Peer Revenue Returns subject to peer review

Papers

Fresh from arXiv, typeset properly. Highlighted words trade on the market.

Computer Science Machine Learning

cs.LG Watch Machine Learning (sign in to use)

Tuesday 20 January

Machine Learning · Cross-listed from cs.CL

Towards Execution-Grounded Automated AI Research

Automated AI research holds great potential to accelerate scientific discovery. However, current LLMs often generate plausible-looking but ineffective ideas. Execution grounding may help, but it is unclear whether automated execution is feasible and whether LLMs can learn from the execution feedback. To investigate these, we first build an automated executor to implement ideas and launch large-scale parallel GPU experiments to verify their effectiveness. We then convert two realistic research problems - LLM pre-training and post-training - into execution environments and demonstrate that our automated executor can implement a large fraction of the ideas sampled from frontier LLMs. We analyze two methods to learn from the execution feedback: evolutionary search and reinforcement learning. Execution-guided evolutionary search is sample-efficient: it finds a method that significantly outperforms the GRPO baseline (69.4% vs 48.0%) on post-training, and finds a pre-training recipe that outperforms the nanoGPT baseline (19.7 minutes vs 35.9 minutes) on pre-training, all within just ten search epochs. Frontier LLMs often generate meaningful algorithmic ideas during search, but they tend to saturate early and only occasionally exhibit scaling trends. Reinforcement learning from execution reward, on the other hand, suffers from mode collapse. It successfully improves the average reward of the ideator model but not the upper-bound, due to models converging on simple ideas. We thoroughly analyze the executed ideas and training dynamics to facilitate future efforts towards execution-grounded automated AI research.
Read the full abstract

All authors Chenglei Si, Zitong Yang, Yejin Choi, Emmanuel Candès, Diyi Yang, Tatsunori Hashimoto

Machine Learning · Cross-listed from cs.AI

Large Language Model-Powered Evolutionary Code Optimization on a Phylogenetic Tree

Optimizing scientific computing algorithms for modern GPUs is a labor-intensive and iterative process involving repeated code modification, benchmarking, and tuning across complex hardware and software stacks. Recent work has explored large language model (LLM)-assisted evolutionary methods for automated code optimization, but these approaches primarily rely on outcome-based selection and random mutation, underutilizing the rich trajectory information generated during iterative optimization. We propose PhyloEvolve, an LLM-agent system that reframes GPU-oriented algorithm optimization as an In-Context Reinforcement Learning (ICRL) problem. This formulation enables trajectory-conditioned reuse of optimization experience without model retraining. PhyloEvolve integrates Algorithm Distillation and prompt-based Decision Transformers into an iterative workflow, treating sequences of algorithm modifications and performance feedback as first-class learning signals. To organize optimization history, we introduce a phylogenetic tree representation that captures inheritance, divergence, and recombination among algorithm variants, enabling backtracking, cross-lineage transfer, and reproducibility. The system combines elite trajectory pooling, multi-island parallel exploration, and containerized execution to balance exploration and exploitation across heterogeneous hardware. We evaluate PhyloEvolve on scientific computing workloads including PDE solvers, manifold learning, and spectral graph algorithms, demonstrating consistent improvements in runtime, memory efficiency, and correctness over baseline and evolutionary methods. Code is published at: https://github.com/annihi1ation/phylo_evolve
Read the full abstract

All authors Leyi Zhao, Weijie Huang, Yitong Guo, Jiang Bian, Chenghong Wang, Xuhong Zhang

