Machine Learning
ALIGN: Aligned Delegation with Performance Guarantees for Multi-Agent LLM Reasoning
Tong Zhu, Baiting Chen, Jin Zhou et al. (+3 more)
LLMs often underperform on complex
reasoning tasks when relying on a single generation-and-selection pipeline. Inference-time ensemble methods can improve performance by sampling diverse
reasoning paths or aggregating multiple candidate answers, but they typically treat candidates independently and provide no formal guarantees that ensembling improves
reasoning quality. We propose a novel method, Aligned Delegation for Multi-
Agent LLM
Reasoning (ALIGN), which formulates LLM
reasoning as an aligned delegation game. In ALIGN, a principal delegates a task to multiple
agents that generate candidate solutions under designed incentives, and then selects among their outputs to produce a final answer. This formulation induces structured interaction among
agents while preserving
alignment between
agent and principal objectives. We establish theoretical guarantees showing that, under a fair comparison with equal access to candidate solutions, ALIGN provably improves expected performance over single-
agent generation. Our analysis accommodates correlated candidate answers and relaxes independence assumptions that are commonly used in prior work. Empirical results across a broad range of LLM
reasoning benchmarks consistently demonstrate that ALIGN outperforms strong single-
agent and ensemble baselines.
Read the full abstract ▾
All authors Tong Zhu, Baiting Chen, Jin Zhou, Hua Zhou, Sriram Sankararaman, Xiaowu Dai
Machine Learning · Cross-listed from cs.GT
Tacit Coordination of Large Language Models
Ido Aharon, Emanuele La Malfa, Michael Wooldridge et al. (+1 more)
Large
Language Models (LLMs) are increasingly deployed in multi-
agent settings that require coordination without communication, from human-AI interaction to safety-critical scenarios. Humans often overcome the absence of communication through focal points: salient solutions that naturally stand out to all participants. We present the first large-scale evaluation of how, when, and why focal points emerge in LLMs, comparing their behaviour with humans across cooperative and competitive games, including realistic search and rescue scenarios, demonstrating when focal points enable effective coordination. Across more than 20 open- and closed-source models, we find that LLMs exhibit a remarkable ability to coordinate without communication, often matching or outperforming humans. However, the same models consistently fail in tasks requiring numerical common sense or culturally nuanced notions of salience. We additionally evaluate simple learning-free strategies that substantially improve coordination both among LLMs and between humans and LLMs. Our results reveal striking coordination capabilities, as well as social limitations in modern LLMs, and offer new insight into the latent notions of salience encoded within them. Our findings caution against assuming that LLMs share humans' cultural and perceptual substrate when deployed in coordination settings.
Read the full abstract ▾
All authors Ido Aharon, Emanuele La Malfa, Michael Wooldridge, Sarit Kraus
Machine Learning · Cross-listed from stat.ML
Multilevel and Sequential Monte Carlo for Training-Free Diffusion Guidance
Aidan Gleich and Scott C. Schmidler
We address the problem of accurate, training-free guidance for conditional generation in trained
diffusion models. Existing methods typically rely on point-estimates to approximate the posterior score, often resulting in biased approximations that fail to capture multimodality inherent to the reverse process of
diffusion models. We propose a sequential Monte Carlo (SMC) framework that integrates over the full denoising distribution to construct an unbiased Monte Carlo
estimator of $p_\theta(y\mid x_t)$, avoiding the bias of existing training-free guidance methods. To ensure computational tractability, we incorporate variance-reduction schemes based on MultiLevel Monte Carlo (MLMC). We evaluate on three image generation tasks. On CIFAR-10 and ImageNet classifier guidance, our approach improves accuracy by 18 and 38 points respectively over the strongest training-free baseline. On CelebA-HQ joint-attribute guidance, three existing methods (DPS, LGD, TFG) used within our framework all outperform the closest SMC-based comparator (TDS) by 25-37 points, and improve over the strongest baseline by 2-14 points.
Read the full abstract ▾
Machine Learning · Cross-listed from cs.AI
Magellan: Autonomous Discovery of Novel Compiler Optimization Heuristics with AlphaEvolve
Hongzheng Chen, Alexander Novikov, Ngân Vũ et al. (+5 more)
Modern compilers rely on hand-crafted heuristics to guide optimization passes. These human-designed rules often struggle to adapt to the complexity of modern software and hardware and lead to high maintenance burden. To address this challenge, we present Magellan, an agentic framework that evolves the compiler pass itself by synthesizing executable C++ decision logic. Magellan couples an LLM coding
agent with evolutionary search and autotuning in a closed loop of generation, evaluation on user-provided macro-
benchmarks, and refinement, producing compact heuristics that integrate directly into existing compilers. Across several production optimization tasks, Magellan discovers policies that match or surpass expert baselines. In LLVM function inlining, Magellan synthesizes new heuristics that outperform decades of manual engineering for both binary-size reduction and end-to-end performance. In register allocation, it learns a concise priority rule for live-range processing that matches intricate human-designed policies on a large-scale workload. We also report preliminary results on XLA problems, demonstrating portability beyond LLVM with reduced engineering effort.
