OpenReview · 实时热榜

期刊：Transactions on Machine Learning Research · 摘要：In this work, we provide a systematic survey of Discrete Diffusion Language Models (dLLMs) and Discrete Diffusion Multimodal Language Models (dMLLMs). Unlike autoregressive (AR) models, dLLMs and dMLLMs adopt a multi-token, parallel decoding paradigm using full attention and a denoising-based generation strategy. This paradigm naturally enables parallel generation, fine-grained output control, and dynamic perception. These capabilities are previously difficult to achieve with AR models. A growing number of industrial-scale proprietary d(M)LLMs, as well as a large number of open-source academi… · 篇幅：Long submission (more than 12 pages of main content) · OpenReview ID：0DsqnkP8Cp

期刊：Transactions on Machine Learning Research · 摘要：AlphaZero has demonstrated that a neural-guided Monte Carlo Tree Search can achieve superhuman performance, but strong play does not imply perfect play. We study this gap in two oracle-evaluable domains with contrasting structure: Connect Four, a solved partisan game with exact game-theoretic values, and Chomp, an impartial game whose optimal play is governed by Grundy-number structure. Under a unified self-play $+$ MCTS pipeline, we compare vanilla AlphaZero, a multi-frame variant (limited to Chomp), and an AlphaZero Auxiliary Loss (AZAL) that adds oracle-derived policy supervision. We find… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/Brentkong/AlphaZero-in-Sparsely-Rewarded-Games-Limits-and-Auxiliary-Supervision · OpenReview ID：1z0CnFiJKg

期刊：Transactions on Machine Learning Research · 摘要：Large language models are known to hallucinate, generating linguistically plausible but incorrect answers to questions. Uncertainty quantification has been proposed as a strategy to detect such behaviour, but existing methods lack a unified framework to assess reliability at both the prompt and answer level. We introduce a geometric framework which quantifies language model uncertainty at both levels by explicitly modelling a prompt-conditioned semantic distribution in answer embedding space. Our approach is black-box and sampling-based; we generate multiple answers per prompt, and use archet… · 篇幅：Long submission (more than 12 pages of main content) · OpenReview ID：5UVv7gkgUD

期刊：Transactions on Machine Learning Research · 摘要：We study what provable privacy attacks can be shown for trained 2-layer ReLU neural networks, focusing on two types of attacks: membership inference and data reconstruction. We prove that theoretical results on the implicit bias of 2-layer neural networks can be used to provably identify with high probability whether a given point was used in the training set in a high-dimensional setting, and can also be used to construct a set of which at least a constant fraction are training points in a univariate setting. To the best of our knowledge, our work is the first to show provable vulnerabilitie… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/guy120494/Provable-Privacy-Attacks-on-Trained-Shallow-Neural-Networks · OpenReview ID：6lCkCCw2ds

期刊：Transactions on Machine Learning Research · 摘要：The concept of Markov chain Monte Carlo (MCMC) cycles, an analogy to cyclic processes in heat engines, is presented in order to examine Bayesian inference problems. In this effort, we develop adaptive ensemble schedulers that allow the tuning of external parameters of a Bayesian canonical ensemble during an MCMC run, realising the MCMC cycles in practice. We run these cycles on different statistical models. As a fundamental insight, we find (both theoretically and in practice) that such systems can produce a non-zero net work output if and only if the considered model is non-Gaussian. As such… · 篇幅：Long submission (more than 12 pages of main content) · OpenReview ID：88KWqihymD

期刊：Transactions on Machine Learning Research · 摘要：Model parallelism (MP) is a widely adopted paradigm for scaling large language model (LLM) training across multiple nodes. Yet, existing methods mainly rely on first-order optimization, which suffer from two key bottlenecks: high communication overhead due to frequent transmission of activations and gradients, and substantial memory consumption caused by caching these intermediate states. Zeroth-order (ZO) optimization offers a compelling alternative by eliminating explicit gradient computation and storage, naturally reducing communication and memory costs. Despite these advantages, ZO method… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：9wpm4fbBJI