Machine Learning

On the Runway Cascade of Transformers for Language Modeling

In decoder-only (causal) transformers, the computation graph created by causal masking routes information through both direct-path attention and indirect paths formed by intermediate tokens. We denote these indirect paths between token pairs as their runways. We argue that certain failure modes of causal transformers as observed by a growing body of recent works are likely exacerbated by a misalignment between these two information propagation modes. We formalize runway cascade as a phenomenon whereby this misalignment results in redundancies and irrelevant information cascading to token representations despite adequately learned attention patterns. As a solution, we propose runway-aware rewiring as a more explicit way of incorporating runway context directly into each token's direct-path attention. This mechanism re-wires the attention pattern for each token based on a summary of its runway landscape, enabling awareness of accumulating representational influences and allowing for more balanced information propagation. Our proposed methodology introduces no additional parameters and can seamlessly be integrated into standard attention mechanism. Empirically, our rewired transformer results in steady improvements in general language modeling as well as noticeably stronger information retrieval and extrapolation abilities compared to standard transformers.
Read the full abstract

Machine Learning

How Worst-Case Are Adversarial Attacks? Linking Adversarial and Perturbation Robustness

Adversarial attacks are widely used to identify model vulnerabilities; however, their validity as proxies for robustness to random perturbations remains debated. We ask whether an adversarial example provides a representative estimate of misprediction risk under stochastic perturbations of the same magnitude, or instead reflects an atypical worst-case event. To address this question, we introduce a probabilistic analysis that quantifies this risk with respect to directionally biased perturbation distributions, parameterized by a concentration factor $\kappa$ that interpolates between isotropic noise and adversarial directions. Building on this, we study the limits of this connection by proposing an attack strategy designed to probe vulnerabilities in regimes that are statistically closer to uniform noise. Experiments on ImageNet and CIFAR-10 systematically benchmark multiple attacks, revealing when adversarial success meaningfully reflects robustness to perturbations and when it does not, thereby informing their use in safety-oriented robustness evaluation.
Read the full abstract

Machine Learning

Learning PDE Solvers with Physics and Data: A Unifying View of Physics-Informed Neural Networks and Neural Operators

Partial differential equations (PDEs) are central to scientific modeling. Modern workflows increasingly rely on learning-based components to support model reuse, inference, and integration across large computational processes. Despite the emergence of various physics-aware data-driven approaches, the field still lacks a unified perspective to uncover their relationships, limitations, and appropriate roles in scientific workflows. To this end, we propose a unifying perspective to place two dominant paradigms: Physics-Informed Neural Networks (PINNs) and Neural Operators (NOs), within a shared design space. We organize existing methods from three fundamental dimensions: what is learned, how physical structures are integrated into the learning process, and how the computational load is amortized across problem instances. In this way, many challenges can be best understood as consequences of these structural properties of learning PDEs. By analyzing advances through this unifying view, our survey aims to facilitate the development of reliable learning-based PDE solvers and catalyze a synthesis of physics and data.
Read the full abstract

All authors Yilong Dai, Shengyu Chen, Ziyi Wang, Xiaowei Jia, Yiqun Xie, Vipin Kumar, Runlong Yu

Machine Learning · Cross-listed from stat.ML

Large Data Limits of Laplace Learning for Gaussian Measure Data in Infinite Dimensions

Laplace learning is a semi-supervised method, a solution for finding missing labels from a partially labeled dataset utilizing the geometry given by the unlabeled data points. The method minimizes a Dirichlet energy defined on a (discrete) graph constructed from the full dataset. In finite dimensions the asymptotics in the large (unlabeled) data limit are well understood with convergence from the graph setting to a continuum Sobolev semi-norm weighted by the Lebesgue density of the data-generating measure. The lack of the Lebesgue measure on infinite-dimensional spaces requires rethinking the analysis if the data aren't finite-dimensional. In this paper we make a first step in this direction by analyzing the setting when the data are generated by a Gaussian measure on a Hilbert space and proving pointwise convergence of the graph Dirichlet energy.
Read the full abstract

