OpenReview · 实时热榜

期刊：Transactions on Machine Learning Research · 摘要：Anomaly detection seeks to identify observations that deviate significantly from an underlying data distribution. While deep learning and ensemble-based approaches have achieved strong empirical performance, their computational and memory requirements limit their applicability in resource-constrained edge environments. Furthermore, many approaches to improving efficiency rely on supervised models, which require labeled anomalies that are often scarce in practice. We propose Yose-Ue, a resource-efficient, fully unsupervised anomaly detection framework based on treap-structured ensemble learnin… · 篇幅：Long submission (more than 12 pages of main content) · 代码：https://github.com/eduardoortegathethird/Yose-Ue · OpenReview ID：17y2ooyemG

期刊：Transactions on Machine Learning Research · 摘要：Symbolic regression (SR) models complex systems by discovering mathematical expressions that capture underlying relationships in observed data. However, most SR methods prioritize minimizing prediction error over identifying the governing equations, often producing overly complex or inaccurate expressions. To address this, we present a decomposable SR method that generates interpretable multivariate expressions leveraging transformer models, genetic algorithms (GAs), and genetic programming (GP). In particular, our explainable SR method distills a trained "opaque'' regression model into mathe… · 篇幅：Long submission (more than 12 pages of main content) · 代码：https://github.com/NISL-MSU/MultiSetSR · OpenReview ID：54EL928uCf

期刊：Transactions on Machine Learning Research · 摘要：Composed Image Retrieval (CIR) is an emerging paradigm in content-based image retrieval that enables users to formulate compositional queries by combining a reference image with an auxiliary modality, usually text-based. This approach supports fine-grained search where the target image shares structural elements with the user-provided image while incorporating the modifications specified by the auxiliary text. Conventional CIR methods rely on multimodal fusion to combine visual and textual features into a joint query embedding, which requires training modules that align composed queries with… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/StevenXuf/PeFuse4CIR · OpenReview ID：6W3pFEQXZc

期刊：Transactions on Machine Learning Research · 摘要：Earth Observation (EO) uses satellite and aerial imagery to monitor the Earth’s surface, supporting critical applications in infrastructure, agriculture, and climate change. As governments and industry scale EO pipelines, reliable automation has become essential. Yet, current Vision-Language Models are limited to coarse-grained perception, struggling to execute the precise, multi-step reasoning required for operational decision-making. Recent evaluations on benchmarks like GeoBench-VLM highlight this shortcoming: even state-of-the-art models show low accuracy and frequently struggle with task… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：7yUrnyFgEq

期刊：Transactions on Machine Learning Research · 摘要：We study in-context learning (ICL) with Transformers for categorical outputs $y_i$, a setting largely unexplored compared to research on real-valued $y_i$. While attention-only Transformers can, in principle, perform functional gradient descent (GD) inference for real-valued outputs, we show that categorical $y_i$ introduce a nonlinear interlayer computation. The MLP layers interleaved with attention in the standard Transformer are a natural architectural component to approximate this computation, providing a concrete role for MLPs that is absent in the real-valued setting. We characterize co… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：8P4V7V1cs4

期刊：Transactions on Machine Learning Research · 摘要：Missing data problems, such as missing modalities in multi-modal brain MRI and missing slices in cardiac MRI, pose significant challenges in clinical practice. Existing methods rely on external guidance to supply detailed missing-state information for instructing generative models to synthesize missing MRIs. However, manual indicators are not always available or reliable in real-world scenarios due to the unpredictable nature of clinical environments. Moreover, these explicit masks are not informative enough to provide guidance for improving semantic consistency. In this work, we argue that g… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/JK-Liu7/CoPeDiT · OpenReview ID：DCaolE9oBN

期刊：Transactions on Machine Learning Research · 摘要：This paper proposes to perform image restoration through a state-dependent mean-reverting forward diffusion (FoD) process. In contrast to traditional diffusion-based approaches that rely on a coupled forward-backward diffusion scheme, FoD directly learns image restoration through a single forward diffusion process, yielding a simple yet efficient framework. The core of FoD is a state-dependent stochastic differential equation (SDE) that involves a mean-reverting term in both the drift and diffusion functions. This mean-reverting structure drives the low-quality data toward the clean endpoint… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/Algolzw/FoD · OpenReview ID：Eq9k6Va3hY