期刊：Transactions on Machine Learning Research · 摘要：Hyperparameter Optimization (HPO) is crucial to developing well-performing machine learning models. In order to ease prototyping and benchmarking of HPO methods, we propose carps, a benchmark framework for Comprehensive Automated Research Performance Studies allowing to evaluate N optimizers on M benchmark tasks. In this first release of carps, we focus on the four most important types of HPO task types: blackbox, multi-fidelity, multi-objective, and multi-fidelity-multi-objective. With 3 336 tasks from 5 community benchmark collections and 28 variants of 9 optimizer families, we offer the bi… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://www.github.com/automl/CARP-S · OpenReview ID：AuA8m4I6zI

期刊：Transactions on Machine Learning Research · 摘要：While many state-of-the-art techniques reconstruct incomplete time series datasets by replacing gaps with modeled estimates, we propose an alternative: encode missing values as an extremal sentinel value, allowing a prediction model to learn from the pattern of missingness. Incomplete data is a common problem in real-world time series forecasting, particularly in environmental monitoring where sensor failures can cause continuous gaps in data. This paper proposes the \emph{Min-Std} method, a novel computationally efficient imputation strategy that encodes missingness in representation-space w… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：DMmCMIrrez

期刊：Transactions on Machine Learning Research · 摘要：Neural networks trained with gradient descent exhibit the Edge of Stability (EoS), where Hessian sharpness rises toward $2/\eta$ and the loss oscillates non-monotonically. This paper asks whether that instability persists under Fisher-preconditioned optimization. We analyze effective sharpness, $S_{\text{eff}} = \lambda_{\max}(F^{-1}H)$, and prove the general bound $S_{\text{eff}} \leq 1 + (\epsilon + \delta)/\mu_{\min}(F)$, where $\epsilon = \|H - G\|_2$ is residual curvature, $\delta = \|G - F\|_2$ is the Gauss-Newton/Fisher gap, and $\mu_{\min}(F)$ is the minimum Fisher eigenvalue. The ide… · 篇幅：Long submission (more than 12 pages of main content) · 代码：https://github.com/990aa/sbesfpgd · OpenReview ID：EabuvggEbb

期刊：Transactions on Machine Learning Research · 摘要：Large Language Models (LLMs) encode factual knowledge within hidden parametric spaces that are difficult to inspect or control. While Sparse Autoencoders (SAEs) can decompose hidden activations into more fine-grained, interpretable features, they often struggle to reliably align these features with human-defined concepts, resulting in entangled and distributed feature representations. To address this, we introduce AlignSAE, a method that aligns SAE features with a predefined ontology through a "pre-train, then post-train" curriculum. After an initial unsupervised training phase, we apply supe… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/yminglai/AlignSAE · OpenReview ID：I9UjKxW4nq

期刊：Transactions on Machine Learning Research · 摘要：We present a reinforcement-learning agent that solves symbolic equations step by step --- both nonlinear \emph{closed} equations (radicals, exponentials, trigonometric) and, for the first time, a controlled class of \emph{restricted-open} families requiring a change of variables (CoV) such as completing the square. We cast algebra as an MDP with a dynamic action space and a tree-structured policy (TreeMLP). The \emph{main} policy learns the full solution procedure from reward alone, with no supervised solution traces; the CoV substitution itself comes from a supervised generator interchangeab… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：JlC5BJiikD

期刊：Transactions on Machine Learning Research · 摘要：Modern large language model (LLM) systems increasingly rely on multi-turn pipelines that are composed of multiple task-specific adapters, yet existing serving frameworks remain inefficient, incurring substantial recomputation overhead when switching between adapters. We present the first LLM serving engine that supports cross-model prefix cache reuse between base and adapted models via Activated LoRA (aLoRA), enabling efficient and fine-grained adapter switching during inference. Our design extends the vLLM framework by introducing base-aligned block hashing and activation-aware masking withi… · 篇幅：Long submission (more than 12 pages of main content) · 代码：https://github.com/tdoublep/vllm/tree/alora · OpenReview ID：Q8nCBmOkyn