Read the full abstract ▾
All authors Hongzheng Chen, Alexander Novikov, Ngân Vũ, Hanna Alam, Zhiru Zhang, Aiden Grossman, Mircea Trofin, Amir Yazdanbakhsh
Machine Learning
Safety Generalization Under Distribution Shift in Safe Reinforcement Learning: A Diabetes Testbed
Minjae Kwon, Josephine Lamp and Lu Feng
Safe Reinforcement Learning (RL)
algorithms are typically evaluated under fixed training conditions. We investigate whether training-time safety guarantees transfer to deployment under distribution shift, using diabetes management as a safety-critical testbed. We
benchmark safe RL
algorithms on a unified clinical simulator and reveal a safety generalization gap: policies satisfying constraints during training frequently violate safety requirements on unseen patients. We demonstrate that test-time shielding, which filters unsafe actions using learned dynamics models, effectively restores safety across
algorithms and patient populations. Across eight safe RL
algorithms, three diabetes types, and three age groups, shielding achieves Time-in-Range gains of 13–14% for strong baselines such as PPO-Lag and CPO while reducing clinical risk index and glucose variability. Our simulator and
benchmark provide a platform for studying safety under distribution shift in safety-critical control domains. Code is available at https://github.com/safe-autonomy-lab/GlucoSim and https://github.com/safe-autonomy-lab/GlucoAlg.
Read the full abstract ▾
Machine Learning · Cross-listed from stat.ML
High-dimensional learning dynamics of multi-pass Stochastic Gradient Descent in multi-index models
Zhou Fan and Leda Wang
The learning dynamics of
Stochastic Gradient Descent (SGD) are believed to play an important role in influencing the generalization properties of trained models. While the precise dynamics of how online (one-pass) SGD optimizes the population risk are now understood in a broad range of high-dimensional problems, a comparable understanding of multi-pass SGD for empirical risk minimization is still largely lacking. In this work, we develop an asymptotically exact characterization of the dynamics of multi-pass SGD for a non-convex multi-index model, in the high-dimensional regime where sample size $n$ and data dimension $d$ increase proportionally. We consider any sub-linear batch size $\kappa \asymp n^\alpha$ with $\alpha \in [0,1)$ and a commensurate “critical” scaling of the learning rate, and show that the coordinate-wise dynamics of SGD admit a novel continuous-time scaling limit, given by a system of dynamical mean-field equations that are driven by a scalar Poisson jump process representing the asymptotic limit of SGD sampling noise. This scaling limit connects several previously disparate theories around SGD, namely (i) the classical interpretation of SGD as an approximation to gradient flow on the empirical risk, (ii) the characterization of online SGD via an autonomous ODE in its overlap statistics, and (iii) the coincidence between multi-pass SGD dynamics and
diffusion approximations in least-squares problems. We illustrate how these perspectives are recovered by specializing our result to small learning rates, large sample sizes, and linear models, respectively. Outside of these specializations, the scaling limit of SGD is distinct both from
diffusion approximations and gradient flow, and may enable SGD to converge to local optima of the non-convex empirical risk with different statistical properties.
Read the full abstract ▾
Machine Learning
MapPFN: Learning Causal Perturbation Maps in Context
Marvin Sextro, Weronika Kłos and Gabriel Dernbach
Planning effective interventions in biological systems requires treatment-effect models that adapt to unseen biological contexts by identifying their specific underlying mechanisms. Yet single-
cell perturbation
datasets span only a handful of biological contexts, and existing methods cannot leverage new interventional evidence at inference time to adapt beyond their training data. To meta-learn a perturbation effect
estimator, we present MapPFN, a prior-data fitted network (PFN) pre-trained on a synthetic biological prior with causal interventions, decoupling pre-training from limited wet-lab data. Unlike existing methods, MapPFN uses in-context learning to map a sequence of experiments to a post-perturbation distribution, enabling a single pre-trained model to adapt to new
datasets and gene sets. Our results show that combining synthetic pre-training with fine-tuning on real single-
cell data improves perturbation predictions across biological contexts. The code, model and data are available at https://marvinsxtr.github.io/MapPFN.
Read the full abstract ▾
Machine Learning · Cross-listed from stat.ML
An efficient, accurate, and interpretable machine learning method for computing probability of failure
Jacob Zhu and Donald Estep
We introduce a novel machine learning method called the Penalized Profile Support Vector Machine based on the Gabriel edited set for the computation of the probability of failure for a complex system as determined by a threshold condition on a computer model of system behavior. The method is designed to minimize the number of evaluations of the computer model while preserving the geometry of the decision boundary that determines the probability. It employs an adaptive sampling strategy designed to strategically allocate points near the boundary determining failure and builds a locally linear surrogate boundary that remains consistent with its geometry by strategic clustering of training points. We prove two convergence results and we compare the performance of the method against a number of state of the art classification methods on four test problems. We also apply the method to determine the probability of survival using the Lotka–Volterra model for competing species.