期刊：Transactions on Machine Learning Research · 摘要：Long-horizon tool-using tasks sometimes benefit from revisiting earlier subtasks, but explicit revisitation also adds routing, coordination, and token cost. We study complete cyclic subtask graphs for large language model (LLM) agents: a workflow controller in which executable subtasks are fully connected and a unified state-analysis-and-routing agent selects transitions from natural-language criteria. We evaluate task-specific (Spec-Cyc) and benchmark-generic (Gen-Cyc) cyclic graphs on TextCraft, ALFWorld, and Finance-Agent against ReAct and dependency-directed acyclic workflows. The results… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/luaygz/cyclic_task_graphs · OpenReview ID：FAkarhXCfI

期刊：Transactions on Machine Learning Research · 摘要：LLM judges are increasingly placed inside an agent's loop, scoring the agent's own attempts and re-prompting until one passes. We show this quietly corrupts measurement: retry-until-PASS is optional stopping against a noisy classifier—it keeps drawing until the judge slips—so the reported pass rate is a biased estimator of true success, upward in the pass-prone retry regimes of interest (and downward under conservative rules such as strict rubrics or unanimous juries). We make this exact. The cap-$K$ gate is a binary classifier with closed-form sensitivity/specificity, and its bias is governe… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：J2Yg9vJcYb

期刊：Transactions on Machine Learning Research · 摘要：Language models have consistently grown to compress more world knowledge into their parameters, but the knowledge that can be pretrained into them is upper-bounded by their parameter size. Especially the capacity of Small Language Models (SLMs) is limited, leading to factually incorrect generations. This problem is often mitigated by giving the SLM access to an outside source: the ability to query a larger model, documents, or a database. Under this setting, we study the fundamental question of \emph{which tokens an SLM can and should learn} during pretraining, versus \emph{which ones it shou… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：JxVxBa3wO5

期刊：Transactions on Machine Learning Research · 摘要：Subjective evaluation tasks including critical analysis and rating remain at the top of Bloom’s Taxonomy. These have emerged as new pathways for evaluating Language Models (LMs) wherein correctness is relative. While LMs present diverse and human-aligned opinions on such tasks, their confidence and reliability in opinions remains unexplored. We take a deeper look at the reliability of LMs for subjective evaluations by selecting one such task of focus group surveys. LMs act as participants by completing survey questionnaires of diverse physical products. Participants must verbalize their opini… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/karush17/focus-groups · OpenReview ID：NGuOZYQZBq

期刊：Transactions on Machine Learning Research · 摘要：Language models can treat semantically distinct inputs as interchangeable at the representation level, creating blind spots that standard sparse autoencoder (SAE) objectives do not target. In safety-critical settings such as clinical dosage extraction, legal clause interpretation, or financial amount verication, such failures can propagate silently into downstream decisions. We study this problem as one of feature-basis orientation. Building on a symmetry-matching view from algebraic error-detection theory, we add a diferentiable V_Gini regularisation term to SAE training that penalises uneve… · 篇幅：Long submission (more than 12 pages of main content) · 代码：https://github.com/csabibme/blind_spot_reduction_SAE_TMLR · OpenReview ID：NWWpKC9CZH

期刊：Transactions on Machine Learning Research · 摘要：Class imbalance has been a critical bottleneck in classification problems, undermining a classifier's identification of minority instances. Data augmentation provides an effective solution by oversampling the minority. Extant methods often generate samples through duplication, perturbation, or interpolation, largely relying on the assumption of local smoothness of the data space to ensure synthetic data reliability. Alternatively, generative models are leveraged for data learning and synthesis. However, both approaches encounter significant limitations in tabular data, primarily due to data h… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/1836897243/TreeSmote · OpenReview ID：OelOS8cbBY

期刊：Transactions on Machine Learning Research · 摘要：Downscaling is essential for generating the high-resolution climate data needed for local planning, but traditional methods remain computationally demanding. Recent years have seen impressive results from AI downscaling models, particularly diffusion models, which have attracted attention due to their ability to generate ensembles and overcome the smoothing problem common in other AI methods. However, these models typically remain computationally intensive. We introduce a Hierarchical Diffusion Downscaling (HDD) model, which introduces an easily-extensible hierarchical sampling process to the… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：OhTYgFpMU2