期刊：Transactions on Machine Learning Research · 摘要：As AI systems become more powerful, there is growing concern that they may act in ways misaligned with human interests. However, this concern presupposes that AI models have consistent preferences and that these preferences influence their behavior. These claims have yet to be rigorously tested. Here, we examine one precondition for misalignment: whether LLM preferences predict downstream behavior. The questions raised in this paper are theoretically motivated by the concept of "sandbagging" from the misalignment literature, though sandbagging itself is not directly measured here. We evaluate… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：RmWSM3sFOr

期刊：Transactions on Machine Learning Research · 摘要：Predicting generalization from quantities available before target-test evaluation remains a central challenge in deep learning. The systematic benchmark of Jiang et al. (2020) evaluated many generalization measures, but it focused on independent and identically distributed (IID) settings. We revisit this problem for image classifiers evaluated under controlled corruptions and perturbations. Our study uses CIFAR-10-C/P, where the label space and task remain fixed while the input images are degraded or perturbed. This setting also allows us to revisit the robustness concerns raised by Dziugaite… · 篇幅：Long submission (more than 12 pages of main content) · OpenReview ID：X4RoujAYnY

期刊：Transactions on Machine Learning Research · 摘要：Bayesian optimization (BO) is an effective technique for black-box optimization. However, its applicability is typically limited to moderate-budget problems due to the cubic complexity of fitting the Gaussian process (GP) surrogate model. In large-budget scenarios, directly employing the standard GP model faces significant challenges in computational time and resource requirements. In this paper, we propose Gradient-based Sample Selection Bayesian Optimization (GSSBO), a subset-maintenance approach designed to enhance the computational efficiency of BO. Here, ``gradient-based'' refers to resp… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/Qiyu-Wei/GSSBO-Official.git · OpenReview ID：Ysr1zUeuxz

期刊：Transactions on Machine Learning Research · 摘要：Quantum process tomography reconstructs an unknown quantum channel from finite measurement counts. For sample-based uncertainty reporting, the sampled candidate channels should also remain physically valid, meaning completely positive and trace preserving (CPTP). We introduce a matrix-free, CPTP-by-construction reconstruction-and-sampling framework evaluated in a controlled full-basis simulator setting. Its primary contribution combines indexed residual evaluation, which avoids materializing the dense Choi design matrix, with normalized Kraus factors that construct physical candidate channels… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/DDDIGHE/matrix-free-cptp-qpt · OpenReview ID：f2LcKcJL2m

期刊：Transactions on Machine Learning Research · 摘要：Conditional generative models map input variables to complex, high-dimensional distributions, enabling realistic sample generation in a diverse set of domains. A critical challenge with these models is the absence of calibrated uncertainty, which undermines trust in individual outputs for high-stakes applications. To address this issue, we propose a systematic conformal prediction approach tailored to conditional generative models, leveraging density estimation on model-generated samples. We introduce a novel method called CP4Gen, which utilizes cluster-based density estimation to construct p… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/Earth-Intelligence-Lab/CP4GenerativeModel · OpenReview ID：goxeVsh9Po

期刊：Transactions on Machine Learning Research · 摘要：Adapting policies in piecewise stationary environments - where the underlying properties remain stable for periods but abruptly change at unknown points - remains a challenge in reinforcement learning (RL). Addressing this problem using context-based offline meta-RL, which enables generalization to new online tasks from offline data, is particularly appealing, as it avoids the risks associated with online exploration. These methods encode transition history (the context) into a task representation and condition the policy and value function to enable generalization. We show that existing appr… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/MohammadrezaNakhaei/CAMEL-official · OpenReview ID：gp1mAySr25

