on Machine Learning Research · 摘要：Model-based reinforcement learning (MBRL) learns world models—internal simulators of environment dynamics—to plan by imagining future trajectories. However, when these models incorrectly predict state transitions, they generate unrealistic states that mislead agents into learning delusional policies. Inspired by human vision, we propose anomaly detection in world model with \textbf{L}inear \textbf{P}rior (LP), a three‐stage approach that 1) enforces a lightweight linear prior on successive latent states, 2) flags generated states that deviate from this prior, and 3) removes their contribution… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/zarifikram/lp-dreamer · OpenReview ID：VLIzLK3CfR

期刊：Transactions on Machine Learning Research · 摘要：We present the inversion variational autoencoder ($\mathcal{I}$-VAE), a conditional generative model for efficient belief-state planning in partially observable sequential decision-making problems. The $\mathcal{I}$-VAE maps partial observations to stochastic posterior state samples by learning an observation-conditioned latent prior, enabling consistent belief updates without an explicit likelihood model. We further fine-tune the belief model with a trajectory-based mutual information objective to improve latent space consistency across observation sequences. To support scalable planning wit… · 篇幅：Long submission (more than 12 pages of main content) · OpenReview ID：K7MPwAV1AT

期刊：Transactions on Machine Learning Research · 摘要：Rule-based models, such as decision trees, appeal to practitioners due to their interpretable nature. However, the learning algorithms that produce such models are often vulnerable to spurious associations, and thus, they are not guaranteed to extract causally relevant insights. This limitation reduces their utility in gaining mechanistic insights into a phenomenon of interest. In this work, we build on ideas from the invariant causal prediction literature to propose Invariant Causal Set Covering Machines, an extension of the classical Set Covering Machine (SCM) algorithm for conjunctions/dis… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/thibgo/icscm-expe-real-data · OpenReview ID：slquR2A8rA

期刊：Transactions on Machine Learning Research · 摘要：In a celebrated paper, Noé, Olsson, K\"ohler and Wu (\cite{noe2019boltzmann}) introduced an efficient method for sampling high-dimensional Boltzmann distributions arising in molecular dynamics via normalizing flow approximation of transport maps. Here, we place this approach on a firm mathematical foundation. We prove the existence of a normalizing flow between the reference measure and the true Boltzmann distribution up to an arbitrarily small error in the Wasserstein distance. This result covers general Boltzmann distributions from molecular dynamics, which have low regularity due to the pr… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/zia7735/boltzmann-sampling-tmlr2026.git · OpenReview ID：4HLZD6LMuJ

期刊：Transactions on Machine Learning Research · 摘要：Efficient deployment of large language models (LLMs) requires balancing inference speed with output quality. Speculative decoding accelerates inference by using a smaller draft model to propose future tokens, whereas reasoning-heavy approaches—such as chain-of-thought prompting, ensembles, and dynamic routing—improve output quality through deep search and verification. Although historically treated as isolated research trajectories, the demands of complex, high-difficulty tasks have forced these domains to converge. This paper presents a structured taxonomy and analysis focused specifically o… · 篇幅：Long submission (more than 12 pages of main content) · OpenReview ID：OIrJI53MvN

期刊：Transactions on Machine Learning Research · 摘要：Large language models generate outputs stochastically and may produce fluent but invalid responses, including factual hallucinations. Existing mitigation strategies reduce average error rates but do not provide explicit control over the \emph{frequency} of such failures under repeated use. We formulate inference as a deployment-time risk control problem and introduce \emph{chance-constrained inference} (CCI), which directly bounds the probability of hallucinations among accepted generations. Hallucinations are modeled as stochastic constraint violations, and we show that confidence-based sele… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：cJDhDC69m9