期刊：Transactions on Machine Learning Research · 摘要：Recently, large language models (LLMs) have demonstrated strong performance, ranging from simple to complex tasks. However, while large models achieve remarkable results across diverse tasks, they often incur substantial monetary inference cost, making frequent use impractical for many applications. In contrast, small models are often freely available and easy to deploy locally, but their performance on complex tasks remains limited. This trade-off raises a natural question: how can small and large models efficiently collaborate to combine their complementary strengths? To bridge this trade-o… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/prinwinter/cope · OpenReview ID：RPzbeL0koP

期刊：Transactions on Machine Learning Research · 摘要：We investigate the finite-time convergence properties of Temporal Difference (TD) learning with linear function approximation, a cornerstone of reinforcement learning. We are interested in the so-called ``robust'' setting, where the convergence guarantee does not depend on the potential function's minimal curvature. While prior work has established convergence guarantees in this setting, these results typically rely on the artificial assumption that each iterate is projected onto a bounded set. Removing such a condition was left as an open problem by Bhandari et al. (COLT'18), hypothesizing t… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：Tj1B5WDyt8

期刊：Transactions on Machine Learning Research · 摘要：It remains a critical challenge to adapt policies across domains with mismatched dynamics in reinforcement learning (RL). In this paper, we study cross-domain offline RL, where an offline dataset from another similar source domain can be accessed to enhance policy learning upon a target domain dataset. Directly merging the two datasets may lead to suboptimal performance due to potential dynamics mismatches. Existing approaches typically mitigate this issue through source domain transition filtering or reward modification, which, however, may lead to insufficient exploitation of the valuable s… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/soulchanging/STC · OpenReview ID：TupiNRpgHw

期刊：Transactions on Machine Learning Research · 摘要：In reinforcement learning (RL), spatial planning is often mediated through rasterized observations processed by convolutional networks, even when the underlying task is continuous and geometric. This discretization can introduce aliasing and obscure topological structure, increasing the difficulty of the spatial problem. We study a hierarchical set-valued geometry-first observation interface for sparse-reward navigation that operates directly on triangulated obstacle geometry. This interface uses learned multi-token aggregation to compress variable-sized geometry into a bounded fixed-size rep… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/Andresrzr/HGARL/tree/main/rl · OpenReview ID：U10DFNcMrW

期刊：Transactions on Machine Learning Research · 摘要：This work considers recovering a sequence of low-rank matrices from undersampled measurements, where the underlying subspace varies across samples over time. Existing works involve concatenating all of the samples from each time point to recover the underlying matrix under the assumption that the data are well-approximated by a single, static subspace. However, this assumption is inappropriate for applications where the best low-rank approximations vary over time. To address this issue, we propose a Riemannian block majorize minimization algorithm that constrains the time-varying subspaces as… · 篇幅：Long submission (more than 12 pages of main content) · OpenReview ID：W0ZNwXoBeh

期刊：Transactions on Machine Learning Research · 摘要：Nutritional interventions are important for managing chronic health conditions, but current computational methods provide limited support for personalized dietary guidance. We identify three key gaps: (1) dietary pattern studies often ignore real-world constraints such as socioeconomic status, comorbidities, and limited food access; (2) recommendation systems rarely explain why a particular food helps a given patient; and (3) no unified benchmark evaluates methods across the connected tasks needed for nutritional interventions. We introduce GLEN-Bench, the first comprehensive graph-language b… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：XMUbOKWsem

期刊：Transactions on Machine Learning Research · 摘要：Causal Bayesian Optimization (CBO) integrates causal inference with Bayesian optimization to enable sample-efficient intervention selection in systems governed by causal structure. This survey provides a comprehensive and systematic review of the CBO landscape, organizing the growing literature through a unified BO-loop perspective that reveals how causal assumptions shape four core components: intervention search spaces, surrogate construction, acquisition design, and decision policies. We organize methods along recurring design axes, including graph and system-knowledge assumptions, environ… · 篇幅：Long submission (more than 12 pages of main content) · 代码：https://github.com/chenfeng-huang/CBO-Benchmark-TMLR-2026 · OpenReview ID：XT6DC37m5I

期刊：Transactions on Machine Learning Research · 摘要：Recent work has analyzed temporal-difference (TD) learning with target networks through an optimization view and established linear convergence under a force-dominance condition, but these results typically rely on global smoothness, i.e., a uniform upper bound on curvature. This assumption can fail even when the inner problem is well posed, since curvature encountered during training can grow with the scale of TD-residual-induced gradients. We retain the stabilized regime in which the inner problem is strongly convex in the optimization variable, in order to isolate upper-curvature growth ef… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：Xvyehi4izc