期刊：Transactions on Machine Learning Research · 摘要：Large Language Models (LLMs) have proven effective in understanding and generating extremely long contexts. Recently, linear attention mechanisms have garnered significant attention, as they can largely reduce the quadratic computational complexity of traditional attention mechanisms to linear complexity relative to token sequence length, thus balancing effectiveness and efficiency in LLM training and inference. This survey mainly focuses on a broad spectrum of linear attention techniques, including traditional linear attention methods, state space models (SSMs), and linear recurrent neural n… · 篇幅：Long submission (more than 12 pages of main content) · OpenReview ID：ilkVX8aGmQ

期刊：Transactions on Machine Learning Research · 摘要：Data rehearsal has emerged as a leading approach for mitigating catastrophic forgetting in Continual Reinforcement Learning (CRL). However, existing work remains confined to policy gradient frameworks, regularizing only actors due to the performance degradation incurred by critic regularization. This actor-centric approach overlooks the potential of data rehearsal for value function approximation. Moreover, existing evaluations in CRL rarely consider multi-cyclic environments where task sequences repeat, a critical real-world scenario that exacerbates forgetting and plasticity. We investigate… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/Bpoole908/Qreg · OpenReview ID：wYayhflqqR

期刊：Transactions on Machine Learning Research · 摘要：In classification with tabular data, one often utilizes tree-based models. Those can be competitive with deep neural networks on tabular data and, under some conditions, explainable. The explainability depends on the tree's depth and the accuracy of each leaf. Decision trees containing leaves with unbalanced accuracy can provide misleading explanations. Low-accuracy leaves provide less useful explanations to the individuals they classify. Here, we train a shallow tree that minimizes the maximum misclassification error across leaf nodes. The shallow tree provides a more useful global explanati… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/Epanemu/LeafLAT · OpenReview ID：y4FJDMpWq3

期刊：Transactions on Machine Learning Research · 摘要：Interactive prompting is an appealing approach to vision–language fusion using frozen uni-modal transformers, yet recent progress often relies on increasingly complex prompting architectures. A natural question arises: instead of refining prompt designs, can fusion be improved more effectively by directly adapting internal representations within attention layers? Our analysis, from a representation-centric perspective, suggests that within each frozen attention layer, prompt tokens have limited direct control over the value representations of original modality tokens and their query–key inter… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/YYJmay/representation-centric-multimodal-fusion · OpenReview ID：yBVwYxHxUq

期刊：Transactions on Machine Learning Research · 摘要：Inferring the causal relationships among a set of variables in the form of a directed acyclic graph (DAG) is an important but notoriously challenging problem. Recently, advancements in high-throughput genomic perturbation screens have inspired the development of methods that leverage interventional data to improve model identification. However, existing methods still suffer from poor performance on large-scale tasks and fail to quantify uncertainty. Here, we propose Interventional Bayesian Causal Discovery (IBCD), an empirical Bayesian framework that infers the causal graph by using intervent… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/bcbg-bio/IBCD · OpenReview ID：yW4T2fsf0l

期刊：Transactions on Machine Learning Research · 摘要：Change point detection (CPD) identifies abrupt and significant changes in sequential data, with applications in human activity recognition, financial markets, cybersecurity, manufacturing, and autonomous systems. Traditional CPD methods often face computational challenges in high-dimensional settings and typically provide limited explanations for detected changes, which can restrict their practical usability. This paper introduces a CPD framework that improves scalability and interpretability by leveraging the Sliced Wasserstein (SW) distance. Our contributions are fourfold: (1) we transform… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/jsve96/SWCPD_Code · OpenReview ID：4ewaiYXoiv

期刊：Transactions on Machine Learning Research · 摘要：Most existing approaches to referring segmentation achieve strong performance only through fine-tuning or by composing multiple pre-trained models, often at the cost of additional training and architectural modifications. Meanwhile, large-scale generative diffusion models encode rich semantic information, making them attractive as general-purpose feature extractors. In this work, we introduce a new method that directly exploits features—attention scores—from diffusion transformers for downstream tasks, requiring neither architectural modifications nor additional training. To systematically ev… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/Annusha/refam · OpenReview ID：5DTaxhAP4h