Read the full abstract ▾
Machine Learning
LOCUS: Low-Dimensional Model Embeddings for Efficient Model Exploration, Comparison, and Selection
Shivam Patel, William Cocke and Gauri Joshi
The rapidly growing ecosystem of Large
Language Models (LLMs) makes it increasingly challenging to manage and utilize the vast and dynamic pool of models effectively. We propose LOCUS, a method that produces low-dimensional vector
embeddings that compactly represent a
language model's capabilities across queries. LOCUS is an
attention-based approach that generates
embeddings by a deterministic forward pass over query encodings and evaluation scores via an encoder model, enabling seamless incorporation of new models to the pool and refinement of existing model
embeddings without having to perform any retraining. We additionally train a correctness predictor that uses model
embeddings and query encodings to achieve state-of-the-art routing accuracy on unseen queries. Experiments show that LOCUS needs up to 4.8x fewer query evaluation samples than baselines to produce informative and robust
embeddings. Moreover, the learned
embedding space is geometrically meaningful: proximity reflects model similarity, enabling a range of downstream applications including model comparison and clustering, model portfolio selection, and resilient proxies of unavailable models.
Read the full abstract ▾
Machine Learning · Cross-listed from math.NA
Parametric Hyperbolic Conservation Laws: A Unified Framework for Conservation, Entropy Stability, and Hyperbolicity
Lizuo Liu, Lu Zhang and Anne Gelb
We propose a parametric
hyperbolic conservation law (SymCLaw) for learning
hyperbolic systems directly from data while ensuring conservation,
entropy stability, and hyperbolicity by design. Unlike existing approaches that typically enforce only conservation or rely on prior knowledge of the governing equations, our method parameterizes the flux functions in a form that guarantees real
eigenvalues and complete eigenvectors of the flux Jacobian, thereby preserving hyperbolicity. At the same time, we embed
entropy-stable design principles by jointly learning a convex
entropy function and its associated flux potential, ensuring
entropy dissipation and the selection of physically admissible weak solutions. A corresponding
entropy-stable numerical flux scheme provides compatibility with standard discretizations, allowing seamless integration into classical solvers. Numerical experiments on
benchmark problems, including Burgers, shallow water, Euler, and KPP equations, demonstrate that SymCLaw generalizes to unseen initial conditions, maintains stability under noisy training data, and achieves accurate long-time predictions, highlighting its potential as a principled foundation for data-driven modeling of
hyperbolic conservation laws.
Read the full abstract ▾
Machine Learning
Out-of-Distribution Generalization in Graph Foundation Models
Haoyang Li, Haibo Chen, Xin Wang et al. (+1 more)
Graphs are a fundamental data structure for representing relational information in domains such as social networks, molecular systems, and knowledge graphs. However, graph learning models often suffer from limited generalization when applied beyond their training distributions. In practice, distribution shifts may arise from changes in graph structure, domain semantics, available modalities, or task formulations. To address these challenges, graph foundation models (GFMs) have recently emerged, aiming to learn general-purpose representations through large-scale pretraining across diverse graphs and tasks. In this survey, we review recent progress on GFMs from the perspective of out-of-distribution (OOD) generalization. We first discuss the main challenges posed by distribution shifts in graph learning and outline a unified problem setting. We then organize existing approaches based on whether they are designed to operate under a fixed task specification or to support generalization across heterogeneous task formulations, and summarize the corresponding OOD handling strategies and pretraining objectives. Finally, we review common evaluation protocols and discuss open directions for future research. To the best of our knowledge, this paper is the first survey for OOD generalization in GFMs.
Read the full abstract ▾
All authors Haoyang Li, Haibo Chen, Xin Wang, Wenwu Zhu
Machine Learning
Textual Equilibrium Propagation for Deep Compound AI Systems
Minghui Chen, Wenlong Deng, James Zou et al. (+2 more)
Large
language models (LLMs) are increasingly deployed as part of compound AI systems that coordinate multiple modules (e.g., retrievers, tools, verifiers) over long-horizon workflows. Recent approaches that propagate textual feedback globally (e.g., TextGrad) make it feasible to optimize such pipelines, but we find that performance degrades as system depth grows. In particular, long-horizon agentic workflows exhibit two depth-scaling failure modes: 1) exploding textual gradient, where textual feedback grows exponentially with depth, leading to prohibitively long message and amplifies evaluation biases; and 2) vanishing textual gradient, where limited long-context ability causes models overemphasize partial feedback and compression of lengthy feedback causes downstream messages to lose specificity gradually as they propagate many hops upstream. To mitigate these issues, we introduce Textual
Equilibrium Propagation (TEP), a local learning principle inspired by
Equilibrium Propagation in energy-based models. TEP includes two phases: 1) a free phase where a local LLM critics iteratively refine
prompts until reaching
equilibrium (no further improvements are suggested); and 2) a nudged phase which applies proximal
prompt edits with bounded modification intensity, using task-level objectives that propagate via forward signaling rather than backward feedback chains. This design supports local
prompt optimization followed by controlled adaptation toward global goals without the computational burden and signal degradation of global textual backpropagation. Across long-horizon QA
benchmarks and multi-
agent tool-use
dataset, TEP consistently improves accuracy and efficiency over global propagation methods such as TextGrad. The gains grows with depth, while preserving the practicality of black-box LLM components in deep compound AI system.