期刊：Transactions on Machine Learning Research · 摘要：We introduce SLayR, Scene Layout Generation with Rectified flow, a novel transformer-based model for text-to-layout generation, which can integrate into a complete text-to-image pipeline. SLayR addresses a domain in which current text-to-image pipelines struggle: generating scene layouts that are of significant variety and plausibility, when the given prompt is ambiguous and does not provide constraints on the scene. In this setting, SLayR surpasses existing baselines, including LLMs. To accurately evaluate the layout generation, we introduce a new benchmark suite, including numerical metrics… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：YGsQxG5ubd

期刊：Transactions on Machine Learning Research · 摘要：We present and test the largest benchmark for vericoding, LLM-generation of formally verified code from formal specifications — in contrast to vibe coding, which generates potentially buggy code from a natural language description. Our benchmark contains 12,504 formal specifications, with 3,029 in Dafny, 2,334 in Verus/Rust and 7,141 in Lean. Of these, 6,174 are new unseen problems. We find vericoding success rates of 27% in Lean, 44% in Verus/Rust and 82% in Dafny using off-the-shelf LLMs. Adding natural-language descriptions does not significantly improve performance. We also find that LLM… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：Zgh5kpGAm8

期刊：Transactions on Machine Learning Research · 摘要：The ability to robustly maintain and update continuous variables is a hallmark of working memory. While classical continuous attractor networks suffer from severe fine-tuning fragility, standard artificial recurrent neural networks (RNNs) like GRUs and LSTMs typically fail to stably learn continuous manifolds, instead shattering the state space into discretized point attractors. To bridge this gap, we draw inspiration from divisive normalization, a canonical neural computation widely observed across cortical circuits, and propose the Recurrent Divisive Normalization Network (RDNN), a minimal… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/mizunashi-sh/RDNN · OpenReview ID：a43l19lyfC

期刊：Transactions on Machine Learning Research · 摘要：Time series reasoning is crucial to decision-making in diverse domains, including finance, energy, and scientific discovery. While existing time series foundation models (TSFMs) can capture low-level dynamic patterns and provide accurate forecasting, further analysis usually requires additional background knowledge and sophisticated reasoning, which are lacking in most TSFMs but can be achieved through Large Language Models (LLMs). On the other hand, without expensive post-training, LLMs often struggle with the numerical understanding of time series data. Although it is intuitive to integrate… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：d6TD0f2xXq

期刊：Transactions on Machine Learning Research · 摘要：Recurrent neural networks are often used as mechanistic models of learning and control, but closed-loop training creates reproducibility challenges because a model's actions alter future inputs. We conduct a claim-level reproducibility study of Ger and Barak's closed-loop RNN learning dynamics, testing independent implementation, seed variation, protocol perturbations, coupled-system diagnostics, and architecture/task transfer. Under a main-text-aligned double-integrator protocol, the trajectory-level peak, not a persistent final gap, reproduces strongly: 50/50 paired seeds show the post-init… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/aaravsinhaofficial/closed-loop-rnn-audit · OpenReview ID：e0pSqIqwXO

期刊：Transactions on Machine Learning Research · 摘要：Many modern deep learning applications require balancing multiple objectives that are often conflicting. Examples include multi-task learning, fairness-aware learning, and the alignment of Large Language Models (LLMs). This leads to multi-objective deep learning, which tries to find optimal trade-offs or Pareto-optimal solutions by adapting mathematical principles from the field of Multi-Objective Optimization (MOO). However, directly applying gradient-based MOO techniques to deep neural networks presents unique challenges, including high computational costs, optimization instability, and the… · 篇幅：Long submission (more than 12 pages of main content) · OpenReview ID：eCUcXXH3PS

期刊：Transactions on Machine Learning Research · 摘要：Computer-use agents (CUAs) hold promise for automating everyday digital tasks, but their performance on long-horizon, complex problems remains unreliable. Single-rollout execution is brittle, with small errors compounding over time and leading to high variance in outcomes. While prior work has attempted to scale within a single rollout, such approaches have yielded limited gains. Scaling over multiple rollouts offers a more promising alternative but doing so effectively is challenging due to the difficulty of evaluating and selecting among long-horizon agent behaviors. We introduce Behavior J… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：eve4jBYa8D