期刊：Transactions on Machine Learning Research · 摘要：Over-parameterized models are often vulnerable to membership inference attacks, which aim to determine whether a specific sample is included in the training set of a target. Previous weight regularization approaches typically impose uniform penalties on all parameters, leading to a suboptimal trade-off between model utility and privacy. In this work, we first show that only a small fraction of the parameters substantially impact privacy risk. Motivated by this analysis, we propose Privacy-aware Sparsity Tuning ($\textbf{PAST}$)—a novel privacy-preserving training method—by employing adaptive… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：KlqlcnwqHw

期刊：Transactions on Machine Learning Research · 摘要：Circuit discovery methods identify subgraphs that explain specific model behaviors, and structural differences between discovered circuits are commonly interpreted as evidence of distinct mechanisms. We test this assumption by drawing input tokens from bands defined by their frequency in the pretraining data while holding the task fixed. The discovered circuits appear specialized by frequency when compared structurally, but functional and representational analyses show no reliable evidence of corresponding differences in their computations. We term this mismatch phantom specialization. Using… · 篇幅：Long submission (more than 12 pages of main content) · OpenReview ID：Ur42Xx1QPv

期刊：Transactions on Machine Learning Research · 摘要：Model-based evaluation systems increasingly use panels of inexpensive LLM judges to make accept-or-escalate decisions. In factuality settings, accepting a claim because several reference-free judges agree can create a hidden risk: agreement may arise from shared false-negative blind spots rather than independent evidence. We introduce JuryProbe, an empirical consensus-risk diagnostic for reference-free factuality judge panels, paired with a calibration-based routing policy. JuryProbe estimates panel-level consensus risk from a labeled calibration probe using false-negative-only (FN-only) judg… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：dgBczhxcZY

期刊：Transactions on Machine Learning Research · 摘要：This paper presents a comprehensive review of industrial anomaly synthesis (IAS). Existing surveys on industrial anomalies mainly focus on anomaly detection, while IAS is typically treated as an auxiliary component rather than as an independent topic. However, owing to its increasing importance in data augmentation, downstream model training, and controllable industrial inspection, IAS has become a research direction of growing interest. To address the lack of a dedicated review, we survey a broad range of representative methods and organize them into four paradigms: hand-crafted synthesis, d… · 篇幅：Long submission (more than 12 pages of main content) · OpenReview ID：f9qjl5xCVW

期刊：Transactions on Machine Learning Research · 摘要：Real-world decision-making often requires optimizing multiple competing objectives simultaneously. In reinforcement learning (RL), this is typically addressed by combining reward signals into a single scalar objective via a scalarization function, which can be fragile: small changes in the weights can induce drastically different policies. Multi-objective reinforcement learning (MORL) instead produces sets of policies that explicitly represent trade-offs between objectives. However, these policies are typically presented to the decision maker only through their value vectors, which can obscur… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/ffelten/Behavior-vs-Objective-Space · OpenReview ID：hfnMLNCCYz

期刊：Transactions on Machine Learning Research · 摘要：Multi-task learning (MTL) concurrently trains a model on diverse task datasets to exploit common features, thereby improving overall performance across the tasks. Recent studies have dedicated efforts to merging multiple independent model parameters into a unified model for MTL, thus circumventing the need for training data and expanding the scope of applicable scenarios of MTL. However, current approaches to model merging predominantly concentrate on enhancing performance within in-domain (ID) datasets, often overlooking their efficacy on out-of-domain (OOD) datasets. In this work, we propos… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：msiXRyQgxO

期刊：Transactions on Machine Learning Research · 摘要：Predictive systems may combine a predictor, a sequential monitor, a prediction set, and an online strategy, each governed by a different optimality criterion. We study when a guarantee for one component can be transferred to another. Four admissibility geometries shape sequential and distribution-free inference: Blackwell risk dominance over convex risk sets; anytime-valid admissibility within the $e$-process class; fixed-level marginal coverage with expected-length efficiency within a declared rank-indexed family under exchangeability; and choice-based approachability (CApp) boundary-feasibi… · 篇幅：Long submission (more than 12 pages of main content) · OpenReview ID：nZVEXyfvyM