Machine Learning · Cross-listed from cs.CR

Uncovering and Understanding FPR Manipulation Attack in Industrial IoT Networks

In the network security domain, due to practical issues – including imbalanced data and heterogeneous legitimate network traffic – adversarial attacks in machine learning-based NIDSs have been viewed as attack packets misclassified as benign. Due to this prevailing belief, the possibility of (maliciously) perturbed benign packets being misclassified as attack has been largely ignored. In this paper, we demonstrate that this is not only theoretically possible, but also a particular threat to NIDS. In particular, we uncover a practical cyberattack, FPR manipulation attack (FPA), especially targeting industrial IoT networks, where domain-specific knowledge of the widely used MQTT protocol is exploited and a systematic simple packet-level perturbation is performed to alter the labels of benign traffic samples without employing traditional gradient-based or non-gradient-based methods. The experimental evaluations demonstrate that this novel attack results in a success rate of 80.19% to 100%. In addition, while estimating impacts in the Security Operations Center, we observe that even a small fraction of false positive alerts, irrespective of different budget constraints and alert traffic intensities, can increase the delay of genuine alerts investigations up to 2 hr in a single day under normal operating conditions. Furthermore, a series of relevant statistical and XAI analyses is conducted to understand the key factors behind this remarkable success. Finally, we explore the effectiveness of the FPA packets to enhance models' robustness through adversarial training and investigate the changes in decision boundaries accordingly.
Read the full abstract

Machine Learning · Cross-listed from cs.CV

GutenOCR: A Grounded Vision-Language Front-End for Documents

GutenOCR is a family of grounded OCR front-ends obtained by fine-tuning Qwen2.5-VL-3B and Qwen2.5-VL-7B. The resulting single-checkpoint vision-language models expose reading, detection, and grounding through a unified, prompt-based interface. Trained on business documents, scientific articles, and synthetic grounding data, the models support full-page and localized reading with line- and paragraph-level bounding boxes and conditional “where is x?” queries. We introduce a grounded OCR evaluation protocol and show that GutenOCR-7B more than doubles the composite grounded OCR score of its Qwen2.5-VL-7B backbone on 10.5K held-out business and scientific pages (0.40 to 0.82). On Fox and OmniDocBench v1.5, our approach substantially improves region- and line-level OCR as well as text-detection recall, but reveals trade-offs in page-level linearization, color-guided OCR, and formula-heavy layouts.
Read the full abstract

All authors Hunter Heidenreich, Ben Elliott, Olivia Dinica, Yosheb Getachew

Machine Learning

Stabilizing autoregressive forecasts in chaotic systems via multi-rate latent recurrence

Long-horizon autoregressive forecasting of chaotic dynamical systems remains challenging due to rapid error amplification and distribution shift: small one-step inaccuracies compound into physically inconsistent rollouts and collapse of large-scale statistics. We introduce MSR-HINE, a hierarchical implicit forecaster that augments multiscale latent priors with multi-rate recurrent modules operating at distinct temporal scales. At each step, coarse-to-fine recurrent states generate latent priors, an implicit one-step predictor refines the state with multiscale latent injections, and a gated fusion with posterior latents enforces scale-consistent updates; a lightweight hidden-state correction further aligns recurrent memories with fused latents. The resulting architecture maintains long-term context on slow manifolds while preserving fast-scale variability, mitigating error accumulation in chaotic rollouts. Across two canonical benchmarks, MSR-HINE yields substantial gains over a U-Net autoregressive baseline: on Kuramoto-Sivashinsky it reduces end-horizon RMSE by 62.8% at H=400 and improves end-horizon ACC by +0.983 (from -0.155 to 0.828), extending the ACC >= 0.5 predictability horizon from 241 to 400 steps; on Lorenz-96 it reduces RMSE by 27.0% at H=100 and improves end horizon ACC by +0.402 (from 0.144 to 0.545), extending the ACC >= 0.5 horizon from 58 to 100 steps.
Read the full abstract

Machine Learning

GPU-accelerated simulated annealing based on p-bits with real-world device-variability modeling