期刊：Transactions on Machine Learning Research · 摘要：``How long can I live and remain free of cancer?'' is often the first question a patient asks after receiving a cancer diagnosis and treatment. Accurate survival prediction helps alleviate psychological distress and supports risk stratification and personalized treatment planning. Recent survival prediction frameworks have shown strong performance using computed tomography (CT) images. However, variations in imaging acquisition introduce out-of-distribution (OOD) samples caused by covariate shifts that undermine model reliability. Despite this challenge, to our knowledge, no existing benchmar… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/WenjieZhao1/CURE-OOD · OpenReview ID：fziI7nE1vO

期刊：Transactions on Machine Learning Research · 摘要：The recently proposed Microcanonical Hamiltonian Monte Carlo algorithm has not yet been studied in detail from a thermodynamic point of view; this work aims to fill that gap. We demonstrate how thermodynamical state variables and potentials can be derived and thereby demonstrate that the construction of the algorithm formally represents a microcanonical thermodynamic ensemble. In particular, we demonstrate (analytically and numerically) that the algorithm fulfils the Helmholtz theorem, an alternative formulation of the first law of thermodynamics. Furthermore, we construct a new sampling algo… · 篇幅：Long submission (more than 12 pages of main content) · OpenReview ID：jbVVrks50D

期刊：Transactions on Machine Learning Research · 摘要：Value-based offline RL methods are prone to overestimate the values of out-of-distribution (OOD) actions, and this is often addressed by regularizing the action-value function in the Bellman update. However, existing regularization methods can suffer from being too conservative, which can arise from over-penalizing the values for both in-distribution actions and out-of-support actions. We present a new regularization method for offline value-based methods, called Density-Scaled (DS) regularization, which penalizes the value function based on the relative action density of the behavior policy.… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/jackyxie5/density-scaled-regularization · OpenReview ID：nDPrzkQTj1

期刊：Transactions on Machine Learning Research · 摘要：Noisy labelled datasets are generally inexpensive compared to clean labelled datasets, and the same is true for graph data. In this paper, we propose a denoising technique DeGLIF: Denoising Graph Data using Leave-One-Out Influence Function. DeGLIF uses a small set of clean data and the leave-one-out influence function to make label noise robust node-level prediction on graph data. Leave-one-out influence function approximates the change in the model parameters if a training point is removed from the training dataset. Recent advances propose a way to calculate the leave-one-out influence funct… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/pintu-dot/DeGLIF · OpenReview ID：pcs5DmBtUJ

期刊：Transactions on Machine Learning Research · 摘要：Video Large Language Models (Vid-LLMs) have recently achieved strong performance across a wide range of video understanding tasks, including question answering, captioning, and multimodal reasoning. However, these models frequently produce outputs that are not faithfully grounded in the underlying video content, a phenomenon commonly referred to as hallucination. Compared with hallucination in text-only or image-based models, hallucination in video understanding is further complicated by temporal dynamics, motion interpretation, long-context dependencies, and event-level reasoning. In this su… · 篇幅：Long submission (more than 12 pages of main content) · OpenReview ID：qbO71rVrIG

期刊：Transactions on Machine Learning Research · 摘要：Numerical preprocessing remains a critical component of tabular deep learning, as the representation of continuous features can strongly affect downstream performance. We systematically study spline-based numerical encodings, including B-splines, M-splines, and integrated splines (I-splines), under uniform, quantile-based, target-aware, and learnable-knot placement. For the learnable variants, we adopt a differentiable knot parameterization that enables stable end-to-end optimization of knot locations jointly with the backbone. We evaluate these encodings on a diverse collection of public reg… · 篇幅：Long submission (more than 12 pages of main content) · 代码：https://github.com/mkumar73/tdl-numerical-encodings/ · OpenReview ID：str7wQt9Qc

期刊：Transactions on Machine Learning Research · 摘要：Large language model pretraining is compute-intensive, yet many tokens contribute marginally to learning, resulting in inefficiency. We introduce Efficient Selective Language Modeling (ESLM), an online, risk-aware batch selection algorithm that improves training efficiency and distributional robustness. ESLM operates in two phases: (i) instance-level selection via a shallow early-exit model pass that computes proxy per-instance statistics (e.g., loss or entropy) and retains data points using value-at-risk thresholding; and (ii) loss shaping with token-level selection via risk-aware thresholdi… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/melisilaydabal/eslm · OpenReview ID：vcynsg6rMN