期刊：Transactions on Machine Learning Research · 摘要：The integration of Large Language Model (LLMs) blocks with Vision Transformers (ViTs) holds significant promise for vision-only tasks by leveraging the rich semantic knowledge and reasoning capabilities of LLMs. However, a fundamental challenge lies in the inherent modality mismatch between the text-centric pre-training of LLMs and the vision-centric training of ViTs. Direct fusion often fails to fully exploit the LLM's potential and suffers from unstable finetuning. Consequently, prior works typically keep LLM blocks frozen while learning only the vision components. To address these challeng… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://www.github.com/selimkuzucu/LAVIE · OpenReview ID：s2T8Kgj6Rd

期刊：Transactions on Machine Learning Research · 摘要：Generative modeling for mixed-type tabular data has recently been dominated by diffusion-based methods, but their gains often come with schedule design, time dependent score parameterization, and multi-step solvers that increase computational overhead and tuning difficulty. We present \textbf{TabFlowM}, a lightweight framework that asks a more targeted question: once mixed-type records are mapped into a decoder compatible continuous transport space, is diffusion style score learning still necessary? TabFlowM answers this by training a single time conditioned velocity field via flow matching t… · 篇幅：Long submission (more than 12 pages of main content) · OpenReview ID：t5kygrpSIz

期刊：Transactions on Machine Learning Research · 摘要：Estimating individual treatment effects (ITEs) in the presence of unobserved confounding remains a central challenge in causal inference. Existing proxy-based methods aim to recover latent confounders from observational proxies, but typically produce only point estimates without uncertainty quantification. This lack of uncertainty modeling provides incomplete and potentially insufficient information for downstream decision-making, especially when uncertainty is inherent in the data. We propose IntervalGP-VAE, a novel framework that combines variational autoencoders with Interval Gaussian Proc… · 篇幅：Long submission (more than 12 pages of main content) · 代码：https://github.com/ZhigaoGuo/IntervalGP-VAE · OpenReview ID：vh7cshQCAE

期刊：Transactions on Machine Learning Research · 摘要：Graphs are foundational abstractions in data-intensive domains, yet the scale of modern datasets strains computation and memory for downstream learning. From recommender systems to biological networks, graphs have emerged as a fundamental substrate for learning. As graph sizes grow, the cost of training and inference becomes prohibitive, thereby necessitating compact surrogates that retain spectral properties and feature semantics. We propose an entropy-regularized, semi-supervised framework for attributed graph coarsening that jointly leverages the original graph’s Laplacian, node features,… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：xZAwzxUP7c

期刊：Transactions on Machine Learning Research · 摘要：Finding Nash equilibria in two-player zero-sum imperfect-information games remains a central challenge in multi-agent reinforcement learning. Recent multi-round regularization methods offer a promising direction, yet existing approaches either require full enumeration of the game tree or rely on non-policy-gradient inner solvers that underperform in practice, leaving a scalable policy-gradient-based solution open. In this paper, we propose a novel multi-round regularization procedure and show that it guarantees strictly monotonic reduction in Bregman divergence to Nash equilibria and eventual… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/ntu-agents/nashpg · OpenReview ID：yIA2Fjs1FK

期刊：Transactions on Machine Learning Research · 摘要：Vision-language research has largely developed along two trajectories: autoregressive architectures for multimodal understanding and diffusion-based frameworks for visual generation. Recent systems increasingly combine these capabilities, but their benefits and costs depend on how visual information is represented and how understanding interacts with generation. This survey organizes unified vision-language models by continuous versus discrete visual representations. For continuous representations, we distinguish serial coupling, in which a language model provides a structured condition to a… · 篇幅：Long submission (more than 12 pages of main content) · OpenReview ID：AIMmeOrVFL