Read the full abstract ▾
All authors Minghui Chen, Wenlong Deng, James Zou, Han Yu, Xiaoxiao Li
Machine Learning
Signal from Structure: Exploiting Submodular Upper Bounds in Generative Flow Networks
Alexandre Larouche and Audrey Durand
Generative Flow Networks (GFlowNets; GFNs) are a class of generative models that learn to sample compositional objects proportional to their a priori unknown value, their reward. We focus on the case where the reward has a specified, actionable structure, namely that it is submodular. We show submodularity can be harnessed to retrieve upper bounds on the reward of compositional objects that have not yet been observed. We provide in-depth analyses of the probability of such bounds occurring, as well as how many unobserved compositional objects can be covered by a bound. Following the Optimism in the Face of Uncertainty principle, we then introduce SUBo-GFN , which uses the submodular upper bounds to train a GFN. We show that SUBo-GFN generates orders of magnitude more learning signal than classical GFNs for the same number of queries to the reward function. We demonstrate the effectiveness of SUBo-GFN in terms of distribution matching and high-quality candidate generation on graph submodular tasks.
Read the full abstract ▾
Machine Learning
Human-LLM Collaborative Feature Engineering for Tabular Data
Zhuoyan Li, Aditya Bansal, Jinzhao Li et al. (+8 more)
Large
language models (LLMs) are increasingly used to automate feature engineering in tabular learning. Given task-specific information, LLMs can propose diverse feature transformation operations to enhance downstream model performance. However, current approaches typically assign the LLM as a black-box optimizer, responsible for both proposing and selecting operations based solely on its internal heuristics, which often lack calibrated estimations of operation utility and consequently lead to repeated exploration of low-yield operations without a principled strategy for prioritizing promising directions. In this paper, we propose a human-LLM collaborative feature engineering framework for tabular learning. We begin by decoupling the transformation operation proposal and selection processes, where LLMs are used solely to generate operation candidates, while the selection is guided by explicitly modeling the utility and uncertainty of each proposed operation. Since accurate utility estimation can be difficult especially in the early rounds of feature engineering, we design a mechanism within the framework that selectively elicits and incorporates human expert preference feedback, comparing which operations are more promising, into the selection process to help identify more effective operations. Our evaluations on both the synthetic study and the real user study demonstrate that the proposed framework improves feature engineering performance across a variety of tabular
datasets and reduces users' cognitive load during the feature engineering process.
Read the full abstract ▾
All authors Zhuoyan Li, Aditya Bansal, Jinzhao Li, Shishuang He, Zhuoran Lu, Mutian Zhang, Qin Liu, Yiwei Yang, Swati Jain, Ming Yin, Yunyao Li
Machine Learning
Snowball: A Scalable All-to-All Ising Machine with Dual-Mode Markov Chain Monte Carlo Spin Selection and Asynchronous Spin Updates for Fast Combinatorial Optimization
Seungki Hong, Kyeongwon Jeong and Taekwang Jang
Ising machines have emerged as accelerators for combinatorial optimization. To enable practical deployment, this work aims to reduce time-to-solution by addressing three challenges: (1) hardware
topology, (2)
spin selection and update
algorithms, and (3) scalable coupling-coefficient precision. Restricted
topologies require minor
embedding; naive parallel updates can oscillate or stall; and limited precision can preclude feasible mappings or degrade solution quality. This work presents Snowball, a digital, scalable, all-to-all coupled Ising machine that integrates dual-mode Markov chain Monte Carlo
spin selection with asynchronous
spin updates to promote convergence and reduce time-to-solution. The digital architecture supports wide, configurable coupling precision, unlike many analog realizations at high bit widths. A prototype on an AMD Alveo U250 accelerator card achieves an 8$\times$ reduction in time-to-solution relative to a state-of-the-art Ising machine on the same
benchmark instance.
Read the full abstract ▾
Machine Learning · Cross-listed from physics.chem-ph
Better without U: Impact of Selective Hubbard U Correction on Foundational MLIPs
Thomas Warford, Fabian L. Thiemann and Gábor Csányi
The training of foundational machine learning interatomic potentials (fMLIPs) relies on diverse databases with energies and forces calculated using ab initio methods. We show that fMLIPs trained on large
datasets such as MPtrj, Alexandria, and OMat24 encode inconsistencies from the Materials Project's selective use of the Hubbard U correction, which is applied to certain transition metals only if O or F atoms are present in the simulation
cell. This inconsistent use of +U creates two incompatible potential-energy surfaces (PES): a lower-energy GGA surface and a higher-energy GGA+U one. When trained on both, MLIPs interpolate between them, leading to systematic underbinding, or even spurious repulsion, between U-corrected metals and oxygen- or fluorine-containing species. Models such as MACE-OMAT and -MPA exhibit repulsion between U-corrected metals and their
oxides, limiting their value for studying catalysis and oxidation. We link the severity of this pathology to the oxygen number density in U-corrected training configurations. This explains why OMAT-trained models are most affected and suggests the issue might worsen as expanding future
datasets increasingly include configurations with low oxygen content, such as those generated through combinatorial exploration of multi-element or defect-containing systems. Our simple per-U-corrected-atom shift aligns PBE+U and PBE energies for identical structures, yielding a smoother PES compared to existing correction schemes, which target phase diagram accuracy. As a result, models trained on
datasets with our shift applied exhibit smaller mean absolute errors for the adsorption energies of oxygen on U-corrected elemental slabs. Since
datasets omitting +U entirely (e.g. MatPES, MP-ALOE) avoid these pathologies, we recommend excluding +U in future fMLIP
datasets. For existing
datasets, our post-hoc correction provides a low-cost improvement.