期刊：Transactions on Machine Learning Research · 摘要：Proteins are fundamental to biological processes, with their function determined by the complex interplay between the amino acid sequence and the three-dimensional structure. Developing generative models capable of understanding this intrinsically multi-modal relationship is crucial for fields like drug discovery and protein engineering. Existing models often rely on a multi-stage training process where autoencoders that tokenize data into latent representations are trained in a first stage. Secondly, a generative model is trained on the latent representation of the autoencoder(s), i.e., gene… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：wPfw7GkMns

期刊：Transactions on Machine Learning Research · 摘要：Transformers are difficult to optimize with stochastic gradient descent (SGD) and largely rely on adaptive optimizers such as Adam. Despite extensive efforts, the mechanisms behind Adam's advantage over SGD in Transformer optimization are still not fully understood. In this study, we analyze the optimization of Transformer models in the fine-tuning setting through the lens of gradient heterogeneity, defined as the variation in gradient norms across parameter blocks. We provide a theoretical analysis showing that gradient heterogeneity, together with Hessian heterogeneity, degrades the converg… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/tom4649/gradient-heterogeneity · OpenReview ID：wZJcQb5m1e

期刊：Transactions on Machine Learning Research · 摘要：We study fixed-budget temporal memory for streams of probability distributions. The proposed representation is a piecewise-linear density protocol on a replay interval $[0,1]$: recent experience is stored near $t=1$, older experience is represented by intermediate-time marginals, and new experience is incorporated by a deterministic \emph{Compress--Add--Smooth} (CAS) recursion. In the Gaussian-mixture instantiation considered here, each protocol node stores a labeled $K$-component Gaussian mixture in $d$ dimensions, and each daily update costs $O(LKd^2)$ arithmetic operations for a fixed temp… · 篇幅：Long submission (more than 12 pages of main content) · 代码：https://github.com/mchertkov/compress-add-smooth · OpenReview ID：wjoixYG0mC

期刊：Transactions on Machine Learning Research · 摘要：We propose TimePre, a simple framework that unifies the efficiency of Multilayer Perceptron (MLP)-based models with the distributional flexibility of Multiple Choice Learning (MCL) for Probabilistic Time-Series Forecasting (PTSF). Stabilized Instance Normalization (SIN), the core of TimePre, is a normalization layer that explicitly addresses the trade-off among accuracy, efficiency, and stability. SIN stabilizes the hybrid architecture by correcting channel-wise statistical shifts, thereby preventing the hypothesis collapse that otherwise destabilizes this combination. Extensive experiments o… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/LyCharles/TimePre/ · OpenReview ID：yQLnvkJMbP

期刊：Transactions on Machine Learning Research · 摘要：Video world models have attracted significant attention for their ability to produce high-fidelity future visual observations conditioned on past observations and navigation actions. However, achieving temporally and spatially consistent generation over long horizons remains an open challenge: existing approaches either compress past frames at fixed rates based on temporal proximity, discarding spatially critical information, or retrieve only a handful of relevant frames without increasing the total amount of retained history. In this paper, we propose WorldPack, a video world model that intr… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：zJuiG3PiNJ

期刊：Transactions on Machine Learning Research · 摘要：Multi-domain thinking verifiers trained via Reinforcement Learning with Verifiable Rewards (RLVR) are a cornerstone of modern post-training. However, their adoption in code generation has lagged behind that of execution feedback due to the prohibitive costs of the full RLVR pipeline. In this work, we ablate three primary choices along the performance-cost trade-off in RLVR: intermediate thinking traces, learning from negative samples, and on-policy training. We introduce Aletheia, a controlled, execution-grounded testbed to facilitate a decontaminated analysis of code verifier training recipe… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/insait-institute/aletheia/ · OpenReview ID：3rVrBGp0mr

期刊：Transactions on Machine Learning Research · 摘要：AI tools are suggested as solutions to assist public agencies with heavy workloads. In public defense---where a constitutional right to counsel meets the complexities of law, overwhelming caseloads, and constrained resources---practitioners face especially taxing conditions. Yet, there is little evidence of how AI could meaningfully support defenders' day-to-day work. In partnership with the New Jersey Office of the Public Defender, we develop the OPD Resource Library, a retrieval tool which surfaces relevant appellate briefs to streamline legal research and writing. We show that existing ret… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/princeton-polaris-lab/PublicDefenderRetrieval · OpenReview ID：HnbKQGRnDt