期刊：Transactions on Machine Learning Research · 摘要：Membership Inference Attacks (MIAs) pose a critical privacy threat by enabling adversaries to determine whether a specific sample was included in a model's training dataset. Despite extensive research on MIAs, systematic comparisons between generative and discriminative classifiers remain limited. This work addresses this gap by first providing theoretical motivation for why generative classifiers exhibit heightened susceptibility to MIAs, then validating these insights through comprehensive empirical evaluation. Our study encompasses discriminative, generative, and pseudo-generative text cla… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/karan25gupta/privacy-attacks-gendisc-classifiers · OpenReview ID：SHMC01wdVM

期刊：Transactions on Machine Learning Research · 摘要：Knowledge tracing (KT) aims to estimate a student's evolving knowledge state and predict their performance on new exercises based on performance history. Many realistic classroom settings for KT are typically low-resource in data and require online updates as students' exercise history grows, which creates significant challenges for existing KT approaches. To restore strong performance under low-resource conditions, we revisit the hierarchical knowledge concept (KC) information, which is typically available in many classroom settings and can provide strong prior when data are sparse. We there… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/UCSB-NLP-Chang/KT2 · OpenReview ID：9kfruXm7e9

期刊：Transactions on Machine Learning Research · 摘要：Despite the importance for interpretability, decision trees face severe scalability challenges. Existing global optimal methods are often limited by binary feature selection and shallow tree depths, whereas traditional heuristic approaches frequently sacrifice predictive accuracy. To overcome these limitations, this paper proposes a moving-horizon approximate branch-and-reduce method to train near-optimal deep classification trees on large-scale datasets with continuous features. Built on a hierarchical root-subtree optimization framework, the method solves the root-level problem via branch-a… · 篇幅：Long submission (more than 12 pages of main content) · 代码：https://github.com/YankaiGroup/MHABR.jl · OpenReview ID：4Sq5Byd4yS

期刊：Transactions on Machine Learning Research · 摘要：The growing application of Large Language Models (LLMs) to social contexts has led to an increase in unjustifiable social group attributions through their own stereotype-based responses; especially when responding to questions where there is little evidence to support a response or ambiguity to context. The lack of sufficient evidence often leads models to hallucinate socially grounded inferences, undermining fairness and trust. In this work, we attempt to mitigate social bias under ambiguity via epistemic uncertainty. We further introduce BHARATBBQ-R, a rationale-augmented extension of BHARA… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：UT5E31pYob

期刊：Transactions on Machine Learning Research · 摘要：Prompt tuning, a leading parameter-efficient adaptation paradigm in NLP, has recently been extended to computer vision. Visual prompt tuning (VPT) adapts pre-trained vision transformers (ViTs) by updating a small set of additional prompt parameters. However, existing visual prompts are randomly initialized and do not exploit prior knowledge, such as instructions in NLP. We address this gap by injecting two complementary semantic priors into VPT. Fundamental image priors, including color, texture, and shape, are extracted with classical hand-crafted operators and injected into the input space,… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://xixiaouab.github.io/Cascaded-Semantics/ · OpenReview ID：SSsobNZJPO

期刊：Transactions on Machine Learning Research · 摘要：Differential privacy (DP) is the gold standard for privacy-preserving analysis but introduces a fundamental trade-off between privacy guarantees and model performance. Selecting the optimal balance is a critical challenge, framed as a multi-objective optimization (MOO) problem of discovering the Pareto front and eliciting a decision-maker's preference. While interactive MOO offers a solution, standard approaches---which model objectives separately and rely on simple pairwise feedback---are suboptimal for DP because they do not utilize the problem structure. In this work, we propose a method,… · 篇幅：Long submission (more than 12 pages of main content) · 代码：https://github.com/YangYaohong1/PACE · OpenReview ID：ire2TQNxfv