Probabilistic computing using probabilistic bits (p-bits) presents an efficient alternative to traditional CMOS logic for complex problem-solving, including simulated annealing and machine learning. Realizing p-bits with emerging devices such as magnetic tunnel junctions (MTJs) introduces device variability, which was expected to negatively impact computational performance. However, this study reveals an unexpected finding: device variability can not only degrade but also enhance algorithm performance, particularly by leveraging timing variability. This paper introduces a GPU-accelerated, open-source simulated annealing framework based on p-bits that models key device variability factors – timing, intensity, and offset – to reflect real-world device behavior. Through CUDA-based simulations, our approach achieves a two-order magnitude speedup over CPU implementations on the MAX-CUT benchmark with problem sizes ranging from 800 to 20,000 nodes. By providing a scalable and accessible tool, this framework aims to advance research in probabilistic computing, enabling optimization applications in diverse fields.
Read the full abstract

Machine Learning

Adaptive KDE for Real-Time Thresholding: Prioritized Queues for Financial Crime Investigation

We study the problem of converting a continuous stream of risk scores into stable decision thresholds under non-stationary score distributions. This problem arises in a wide range of detection systems where scores must be partitioned into prioritized processing regions while preserving semantic consistency over time.
Read the full abstract

Machine Learning · Cross-listed from cs.AI

On the Generalization Gap in LLM Planning: Tests and Verifier-Reward RL

Recent work shows that fine-tuned Large Language Models (LLMs) can achieve high valid plan rates on PDDL planning tasks. However, it remains unclear whether this reflects transferable planning competence or domain-specific memorization. In this work, we fine-tune a 1.7B-parameter LLM on 40,000 domain-problem-plan tuples from 10 IPC 2023 domains, and evaluate both in-domain and cross-domain generalization. While the model reaches 82.9% valid plan rate in in-domain conditions, it achieves 0% on two unseen domains. To analyze this failure, we introduce three diagnostic interventions, namely (i) instance-wise symbol anonymization, (ii) compact plan serialization, and (iii) verifier-reward fine-tuning using the VAL validator as a success-focused reinforcement signal. Symbol anonymization and compact serialization cause significant performance drops despite preserving plan semantics, thus revealing strong sensitivity to surface representations. Verifier-reward fine-tuning reaches performance saturation in half the supervised training epochs, but does not improve cross-domain generalization. For the explored configurations, in-domain performance plateaus around 80%, while cross-domain performance collapses, suggesting that our fine-tuned model relies heavily on domain-specific patterns rather than transferable planning competence in this setting. Our results highlight a persistent generalization gap in LLM-based planning and provide diagnostic tools for studying its causes.
Read the full abstract

All authors Valerio Belcamino, Nicholas Attolino, Alessio Capitanelli, Fulvio Mastrogiovanni

Machine Learning · Cross-listed from cs.AI

VisTIRA: Closing the Image-Text Modality Gap in Visual Math Reasoning via Structured Tool Integration

Vision-language models (VLMs) lag behind text-only language models on mathematical reasoning when the same problems are presented as images rather than text. We empirically characterize this as a modality gap: the same question in text form yields markedly higher accuracy than its visually typeset counterpart, due to compounded failures in reading dense formulas, layout, and mixed symbolic-diagrammatic context. First, we introduce VisTIRA (Vision and Tool-Integrated Reasoning Agent), a tool-integrated reasoning framework that enables structured problem solving by iteratively decomposing a given math problem (as an image) into natural language rationales and executable Python steps to determine the final answer. Second, we build a framework to measure and improve visual math reasoning: a LaTeX-based pipeline that converts chain-of-thought math corpora (e.g., NuminaMath) into challenging image counterparts, and a large set of synthetic tool-use trajectories derived from a real-world, homework-style image dataset (called SnapAsk) for fine-tuning VLMs. Our experiments show that tool-integrated supervision improves image-based reasoning, and OCR grounding can further narrow the gap for smaller models, although its benefit diminishes at scale. These findings highlight that modality gap severity inversely correlates with model size, and that structured reasoning and OCR-based grounding are complementary strategies for advancing visual mathematical reasoning.
Read the full abstract