期刊：Transactions on Machine Learning Research · 摘要：Integrating combinatorial optimization layers into neural networks has recently attracted significant research interest. However, many existing approaches lack theoretical guarantees or fail to perform adequately when relying on inexact solvers. This is a critical limitation, as many operations research problems are NP-hard, often necessitating the use of neighborhood-based local search heuristics. In this paper, we introduce a principled approach for learning with such inexact solvers. Inspired by the connection between simulated annealing and Metropolis-Hastings, we transform the problem sp… · 篇幅：Long submission (more than 12 pages of main content) · OpenReview ID：M2a0mbACNf

期刊：Transactions on Machine Learning Research · 摘要：How much information about an agent's underlying values can be recovered from its observable behavior? This question matters for any approach that infers agent properties from action sequences, yet remains empirically open for LLM-based agents at scale. We address it through controlled experiments: LLM-based agents (Llama~3.1-8B) assigned one of 36 behavioral profiles (9 belief systems x 4 motivations) generate over 1.5 million behavioral sequences across 36 behavioral profiles in grid-world environments, providing ground truth unavailable in human behavioral studies. After filtering, both cl… · 篇幅：Long submission (more than 12 pages of main content) · OpenReview ID：aDMDqtw63H

期刊：Transactions on Machine Learning Research · 摘要：We define an information-complexity property for aggregation functions, capturing a vast range of practical aggregations, and prove that any Message-Passing Graph Neural Network (MP-GNN) model with such aggregations induces only a polynomial number of equivalence classes on all graphs - while the number of non-isomorphic graphs is super-exponential (in number of vertices). Adding a familiar perspective, we observe that merely 2 iterations of Color Refinement (CR) induce at least an exponential number of equivalence classes, making the aforementioned MP-GNNs relatively infinitely weaker. Previ… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：F4OlpAJ0o2

期刊：Transactions on Machine Learning Research · 摘要：We introduce PLACE (Persistence-Landmark Analytic Classification Engine), a closed-form pipeline for classifying point clouds and graphs through their persistent-homology signatures. Three quantitative guarantees—a margin-based excess-risk rate, a closed-form descriptor-selection rule, and a per-prediction certificate—are derived from training labels alone, with no learned weights or held-out calibration. The embedding sums Mitra–Virk single-point coordinate functions over a sparse landmark grid; the closed-form weight rule $w_k^2 \propto (d_{k+1}^2 - d_k^2)/R_k^2$ maximizes the distortion sl… · 篇幅：Long submission (more than 12 pages of main content) · 代码：https://github.com/akritihq/place-palace · OpenReview ID：4kZxNlE5Ve

期刊：Transactions on Machine Learning Research · 摘要：Urban mobility digital twins are revolutionizing how cities manage increasingly complex transportation systems, enabling real-time optimization across multiple stakeholders, services, and dynamic operations. Central to these digital twins is the origin-destination (OD) calibration problem—estimating travel demand patterns that produce realistic traffic simulations matching observed conditions. However, existing calibration methods face critical limitations: they require a prohibitively large number of expensive simulation runs and struggle with high-dimensional city-scale networks. To mitigat… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：r5oS1XXbT3

期刊：Transactions on Machine Learning Research · 摘要：We develop a learning-theoretic framework for analyzing self-improving agents by decomposing self-modification into five axes. Within this framework, we prove a sharp boundary: under standard i.i.d. assumptions, distribution-free PAC learnability is preserved if and only if the policy-reachable family remains uniformly capacity-bounded. If reachable capacity can grow without bound, utility-rational self-changes can make learnable tasks unlearnable. We further introduce a simple Two-Gate guardrail—a validation-improvement requirement plus a capacity cap—that preserves this boundary and yields… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：q4vuDMtYgF

期刊：Transactions on Machine Learning Research · 摘要：The widespread adoption of deep learning models in computer vision has intensified concerns about interpretability. Despite strong performance, these models are often treated as black boxes, with limited systematic investigation of their decision-making processes. While many interpretability methods exist, objective evaluation of learned representations remains limited, particularly for approaches that rely on sparsity to ``induce'' interpretability. In this work, we investigate how modeling choices in Concept Bottleneck Models (CBMs) affect the semantic alignment of concept representations.… · 篇幅：Long submission (more than 12 pages of main content) · 代码：https://github.com/konpanousis/cbm-clarity · OpenReview ID：IyQEQBRR4M