期刊：Transactions on Machine Learning Research · 摘要：Decentralized training of large language models offers the opportunity to leverage computational resources across geographically distributed participants, but is often bottlenecked by network communication, particularly in pipeline-parallel settings. While pipeline parallelism partitions model layers across devices to handle large-scale models, it requires frequent communication of intermediate activations, which can be challenging when network bandwidth is limited. To address these issues, we propose TAH-Quant (Tile-wise Adaptive Hadamard Quantization), a novel activation quantization framew… · 篇幅：Regular submission (no more than 12 pages of main content) · OpenReview ID：6ysPGq2RVD

期刊：Transactions on Machine Learning Research · 摘要：Approximate Machine Unlearning (MU) methods typically forget specific data by modifying model parameters or learning data-dependent augmentations. However, each unlearning request requires a dedicated optimization process, resulting in high computational overhead and cumulative degradation of model performance over time. Recent approaches have proposed manipulating a small subset of neural activations as a more targeted alternative, yet these methods still rely on on-demand searches for relevant parameters and remain computationally expensive. We propose Minimally Invasive Machine Unlearning… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/layer6ai-labs/MIMU · OpenReview ID：bVc04v8dV7

期刊：Transactions on Machine Learning Research · 摘要：Training stable biological foundation models requires rethinking attention mechanisms: we find that using sigmoid attention as a drop in replacement for softmax attention a) produces better learned representations: on six diverse single-cell datasets, sigmoid achieves 25% higher cell-type separation, better cell-type cohesion metrics, and lower validation loss, b) faster training, models with sigmoid attention train up to 10% faster than their softmax counterparts, and c) more stable training by eliminating inherent sources of instability in softmax attention. We establish that sigmoid attent… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/MSDLLCpapers/triton-sigmoid · OpenReview ID：GxmQsP5CcU

期刊：Transactions on Machine Learning Research · 摘要：Geospatial foundation-model (GeoFM) benchmarks for agricultural field-extent segmentation often rely on cropland land-cover proxies and compare against per-pixel spectral baselines. We measure how these choices affect conclusions using identical single-date Sentinel-2 pixels from six countries. Replacing polygon-derived field labels with ESA WorldCover cropland raises random-forest AUROC from 0.55–0.82 to 0.79–0.96, showing that the proxy creates a more spectrally separable task. On the actual field-extent labels, a from-scratch U-Net reaches 0.89–0.98 AUROC and has higher point AUROC than a… · 篇幅：Long submission (more than 12 pages of main content) · 代码：https://github.com/thesantoshpant/auditing-geofm-evaluation · OpenReview ID：qRXVTe1yYp

期刊：Transactions on Machine Learning Research · 摘要：Deep neural networks are highly susceptible to backdoor attacks, yet most defense methods to date rely on balanced data, overlooking the pervasive class imbalance in real-world scenarios that can amplify backdoor threats. This paper presents the first in-depth investigation of how the dataset imbalance amplifies backdoor vulnerability, showing that (i) the imbalance induces a majority-class bias that increases susceptibility and (ii) conventional defenses degrade significantly as the imbalance grows. To address this, we propose Randomized Probability Perturbation (RPP), a certified poisoned-s… · 篇幅：Regular submission (no more than 12 pages of main content) · 代码：https://github.com/Lin23508/RPP · OpenReview ID：WyPkotlgqQ

期刊：Transactions on Machine Learning Research · 摘要：Video Anomaly Detection (VAD) aims to temporally localize abnormal events in videos. Most existing approaches rely on dataset-specific training and curated annotations, limiting generalization in open-set scenarios. Recent zero-shot methods based on Large Vision-Language Models (LVLMs) alleviate this dependency but often lack temporal continuity and structured reasoning. We propose \textbf{Cog-VADU}, a fully training-free framework that reformulates VAD as a sequential cognitive reasoning task. Cog-VADU introduces \emph{Chain-of-Anomaly Detection Thought Prompting} (CoADTP), which unrolls an… · 篇幅：Long submission (more than 12 pages of main content) · 代码：https://github.com/MohdUbaidwani/Cog-VADU · OpenReview ID：QcuSMNG7J8