Read the full abstract ▾
Machine Learning · Cross-listed from cs.AI
Llama-3.1-FoundationAI-SecurityLLM-Reasoning-8B Technical Report
Zhuoran Yang, Ed Li, Jianliang He et al. (+18 more)
We present Foundation-Sec-8B-
Reasoning, the first open-source native
reasoning model for cybersecurity. Built upon our previously released Foundation-Sec-8B base model (derived from Llama-3.1-8B-Base), the model is trained through a two-stage process combining supervised fine-tuning (SFT) and reinforcement learning from verifiable rewards (RLVR). Our training leverages proprietary
reasoning data spanning cybersecurity analysis, instruction-following, and mathematical
reasoning. Evaluation across 10 cybersecurity
benchmarks and 10 general-purpose
benchmarks demonstrates performance competitive with significantly larger models on cybersecurity tasks while maintaining strong general capabilities. The model shows effective generalization on multi-hop
reasoning tasks and strong safety performance when deployed with appropriate system
prompts and guardrails. This work demonstrates that domain-specialized
reasoning models can achieve strong performance on specialized tasks while maintaining broad general capabilities. We release the model publicly at https://huggingface.co/fdtn-ai/Foundation-Sec-8B-
Reasoning.
Read the full abstract ▾
All authors Zhuoran Yang, Ed Li, Jianliang He, Aman Priyanshu, Baturay Saglam, Paul Kassianik, Sajana Weerawardhena, Anu Vellore, Blaine Nelson, Neusha Javidnia, Arthur Goldblatt, Fraser Burch, Avi Zohary, Assaf Eisenman, Mahdi Sabbaghi, Supriti Vijay, Rahim Dharssi, Dhruv Kedia, Kojin Oshiba, Yaron Singer, Amin Karbasi
Machine Learning
SMKC: Sketch Based Kernel Correlation Images for Variable Cardinality Time Series Anomaly Detection
Haokun Zhou
Conventional anomaly detection in multivariate time series relies on the assumption that the set of observed variables remains static. In operational environments, however, monitoring systems frequently experience sensor churn. Signals may appear, disappear, or be renamed, creating data windows where the cardinality varies and may include values unseen during training. To address this challenge, we propose SMKC, a framework that decouples the dynamic input structure from the anomaly detector. We first employ permutation-invariant feature hashing to sketch raw inputs into a fixed size state sequence. We then construct a hybrid kernel image to capture global temporal structure through pairwise comparisons of the sequence and its derivatives. The model learns normal patterns using masked reconstruction and a teacher-student prediction objective. Our evaluation reveals that robust log-distance channels provide the primary discriminative signal, whereas cosine representations often fail to capture sufficient contrast. Notably, we find that a detector using random projections and nearest neighbors on the SMKC representation performs competitively with fully trained baselines without requiring gradient updates. This highlights the effectiveness of the representation itself and offers a practical cold-start solution for resource-constrained deployments.
Read the full abstract ▾
Machine Learning
Test-Time Adaptation for Unsupervised Combinatorial Optimization
Yiqiao Liao, Farinaz Koushanfar and Parinaz Naghizadeh
Unsupervised neural combinatorial optimization (NCO) enables learning powerful solvers without access to ground-truth solutions. Existing approaches fall into two disjoint paradigms: models trained for generalization across instances, and instance-specific models optimized independently at test time. While the former are efficient during inference, they lack effective instance-wise adaptability; the latter are flexible but fail to exploit learned inductive structure and are prone to poor local optima. This motivates the central question of our work: how can we leverage the inductive bias learned through generalization while unlocking the flexibility required for effective instance-wise adaptation? We first identify a challenge in bridging these two paradigms: generalization-focused models often constitute poor warm starts for instance-wise optimization, potentially underperforming even randomly initialized models when fine-tuned at test time. To resolve this incompatibility, we propose TACO, a model-agnostic test-time adaptation framework that unifies and extends the two existing paradigms for unsupervised NCO. TACO applies strategic warm-starting to partially relax trained parameters while preserving inductive bias, enabling rapid and effective unsupervised adaptation. Crucially, compared to naively fine-tuning a trained generalizable model or optimizing an instance-specific model from scratch, TACO achieves better solution quality while incurring negligible additional computational cost. Experiments on the canonical problems of minimum vertex cover, maximum clique, maximum independent set, and max cut demonstrate the effectiveness and robustness of TACO across static, distribution-shifted, and dynamic combinatorial optimization problems, establishing it as a practical bridge between generalizable and instance-specific unsupervised NCO.