All authors Saeed Khaki, Ashudeep Singh, Nima Safaei, Kamal Ginotra

Machine Learning · Cross-listed from cs.CV

Vision-Based Natural Language Scene Understanding for Autonomous Driving: An Extended Dataset and a New Model for Traffic Scene Description Generation

Traffic scene understanding is essential for enabling autonomous vehicles to accurately perceive and interpret their environment, thereby ensuring safe navigation. This paper presents a novel framework that transforms a single frontal-view camera image into a concise natural language description, effectively capturing spatial layouts, semantic relationships, and driving-relevant cues. The proposed model leverages a hybrid attention mechanism to enhance spatial and semantic feature extraction and integrates these features to generate contextually rich and detailed scene descriptions. To address the limited availability of specialized datasets in this domain, a new dataset derived from the BDD100K dataset has been developed, with comprehensive guidelines provided for its construction. Furthermore, the study offers an in-depth discussion of relevant evaluation metrics, identifying the most appropriate measures for this task. Extensive quantitative evaluations using metrics such as CIDEr and SPICE, complemented by human judgment assessments, demonstrate that the proposed model achieves strong performance and effectively fulfills its intended objectives on the newly developed dataset.
Read the full abstract

Machine Learning · Cross-listed from quant-ph

Quantum Super-resolution by Adaptive Non-local Observables

Super-resolution (SR) seeks to reconstruct high-resolution (HR) data from low-resolution (LR) observations. Classical deep learning methods have advanced SR substantially, but require increasingly deeper networks, large datasets, and heavy computation to capture fine-grained correlations. In this work, we present the first study to investigate quantum circuits for SR. We propose a framework based on Variational Quantum Circuits (VQCs) with Adaptive Non-Local Observable (ANO) measurements. Unlike conventional VQCs with fixed Pauli readouts, ANO introduces trainable multi-qubit Hermitian observables, allowing the measurement process to adapt during training. This design leverages the high-dimensional Hilbert space of quantum systems and the representational structure provided by entanglement and superposition. Experiments demonstrate that ANO-VQCs achieve up to five-fold higher resolution with a relatively small model size, suggesting a promising new direction at the intersection of quantum machine learning and super-resolution.
Read the full abstract

All authors Hsin-Yi Lin, Huan-Hsin Tseng, Samuel Yen-Chi Chen, Shinjae Yoo

Machine Learning · Cross-listed from stat.ML

Meta Flow Maps enable scalable reward alignment

Controlling generative models is computationally expensive. This is because optimal alignment with a reward function–whether via inference-time steering or fine-tuning–requires estimating the value function. This task demands access to the conditional posterior $p_{1|t}(x_1|x_t)$, the distribution of clean data $x_1$ consistent with an intermediate state $x_t$, a requirement that typically compels methods to resort to costly trajectory simulations. To address this bottleneck, we introduce Meta Flow Maps (MFMs), a framework extending consistency models and flow maps into the stochastic regime. MFMs are trained to perform stochastic one-step posterior sampling, generating arbitrarily many i.i.d. draws of clean data $x_1$ from any intermediate state. Crucially, these samples provide a differentiable reparametrization that unlocks efficient value function estimation. We leverage this capability to solve bottlenecks in both paradigms: enabling inference-time steering without inner rollouts, and facilitating unbiased, off-policy fine-tuning to general rewards. Empirically, our single-particle steered-MFM sampler outperforms a Best-of-1000 baseline on ImageNet across multiple rewards at a fraction of the compute.
Read the full abstract

All authors Peter Potaptchik, Adhi Saravanan, Abbas Mammadov, Alvaro Prat, Michael S. Albergo, Yee Whye Teh

Machine Learning · Cross-listed from astro-ph.CO

Cosmo-FOLD: Fast generation and upscaling of field-level cosmological maps with overlap latent diffusion