期刊：Transactions on Machine Learning Research · 摘要：Reward fine-tuning of diffusion and flow models and sampling from tilted or Boltzmann distributions can both be formulated as stochastic optimal control (SOC) problems, where learning an optimal generative dynamics corresponds to optimizing a control under SDE constraints. In this work, we revisit and generalize \emph{Adjoint Matching}, a recently proposed SOC-based method for learning optimal controls, and place it on a rigorous footing by deriving it from the \emph{Stochastic Maximum Principle} (SMP). We formulate a general Hamiltonian adjoint matching objective for SOC problems with contro… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/frankhan91/smp-adjoint-matching · OpenReview ID：tR5VsdQFhK

期刊：Transactions on Machine Learning Research · 摘要：While Large Language Models (LLMs) have substantially advanced text-to-code generation, many real programming tasks specify intent through visual artifacts such as screenshots, charts, and videos. These tasks require models to connect visual perception to executable programs, as correctness depends not only on syntax but also on layout, data semantics, and domain-specific constraints that apply after execution. This survey reviews Multimodal Code Intelligence, covering systems that generate, edit, refine, or reason with code under visually grounded inputs and outputs. We first formulate the f… · 篇幅：Long submission (more than 12 pages of main content) · 代码：https://github.com/xjywhu/Awesome-Multimodal-LLM-for-Code · OpenReview ID：pn2sdu3Vrf

期刊：Transactions on Machine Learning Research · 摘要：In the evolving landscape of deep learning, selecting the best pre-trained models from a growing number of choices is a challenge. Transferability scorers propose an efficient alternative to this challenge by calculating a proxy to rank a pool of pre-trained model candidates. Despite their promise, the field currently lacks standardized evaluation protocols, consistent baselines, and reproducible methodologies. This has led to contradictory findings across studies, with the best scorer in one study ranking among the worst in another. In this work, we introduce a benchmark for transferability… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/VirtualSpaceman/tmlr-transferability-benchmark · OpenReview ID：3i2ZRk8GDN

期刊：Transactions on Machine Learning Research · 摘要：Large language models (LLMs) struggle with multi-step mathematical reasoning, for which inference-time scaling—via sequential or parallel scaling—has emerged as a promising strategy. While recent advances have focused on sequential scaling, we revisit the less-explored parallel scaling approach, verifier-guided beam search, to examine its limitations. In this paper, we argue that its strength is, paradoxically, also its limitation: verifiers can boost performance under limited sample sizes by elevating promising reasoning paths, yet the same mechanism can also hide or cut off the valid paths… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：D5VKbIzlrR

期刊：Transactions on Machine Learning Research · 摘要：This paper presents a method to infer a 3D face avatar model from a single arbitrarily posed image, using the 3D Gaussian Splatting (3DGS) framework. Inference of a full 3DGS face model from one image is a highly ill-posed problem, requiring the estimation of hundreds of thousands, often well over a million, per-Gaussian appearance and structural parameters. To address this challenge, we draw inspiration from the classical morphable face models literature, in which individual identities are well-described as compact deformations (residuals) with respect to a canonical template face model, the… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/hliang2/FastAvatar · OpenReview ID：4WnbCj0v0K

期刊：Transactions on Machine Learning Research · 摘要：Foundation models have achieved remarkable success across many domains, relying on pretraining over vast amounts of data. Graph-structured data often lacks the same scale as unstructured data, making the development of graph foundation models challenging. In this work, we propose Foundation-Informed Message Passing (FIMP), a message-passing framework that repurposes existing pretrained non-textual foundation models for graph-based tasks in multiple capacities, including tokenization, representation extraction, and weight initialization. We show that the self-attention layers of foundation mod… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：fj7sjOwtXc

期刊：Transactions on Machine Learning Research · 摘要：Differential privacy (DP) is a formal privacy framework that enables training machine learning (ML) models while protecting individuals' data. As pointed out by prior work, ML models are part of larger systems, which can lead to so-called privacy side-channels even if the model training itself is DP. We identify the output label space of a classification model as such a privacy side-channel and show a concrete privacy attack that exploits it. The side-channel becomes highly relevant in continual learning (CL), where the output label space changes over time. To reason about privacy guarantees… · 篇幅：Long submission (more than 12 pages of main content) · 代码：https://github.com/TrustworthyMLHelsinki/dp-continual-learning · OpenReview ID：ZshFgRQWrm