期刊：Transactions on Machine Learning Research · 摘要：Fine-Tuning is the dominant paradigm for adapting pretrained large language models (LLMs) to downstream NLP tasks. In practice, fine-tuning datasets may contain various forms of noise that arise from annotation errors or automated data collection. Although prior work has concentrated on designing robust learning algorithms to mitigate performance degradation under noisy conditions, comparatively little is known about how different types of noise affect the internal learning dynamics of LLMs during fine-tuning. In this work, we systematically study the impact of noise on model behaviour across… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/LingfangLi/analyzing-noise-llm-finetuning · OpenReview ID：NlSBeHZEz5

期刊：Transactions on Machine Learning Research · 摘要：Mixture-of-Experts (MoE) models achieve scalability through sparse expert routing, but experts process tokens independently. A natural hypothesis is that enabling expert communication—through learned topologies, message passing, or sequential chains—should improve performance. We test this hypothesis on WMT14 En-De translation with a small decoder-only Transformer, evaluating ten communication approaches across seven exper-imental axes. We find no clear evidence that any variant improves over standard MoE, though modest sample sizes limit power for detecting small effects; several variants de… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：hPD4MjMfoN

期刊：Transactions on Machine Learning Research · 摘要：Accurate and rapid estimation of hemodynamic metrics, such as pressure and wall shear stress (WSS), is important for assessing the severity of Coronary Artery Disease (CAD). Existing approaches, including invasive Fractional Flow Reserve (FFR) measurements and computationally expensive Computational Fluid Dynamics (CFD) simulations, face challenges in invasiveness, cost, and speed. We present a learned surrogate for fast prediction of CFD-simulated coronary hemodynamics from vessel centerline geometry. The model encodes 1D vessel centerlines together with inlet flow rate using a transformer-b… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：AoJUrVjufP

期刊：Transactions on Machine Learning Research · 摘要：Forecasting how a patient’s condition is likely to evolve, including possible deterioration, recovery, treatment needs, and care transitions, could support more proactive and personalized care, but requires modeling heterogeneous and longitudinal electronic health record (EHR) data. Yet, existing approaches typically focus on isolated prediction tasks, narrow feature spaces, or short context windows, limiting their ability to model full patient pathways. To address this gap, we introduce EHR2Path, a multimodal framework for forecasting and simulating full in-hospital patient pathways from rou… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/ChantalMP/EHR2Path · OpenReview ID：ywa71iOykg

期刊：Transactions on Machine Learning Research · 摘要：We aim at designing language agents with greater autonomy for crystal materials discovery. While most of existing studies restrict the agents to perform specific tasks within predefined workflows, we aim to automate workflow planning given high-level goals and scientist intuition. To this end, we propose Materials Agent unifying Planning, Physics, and Scientists, known as MAPPS. MAPPS consists of a Workflow Planner, a Tool Code Generator, and a Scientific Mediator. The Workflow Planner uses large language models (LLMs) to generate structured and multi-step workflows. The Tool Code Generator s… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：Cwq1U8tbWW

期刊：Transactions on Machine Learning Research · 摘要：We ask whether the geometry of a reasoning trajectory, that is, how a chain-of-thought (CoT) trace moves through semantic space beyond its raw length, predicts whether the final answer is correct, and whether that prediction is useful in practice. Across 2,800 CoT traces spanning three reasoning benchmarks (FOLIO, GSM8K, and PrOntoQA) and five language models, we extract interpretable trajectory-level features (adjacent-step transition energy, path entropy, semantic drift, loopiness, discourse-graph spectra, and direction-sensitive drift) and predict per-trace correctness. Under problem-group… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：H9cBkEqVeY