We demonstrate the capabilities of probabilistic diffusion models to reduce dramatically the computational cost of expensive hydrodynamical simulations to study the relationship between observable baryonic cosmological probes and dark matter at field level and well into the non-linear regime. We introduce a novel technique, Cosmo-FOLD (Cosmological Fields via Overlap Latent Diffusion) to rapidly generate accurate and arbitrarily large cosmological and astrophysical 3-dimensional fields, conditioned on a given input field. We are able to generate TNG300-2 dark matter density and gas temperature fields from a model trained only on ~1% of the volume (a process we refer to as `upscaling'), reproducing both large scale coherent dark matter filaments and power spectra to within 10% for wavenumbers k <= 5 h Mpc^-1. These results are obtained within a small fraction of the original simulation cost and produced on a single GPU. Beyond one and two points statistics, the bispectrum is also faithfully reproduced through the inclusion of positional encodings. Finally, we demonstrate Cosmo-FOLD's generalisation capabilities by upscaling a CAMELS volume of 25 (Mpc h^-1)^3 to a full TNG300-2 volume of 205 (Mpc h^-1)^3$ with no fine-tuning. Cosmo-FOLD opens the door to full field-level simulation-based inference on cosmological scale.
Read the full abstract

Machine Learning

VJEPA: Variational Joint Embedding Predictive Architectures as Probabilistic World Models

Joint Embedding Predictive Architectures (JEPA) offer a scalable paradigm for self-supervised learning by predicting latent representations rather than reconstructing high-entropy observations. However, existing formulations rely on deterministic regression objectives, which mask probabilistic semantics and limit its applicability in stochastic control. In this work, we introduce Variational JEPA (VJEPA), a probabilistic generalization that learns a predictive distribution over future latent states via a variational objective. We show that VJEPA unifies representation learning with Predictive State Representations (PSRs) and Bayesian filtering, establishing that sequential modeling does not require autoregressive observation likelihoods. Theoretically, we prove that VJEPA representations can serve as sufficient information states for optimal control without pixel reconstruction, while providing formal guarantees for collapse avoidance. We further propose Bayesian JEPA (BJEPA), an extension that factorizes the predictive belief into a learned dynamics expert and a modular prior expert, enabling zero-shot task transfer and constraint (e.g. goal, physics) satisfaction via a Product of Experts. Empirically, through a noisy environment experiment, we demonstrate that VJEPA and BJEPA successfully filter out high-variance nuisance distractors that cause representation collapse in generative baselines. By enabling principled uncertainty estimation (e.g. constructing credible intervals via sampling) while remaining likelihood-free regarding observations, VJEPA provides a foundational framework for scalable, robust, uncertainty-aware planning in high-dimensional, noisy environments.
Read the full abstract

Machine Learning · Cross-listed from cs.MA

MARBLE: Multi-Agent Reasoning for Bioinformatics Learning and Evolution

Motivation: Developing high-performing bioinformatics models typically requires repeated cycles of hypothesis formulation, architectural redesign, and empirical validation, making progress slow, labor-intensive, and difficult to reproduce. Although recent LLM-based assistants can automate isolated steps, they lack performance-grounded reasoning and stability-aware mechanisms required for reliable, iterative model improvement in bioinformatics workflows. Results: We introduce MARBLE, an execution-stable autonomous model refinement framework for bioinformatics models. MARBLE couples literature-aware reference selection with structured, debate-driven architectural reasoning among role-specialized agents, followed by autonomous execution, evaluation, and memory updates explicitly grounded in empirical performance. Across spatial transcriptomics domain segmentation, drug-target interaction prediction, and drug response prediction, MARBLE consistently achieves sustained performance improvements over strong baselines across multiple refinement cycles, while maintaining high execution robustness and low regression rates. Framework-level analyses demonstrate that structured debate, balanced evidence selection, and performance-grounded memory are critical for stable, repeatable model evolution, rather than single-run or brittle gains. Availability: Source code, data and Supplementary Information are available at https://github.com/PRISM-DGU/MARBLE.
Read the full abstract