Read the full abstract ▾
Machine Learning
Thinking in Frames: How Visual Context and Test-Time Scaling Empower Video Reasoning
Chengzu Li, Zanyi Wang, Jiaang Li et al. (+9 more)
Vision-
Language Models have excelled at textual
reasoning, but they often struggle with fine-grained spatial understanding and continuous action planning, failing to simulate the dynamics required for complex visual
reasoning. In this work, we formulate visual
reasoning by means of
video generation models, positing that generated frames can act as intermediate
reasoning steps between initial states and solutions. We evaluate their capacity in two distinct regimes: Maze
Navigation for sequential discrete planning with low visual change and Tangram Puzzle for continuous manipulation with high visual change. Our experiments reveal three critical insights: (1) Robust Zero-Shot Generalization: In both tasks, the model demonstrates strong performance on unseen data distributions without specific finetuning. (2) Visual Context: The model effectively uses visual context as explicit control, such as
agent icons and tangram shapes, enabling it to maintain high visual consistency and adapt its planning capability robustly to unseen patterns. (3) Visual Test-Time Scaling: We observe a test-time scaling law in sequential planning; increasing the generated
video length (visual inference budget) empowers better zero-shot generalization to spatially and temporally complex paths. These findings suggest that
video generation is not merely a media tool, but a scalable, generalizable paradigm for visual
reasoning.
Read the full abstract ▾
All authors Chengzu Li, Zanyi Wang, Jiaang Li, Yi Xu, Han Zhou, Huanyu Zhang, Ruichuan An, Dengyang Jiang, Zhaochong An, Ivan Vulić, Serge Belongie, Anna Korhonen
Machine Learning
Predict-Project-Renoise: Sampling Diffusion Models under Hard Constraints
Omer Rochman-Sharabi and Gilles Louppe
Diffusion models cannot enforce hard constraints, yet applications in the physical sciences demand exact satisfaction of conservation laws, boundary conditions, and observational consistency. In this work, we identify a corrector kernel whose unique stationary distribution is the constrained marginal at each noise level, and approximate it by iteratively projecting through the denoiser and renoising via the forward kernel. The resulting Predict-Project-Renoise (PPR)
algorithm enables sampling from pretrained
diffusion models under hard constraints. Its three components are each necessary: projecting through the denoiser keeps samples close to the data
manifold, while renoising and iterating drive samples toward the constrained marginal. On 2D distributions, the Kuramoto-Sivashinsky equation, and global weather
forecasting with a $10^8$-dimensional atmospheric model, PPR simultaneously achieves low constraint violations and high distributional fidelity, a combination that existing methods fail to deliver.
Read the full abstract ▾
Machine Learning
SIGMA-PPG: Statistical-prior Informed Generative Masking Architecture for PPG Foundation Model
Zongheng Guo, Tao Chen, Yang Jiao et al. (+3 more)
Current foundation model for photoplethysmography (PPG) signals is challenged by the intrinsic redundancy and noise of the signal. Standard masked modeling often yields trivial solutions while contrastive methods lack morphological precision. To address these limitations, we propose a Statistical-prior Informed Generative Masking Architecture (SIGMA-PPG), a generative foundation model featuring a Prior-Guided
Adversarial Masking mechanism, where a reinforcement learning-driven teacher leverages statistical priors to create challenging learning paths that prevent overfitting to noise. We also incorporate a semantic consistency constraint via vector quantization to ensure that physiologically identical waveforms (even those altered by recording artifacts or minor perturbations) map to shared indices. This enhances codebook semantic density and eliminates redundant feature structures. Pre-trained on over 120,000 hours of data, SIGMA-PPG achieves superior average performance compared to five state-of-the-art baselines across 12 diverse downstream tasks. The code is available at https://github.com/ZonghengGuo/SigmaPPG.
Read the full abstract ▾
All authors Zongheng Guo, Tao Chen, Yang Jiao, Yi Pan, Xiao Hu, Manuela Ferrario
Machine Learning · Cross-listed from stat.ML
Diffusion-based Annealed Boltzmann Generators : benefits, pitfalls and hopes
Louis Grenioux and Maxence Noble
Sampling configurations at thermodynamic
equilibrium is a central challenge in statistical physics. Boltzmann Generators (BGs) tackle it by combining a generative model with a Monte Carlo (MC) correction step to obtain asymptotically unbiased samples from an unnormalized target. Most current BGs use classic MC mechanisms such as importance sampling, which both require tractable likelihoods from the backbone model and scale poorly in high-dimensional, multi-modal targets. We study BGs built on annealed Monte Carlo (aMC), which is designed to overcome these limitations by bridging a simple reference to the target through a sequence of intermediate densities.