期刊：Transactions on Machine Learning Research · 摘要：Coherent imaging modalities such as ultrasound and synthetic aperture radar (SAR) images are degraded by signal-dependent multiplicative noise, where the noise distributions vary widely across acquisition scenarios. Existing self-supervised image denoising methods either assume zero-mean additive noise, independence across pixels or require the noise distribution to be known, which often limit their applicability in real-world image denoising systems. We propose a Generalized Multiplicative Unbiased Risk Estimate (Gen-MURE), a model-agnostic self-supervised image denoising framework for enhan… · 篇幅：Long submission (more than 12 pages of main content) · OpenReview ID：Hie13qRm1x

期刊：Transactions on Machine Learning Research · 摘要：Large language models increasingly mediate decisions in healthcare, legal advisory, and financial analysis, settings in which a model’s willingness to answer an inadequate prompt can matter as much as the accuracy of its answer. Yet systematic cross-model evidence on this behavior remains scarce. The present study examined over-compliance, understood as the generation of substantive content when the input warrants clarification, refusal, or deferral. Four frontier models from Ope- nAI, Google, Meta, and Anthropic were evaluated on a benchmark of 400 prompts spanning under- specification, ambi… · 篇幅：Long submission (more than 12 pages of main content) · 代码：https://github.com/hgus107/LLM-Over-Complaince · OpenReview ID：LnUP74YNze

期刊：Transactions on Machine Learning Research · 摘要：Large Language Models (LLMs) have recently achieved impressive performance in math and reasoning benchmarks. However, they often struggle with logic problems and puzzles that are relatively easy for humans. To further investigate this, we introduce a new benchmark, SearchBench, which contains 11 unique search problems inspired by intuitive puzzles. Each SearchBench problem type is equipped with automated pipelines to generate an arbitrary number of instances and analyze the feasibility, correctness, and optimality of LLM-generated solutions. We show that using step-by-step, language-only reas… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：oub2I1ioL5

期刊：Transactions on Machine Learning Research · 摘要：Hypergraph is a data structure that enables us to model higher-order associations among data entities. Conventional graph-structured data can represent pairwise relationships only, whereas hypergraph enables us to associate any number of entities, which is essential in many real-life applications. Hypergraph learning algorithms have been well-studied for numerous problem settings, such as node classification, link prediction, etc. However, much less research has been conducted on anomaly detection from hypergraphs. Anomaly detection identifies events that deviate from the usual pattern and ca… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：etPYIk1BqO

期刊：Transactions on Machine Learning Research · 摘要：Current methods for stochastic hyperparameter learning in Gaussian Processes (GPs) rely onapproximations, suchascomputingbiasedstochasticgradientsorusinginducingpointsin stochastic variational inference. However, when using such methods, we are not guaranteed to converge to a stationary point of the true marginal likelihood. In this work, we propose algorithms for exact stochastic inference of GPs with kernels that induce a Reproducing Kernel Hilbert Space (RKHS) of moderate finite dimension. Our approach can also be extendedtoinfinitedimensionalRKHSsatthecostofforgoingexactness. Bothforfinit… · 篇幅：Long submission (more than 12 pages of main content) · OpenReview ID：wDCulUZla4

期刊：Transactions on Machine Learning Research · 摘要：We present a reproducibility study of XFeat, a lightweight local feature extractor and matcher designed to identify corresponding points across images efficiently on resource-constrained hardware. We re-implement the architecture based on the paper and supplementary material, re-evaluate the authors' released checkpoint alongside our re-implementation, and conduct additional architectural ablations to examine design choices that were not fully justified in the original work. This distinction between re-evaluation and reproduction is important, as the paper, supplement, and public code differ… · 篇幅：Long submission (more than 12 pages of main content) · 代码：https://github.com/GalaxyGHz/xfeat-revisited · OpenReview ID：2WI889Ulin