All authors Sunghyun Kim, Seokwoo Yun, Youngseo Yun, Youngrak Lee, Sangsoo Lim

Machine Learning

DiSPA: Differential Substructure-Pathway Attention for Drug Response Prediction

Accurate prediction of drug response in precision medicine requires models that capture how specific chemical substructures interact with cellular pathway states. However, most existing deep learning approaches treat chemical and transcriptomic modalities independently or combine them only at late stages, limiting their ability to model fine-grained, context-dependent mechanisms of drug action. In addition, vanilla attention mechanisms are often sensitive to noise and sparsity in high-dimensional biological networks, hindering both generalization and interpretability. We present DiSPA (Differential Substructure-Pathway Attention), a framework that models bidirectional interactions between chemical substructures and pathway-level gene expression. DiSPA introduces differential cross-attention to suppress spurious associations while enhancing context-relevant interactions. On the GDSC benchmark, DiSPA achieves state-of-the-art performance, with strong improvements in the disjoint setting. These gains are consistent across random and drug-blind splits, suggesting improved robustness. Analyses of attention patterns indicate more selective and concentrated interactions compared to standard cross-attention. Exploratory evaluation shows that differential attention better prioritizes predefined target-related pathways, although this does not constitute mechanistic validation. DiSPA also shows promising generalization on external datasets (CTRP) and cross-dataset settings, although further validation is needed. It further enables zero-shot application to spatial transcriptomics, providing exploratory insights into region-specific drug sensitivity patterns without ground-truth validation.
Read the full abstract

All authors Yewon Han, Sunghyun Kim, Eunyi Jeong, Sungkyung Lee, Seokwoo Yun, Sangsoo Lim

Machine Learning · Cross-listed from cs.CR

Turn-Based Structural Triggers: Structure-Conditioned Backdoors in Multi-Turn LLMs

Large Language Models (LLMs) are increasingly deployed as multi-turn assistants and customized through instruction tuning with project-specific training components. This practice creates a supply-chain risk when organizations reuse third-party fine-tuning frameworks, trainer extensions, or outsourced training code: an adversary who subtly compromises the loss-computation component can inject malicious supervision during fine-tuning while leaving the stored training corpus, model architecture, and deployment interface unchanged. Existing LLM backdoors and defenses are largely prompt-centric, relying on lexical, syntactic, or semantic patterns in user inputs while overlooking structural signals in multi-turn conversations. We propose Turn-based Structural Trigger (TST), a prompt-free backdoor that uses dialogue turn position as its activation condition. TST exploits structural cues implicitly encoded by chat templates and is implanted without modifying the stored dialogue corpus. Its trigger is automatically present once the conversation reaches the attacker-specified turn, making activation independent of downstream user inputs and resistant to prompt filtering, sanitization, and paraphrasing. In our primary setting, the model behaves normally during early interactions and activates the attacker-defined behavior only after the conversation reaches the designated structural condition. Across four open-source LLM families, TST achieves an average Attack Success Rate of 98.10% on target turns and a Clean Rate of 99.96% on non-target turns, while retaining 97.78% of clean-model utility. These results identify dialogue structure as an overlooked attack surface and motivate structure-aware auditing beyond prompt inspection.
Read the full abstract

All authors Yiyang Lu, Jinwen He, Yue Zhao, Kai Chen, Ruigang Liang, Cheng Hong, Yingjun Zhang

Machine Learning

Log anomaly detection via Meta Learning and Prototypical Networks for Cross domain generalization

Log anomaly detection is essential for system reliability, but it is extremely challenging to do considering it involves class imbalance. Additionally, the models trained in one domain are not applicable to other domains, necessitating the need for cross-domain adaptation (such as HDFS and Linux). Traditional detection models often fail to generalize due to significant data drift and the inherent absence of labeled anomalies in new target domains. To handle the above challenges, we proposed a new end-to-end framework based on a meta-learning approach. Our methodology first gets the data ready by combining a Drain3 log parsing mechanism with a dynamic drift-based labeling technique that uses semantic and fuzzy matching to move existing anomaly knowledge from one source to another. BERT-based semantic embeddings are obtained, and the feature selection is invoked to reduce the dimensionality. Later, Model Agnostic Meta-Learning (MAML) and Prototypical Networks models are trained to adapt quickly and effectively. The SMOTE oversampling method is employed to handle imbalances in the data. All the results are obtained by employing the leave-one-out source method, and the corresponding mean F1 scores are reported. Our empirical findings validate that the proposed meta-learning-driven approach yielded the highest mean F1 score and proved to be effective for cross-domain settings.
Read the full abstract