Diffusion models (DMs) are powerful generative models and have already been incorporated into aMC-based recalibration schemes via the
diffusion-induced density path, making them appealing backbones for aMC-BGs. We provide an empirical meta-analysis of DM-based aMC-BGs on controlled multi-modal Gaussian mixtures (varying mode separation, number of modes, and dimension), explicitly disentangling inference effects from learning effects by comparing (i) a perfectly learned DM and (ii) a DM trained from data. Even with a perfect DM, standard integrations using only first-order
stochastic denoising kernels fail systematically, whereas second-order denoising kernels can substantially improve performance when covariance information is available. We further propose a deterministic aMC integration based on first-order transport maps derived from DMs, which outperforms the
stochastic first-order variant at higher computational cost. Finally, in the learned-DM setting, all DM-aMC variants struggle to produce accurate BGs; we trace the main bottleneck to inaccurate DM log-density estimation. Code available at https://github.com/h2o64/dabg.
Read the full abstract ▾
Machine Learning · Cross-listed from stat.ML
A Diffusive Classification Loss for Learning Energy-based Generative Models
RuiKang OuYang, Louis Grenioux and José Miguel Hernández-Lobato
Score-based generative models have recently achieved remarkable success. While they are usually parameterized by the score, an alternative way is to use a series of time-dependent energy-based models (EBMs), where the score is obtained from the negative input-gradient of the energy. Crucially, EBMs can be leveraged not only for generation, but also for tasks such as compositional sampling or building Boltzmann Generators via Monte Carlo methods. However, training EBMs remains challenging. Direct maximum likelihood is computationally prohibitive due to the need for nested sampling, while score matching, though efficient, suffers from mode blindness. To address these issues, we introduce the Diffusive Classification (DiffCLF) objective, a simple method that avoids blindness while remaining computationally efficient. DiffCLF reframes EBM learning as a supervised classification problem across noise levels, and can be seamlessly combined with standard score-based objectives. We validate the effectiveness of DiffCLF by comparing the estimated energies against ground truth in analytical Gaussian mixture cases, and by applying the trained models to tasks such as model composition and Boltzmann Generator sampling. Our results show that DiffCLF enables EBMs with higher fidelity and broader applicability than existing approaches.
Read the full abstract ▾
Machine Learning
Conditional Denoising Model as a Physical Surrogate Model
José Afonso, Pedro Viegas, Rodrigo Ventura et al. (+1 more)
Surrogate modeling for complex physical systems typically faces a trade-off between data-fitting accuracy and physical consistency. Physics-consistent approaches typically treat physical laws as soft constraints within the loss function, a strategy that frequently fails to guarantee strict adherence to the governing equations, or rely on post-processing corrections that do not intrinsically learn the underlying solution geometry. To address these limitations, we introduce the {Conditional Denoising Model (CDM)}, a generative model designed to learn the geometry of the physical
manifold itself. By training the network to restore clean states from noisy ones, the model learns a vector field that points continuously towards the valid solution subspace. We introduce a time-independent formulation that transforms inference into a deterministic fixed-point iteration, effectively projecting noisy approximations onto the
equilibrium manifold. Validated on a low-temperature
plasma physics and chemistry
benchmark, the CDM achieves higher parameter and data efficiency than physics-consistent baselines. Crucially, we demonstrate that the denoising objective acts as a powerful implicit regularizer: despite never seeing the governing equations during training, the model adheres to physical constraints more strictly than baselines trained with explicit physics losses.
Read the full abstract ▾
All authors José Afonso, Pedro Viegas, Rodrigo Ventura, Vasco Guerra
Machine Learning · Cross-listed from math.ST
Towards regularized learning from functional data with covariate shift
Markus Holzleitner, Sergiy Pereverzyev Jr., Sergei V. Pereverzyev et al. (+2 more)
This paper investigates a general regularization framework for unsupervised domain adaptation in vector-valued regression under the
covariate shift assumption, utilizing vector-valued reproducing kernel
Hilbert spaces (vRKHS).
Covariate shift occurs when the input distributions of the training and test data differ, introducing significant challenges for reliable learning. By restricting the hypothesis space, we develop a practical operator learning
algorithm capable of handling functional outputs. We establish optimal convergence rates for the proposed framework under a general source condition, providing a theoretical foundation for regularized learning in this setting. We also propose an aggregation-based approach that forms a linear combination of
estimators corresponding to different regularization parameters and different kernels. The proposed approach addresses the challenge of selecting appropriate tuning parameters, which is crucial for constructing a good
estimator, and we provide a theoretical justification for its effectiveness. Furthermore, we illustrate the proposed method on a real-world face image
dataset, demonstrating robustness and effectiveness in mitigating distributional discrepancies under
covariate shift.