期刊：Transactions on Machine Learning Research · 摘要：Quantization of Deep Neural Network (DNN) activations is a commonly used technique to reduce compute and memory demands during DNN inference, which can be particularly beneficial on resource-constrained edge devices. To achieve high accuracy, existing methods for quantizing activations rely on complex mathematical computations or perform extensive online searches for the best hyperparameters. However, these expensive operations are impractical on edge devices with limited computational capabilities, memory capacities, and energy budgets. Furthermore, many existing methods either do not focus… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/gicLAB/eDQA · OpenReview ID：SEIBCdgE5W

期刊：Transactions on Machine Learning Research · 摘要：Emergent Language (EL) focuses on the emergence of communication among artificial agents. Although symbolic communication channels more closely mirror the discrete nature of human language, learning such protocols remains fundamentally difficult due to the non-differentiability of symbol sampling. Existing approaches typically rely on high-variance gradient estimators such as REINFORCE or on continuous relaxations such as Gumbel–Softmax, both of which suffer from limitations in training stability and scalability when learning a language from scratch. Motivated by cognitive theories that empha… · 篇幅：Long submission (more than 12 pages of main content) · OpenReview ID：5nqQlGWlsW

期刊：Transactions on Machine Learning Research · 摘要：Quantum mixture of experts (QMoE) extends conditional computation to the NISQ setting by distributing learning across parameterized quantum circuit (PQC) experts selected via a routing mechanism. Existing approaches are limited by single-block experts, lack of load balancing, and aggregation schemes that ignore routing amplitudes. We propose QMoE+, which uses two-block data re-uploading experts with learnable offsets, a coherent aggregation circuit over the joint routing-data Hilbert space, and a Switch-style load-balancing loss. Under top-k=1 sparse routing, QMoE+ activates only ∼28% of its… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/HbjNiser/qmoe-plus · OpenReview ID：l1JaPqZ6K5

期刊：Transactions on Machine Learning Research · 摘要：Safe reinforcement learning (Safe RL) seeks to maximize rewards while satisfying safety constraints, typically addressed through Lagrangian-based methods. However, existing approaches, including PID and classical Lagrangian methods, suffer from oscillations and frequent safety violations due to parameter sensitivity and inherent phase lag. To address these limitations, we propose ADRC-Lagrangian methods that leverage Active Disturbance Rejection Control (ADRC) for enhanced robustness and reduced oscillations. Our unified framework subsumes a broad class of PID Lagrangian updates as frozen-par… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：3IbuT8uzYS

期刊：Transactions on Machine Learning Research · 摘要：While large language models (LLMs) have shown promise in automating data science, existing agents often struggle with the complexity of real-world workflows that require exploring multiple sources and synthesizing open-ended insights. In this paper, we introduce DS-STAR, a specialized agent to bridge this gap. Unlike prior approaches, DS-STAR is designed to (1) seamlessly process and integrate data across diverse, heterogeneous formats, and (2) move beyond simple QA to generate comprehensive research reports for open-ended queries. Extensive evaluation shows that DS-STAR achieves state-of-the… · 篇幅：Long submission (more than 12 pages of main content) · 代码：https://github.com/google-research/ds-star · OpenReview ID：Yz3ZPLzYaU

期刊：Transactions on Machine Learning Research · 摘要：Foundation model developers are among the world’s most important companies. As these companies become increasingly consequential, how do their transparency practices evolve? The 2025 Foundation Model Transparency Index is the third edition of an annual effort to characterize and quantify the transparency of foundation model developers. The 2025 FMTI introduces new indicators related to data acquisition, usage data, and monitoring and evaluates companies like Alibaba, DeepSeek, and xAI for the first time. The 2024 FMTI reported that transparency was improving, but the 2025 FMTI finds this prog… · 篇幅：Long submission (more than 12 pages of main content) · OpenReview ID：1jT253Xtyf