Read the full abstract ▾
All authors Markus Holzleitner, Sergiy Pereverzyev Jr., Sergei V. Pereverzyev, Vaibhav Silmana, S. Sivananthan
Machine Learning · Cross-listed from stat.ML
Efficient Causal Structure Learning via Modular Subgraph Integration
Haixiang Sun, Pengchao Tian, Zihan Zhou et al. (+3 more)
Learning causal structures from observational data remains a fundamental yet computationally intensive task, particularly in high-dimensional settings where existing methods face challenges such as the super-exponential growth of the search space and increasing computational demands. To address this, we introduce VISTA (Voting-based Integration of Subgraph
Topologies for Acyclicity), a modular framework that decomposes the global causal structure learning problem into local subgraphs based on Markov Blankets. The global integration is achieved through a weighted voting mechanism that penalizes low-support edges via exponential decay, filters unreliable ones with an adaptive threshold, and ensures acyclicity using a Feedback Arc Set (FAS)
algorithm. The framework is model-agnostic, imposing no assumptions on the inductive biases of base learners, is compatible with arbitrary data settings without requiring specific structural forms, and fully supports parallelization. We also theoretically establish finite-sample error bounds for VISTA, and prove its asymptotic consistency under mild conditions. Extensive experiments on both synthetic and real
datasets consistently demonstrate the effectiveness of VISTA, yielding notable improvements in both accuracy and efficiency over a wide range of base learners.
Read the full abstract ▾
All authors Haixiang Sun, Pengchao Tian, Zihan Zhou, Jielei Zhang, Peiyi Li, Andrew L. Liu
Machine Learning
Order-Aware Test-Time Adaptation: Leveraging Temporal Dynamics for Robust Streaming Inference
Young Kyung Kim, Oded Schlesinger, Qiangqiang Wu et al. (+2 more)
Test-Time Adaptation (TTA) enables pre-trained models to adjust to distribution shift by learning from unlabeled test-time streams. However, existing methods typically treat these streams as independent samples, overlooking the supervisory signal inherent in temporal dynamics. To address this, we introduce Order-Aware Test-Time Adaptation (OATTA). We formulate test-time adaptation as a gradient-free recursive
Bayesian estimation task, using a learned dynamic transition
matrix as a temporal prior to refine the base model's predictions. To ensure safety in weakly structured streams, we introduce a likelihood-ratio gate (LLR) that reverts to the base predictor when temporal evidence is absent. OATTA is a lightweight, model-agnostic module that incurs negligible computational overhead. Extensive experiments across image classification, wearable and physiological signal analysis, and
language sentiment analysis demonstrate its universality; OATTA consistently boosts established baselines, improving accuracy by up to 6.35%. Our findings establish that modeling temporal dynamics provides a critical, orthogonal signal beyond standard order-agnostic TTA approaches.
Read the full abstract ▾
All authors Young Kyung Kim, Oded Schlesinger, Qiangqiang Wu, J. Matías Di Martino, Guillermo Sapiro
Machine Learning
ORLoopBench: Solver-in-the-Loop Benchmarks for Self-Correction and Behavioral Rationality in Operations Research
Ruicheng Ao, David Simchi-Levi and Xinshang Wang
Operations Research practitioners debug infeasible models through an iterative process: inspecting Irreducible Infeasible Subsystems ( IIS), identifying constraint conflicts, and repairing formulations until feasibility is restored. Existing LLM
benchmarks mostly treat OR as one-shot translation from problem descriptions to solver code, omitting this diagnostic loop. We formalize infeasible-model repair as a solver-in-the-loop Markov Decision Process in which each action triggers solver re-execution and IIS recomputation, yielding deterministic, verifiable feedback. We introduce ORLoopBench, a
benchmark suite with two components: OR-Debug-Bench releases 5,362 LP/MILP repair instances, while OR-Bias-Bench evaluates closed-form operational decision rationality across inventory settings. Solver-verified RLVR training enables an 8B model to surpass frontier APIs on LP repair (95.3% vs 92.4% RR @5), improves diagnostic behavior, and transfers to MILP repair. The same evaluation exposes semantic drift in whole-model code regeneration: feasible regenerated MILPs can solve the wrong problem. Process-level evaluation with solver oracles enables targeted training for reliable OR self-correction.
Read the full abstract ▾
Machine Learning
MADE: Benchmark Environments for Closed-Loop Materials Discovery
Shreshth A Malik, Tiarnan Doherty, Panagiotis Tigas et al. (+4 more)
Existing
benchmarks for computational materials discovery primarily evaluate static predictive tasks or isolated computational sub-tasks. While valuable, these evaluations neglect the inherently iterative and adaptive nature of scientific discovery. We introduce MAterials Discovery Environments (MADE), a novel framework for benchmarking end-to-end autonomous materials discovery pipelines. MADE simulates closed-loop discovery campaigns in which an
agent or
algorithm proposes, evaluates, and refines candidate materials under a constrained oracle budget, capturing the sequential and resource-limited nature of real discovery workflows. We formalize discovery as a search for thermodynamically stable compounds relative to a given convex hull, and evaluate efficacy and efficiency via comparison to baseline
algorithms. The framework is flexible; users can compose discovery
agents from interchangeable components such as generative models, filters, and planners, enabling the study of arbitrary workflows ranging from fixed pipelines to fully agentic systems with tool use and adaptive decision making. We demonstrate this by conducting systematic experiments across a family of systems, enabling ablation of components in discovery pipelines, and comparison of how methods scale with system complexity.
Read the full abstract ▾
All authors Shreshth A Malik, Tiarnan Doherty, Panagiotis Tigas, Muhammed Razzak, Stephen J. Roberts, Aron Walsh, Yarin Gal