期刊：Transactions on Machine Learning Research · 摘要：Recently there has been a surge of interest in optimal decision tree (ODT) methods that globally optimize accuracy directly, in contrast to traditional approaches that locally optimize an impurity or information metric. However, the literature shows conflicting evidence on the value of ODTs, with some demonstrating superior out-of-sample performance of ODTs over greedy approaches, while others show the opposite. The value and performance of ODTs therefore remains one of several open question regarding ODTs, most of which could not be answered before due to lack of scalability. With our experi… · 篇幅：Long submission (more than 12 pages of main content) · 代码：https://github.com/ConSol-Lab/opt-vs-greedy-dts · OpenReview ID：DvDOAtskXl

期刊：Transactions on Machine Learning Research · 摘要：Large Language Models (LLMs) are increasingly applied to tasks involving structured inputs such as semantic graphs, yet adapting them to such inputs remains non-trivial. Common approaches either linearize graphs, discarding structural information, or rely on specialized architectures that are not directly compatible with standard pretrained LLMs. We present SAFT, a structure-aware fine-tuning method that augments LLMs with graph positional encodings derived from the magnetic Laplacian of the input graph. These encodings are projected into the LLM embedding space, introducing relational induct… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/guerrantif/saft · OpenReview ID：QZoUMyzYDB

期刊：Transactions on Machine Learning Research · 摘要：We study classification problems using binary estimators where the decision boundary is described by horizon functions and where the data distribution satisfies a geometric margin condition. A key novelty of our work is the derivation of lower bounds for the worst-case learning rates over broad classes of functions, under a geometric margin condition---a setting that remains theoretically challenging. Moreover, we work in the noiseless setting, where lower bounds are particularly hard to establish. Our general results cover, in particular, classification problems with decision boundaries belo… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：ZIshsqojB6

期刊：Transactions on Machine Learning Research · 摘要：Designing a reinforcement learning from human feedback (RLHF) algorithm to approximate a human's unobservable reward function requires assuming, implicitly or explicitly, a model of human preferences. In sequential decision making tasks, a preference model that poorly describes how humans generate preferences risks learning a poor approximation of the human’s reward function. In this paper, we conduct human studies to assess whether one can influence the expression of real human preferences to more closely conform to a desired preference model. Importantly, our approach does not seek to alter… · 篇幅：Long submission (more than 12 pages of main content) · OpenReview ID：7YPlw1nUmW

期刊：Transactions on Machine Learning Research · 摘要：We study the maximization of nonnegative, non-monotone $\gamma$-weakly diminishing-returns (DR) submodular functions over down-closed convex bodies. The weakly DR model relaxes classical diminishing returns by allowing marginal gains to decay up to a multiplicative factor $\gamma \in (0,1]$, capturing a broad class of objectives that interpolate between monotone and fully non-monotone DR submodularity. Existing methods in this regime achieve guarantees that deteriorate rapidly as $\gamma$ decreases and fail to recover the best known bounds in the fully DR case. We develop a $\gamma$-aware alg… · 篇幅：Long submission (more than 12 pages of main content) · OpenReview ID：yS78Cb1CnX

期刊：Transactions on Machine Learning Research · 摘要：Natural gradient methods are appealing in policy optimization due to their invariance to smooth reparameterization and their ability to account for the local geometry of the policy manifold. These properties often lead to improved conditioning of the optimization problem compared to Euclidean policy gradients. However, their reliance on Monte Carlo estimation introduces high variance and sensitivity to hyperparameters. In this paper, we address these limitations by integrating Randomized Quasi-Monte Carlo (RQMC) sampling into the natural actor-critic (NAC) framework. We revisit the NAC linear… · 篇幅：Long submission (more than 12 pages of main content) · OpenReview ID：kOSx9v6dfb

期刊：Transactions on Machine Learning Research · 摘要：Recent progress in large language models demonstrates that hybrid architectures--combining self-attention mechanisms with state-space layers--can achieve a compelling balance between modeling quality and computational efficiency, particularly for long-context tasks. While these Transformer–Mamba-2 hybrid models show promising performance, systematic comparisons of hybridization strategies and analyses on the key factors behind their effectiveness have not been clearly shared with the community. In this work, we present a holistic evaluation of hybrid architectures based on inter-layer (sequen… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：x7qyXl8ecT