Machine Learning

Hierarchical Contextual Uplift Bandits for Catalog Personalization

Contextual Bandit (CB) algorithms are widely adopted for personalized recommendations but often struggle in dynamic environments typical of fantasy sports, where rapid changes in user behavior and dramatic shifts in reward distributions due to external influences necessitate frequent retraining. To address these challenges, we propose a Hierarchical Contextual Uplift Bandit framework. Our framework dynamically adjusts contextual granularity from broad, system-wide insights to detailed, user-specific contexts, using contextual similarity to facilitate effective policy transfer and mitigate cold-start issues. Additionally, we integrate uplift modeling principles into our approach. Results from large-scale A/B testing on the Dream11 fantasy sports platform show that our method significantly enhances recommendation quality, achieving a 0.4% revenue improvement while also improving user satisfaction metrics compared to the current production system. We subsequently deployed this system to production as the default catalog personalization system in May 2025 and observed a further 0.5% revenue improvement.
Read the full abstract

All authors Anupam Agrawal, Rajesh Mohanty, Shamik Bhattacharjee, Abhimanyu Mittal

Machine Learning · Cross-listed from cs.CV

LURE: Latent Space Unblocking for Multi-Concept Reawakening in Diffusion Models

Concept erasure aims to suppress sensitive content in diffusion models, but recent studies show that erased concepts can still be reawakened, revealing vulnerabilities in erasure methods. Existing reawakening methods mainly rely on prompt-level optimization to manipulate sampling trajectories, neglecting other generative factors, which limits a comprehensive understanding of the underlying dynamics. In this paper, we model the generation process as an implicit function to enable a comprehensive theoretical analysis of multiple factors, including text conditions, model parameters, and latent states. We theoretically show that perturbing each factor can reawaken erased concepts. Building on this insight, we propose a novel concept reawakening method: Latent space Unblocking for concept REawakening (LURE), which reawakens erased concepts by reconstructing the latent space and guiding the sampling trajectory. Specifically, our semantic re-binding mechanism reconstructs the latent space by aligning denoising predictions with target distributions to reestablish severed text-visual associations. However, in multi-concept scenarios, naive reconstruction can cause gradient conflicts and feature entanglement. To address this, we introduce Gradient Field Orthogonalization, which enforces feature orthogonality to prevent mutual interference. Additionally, our Latent Semantic Identification-Guided Sampling (LSIS) ensures stability of the reawakening process via posterior density verification. Extensive experiments demonstrate that LURE enables simultaneous, high-fidelity reawakening of multiple erased concepts across diverse erasure tasks and methods.
Read the full abstract

All authors Mengyu Sun, Ziyuan Yang, Andrew Beng Jin Teoh, Junxu Liu, Haibo Hu, Yi Zhang

Machine Learning

Yuan3.0 Ultra: A Trillion-Parameter Enterprise-Oriented MoE LLM

We introduce Yuan3.0 Ultra, an open-source Mixture-of-Experts (MoE) large language model featuring 68.8B activated parameters and 1010B total parameters, specially designed to enhance performance on enterprise scenarios tasks while maintaining competitive capabilities on general purpose tasks. We propose Layer-Adaptive Expert Pruning (LAEP) algorithm designed for the pre-training stage of MoE LLMs. In contrast to previous expert pruning approaches that operate primarily in the post-training phase, the proposed algorithm enhances training efficiency by selectively pruning underutilized experts and reorganizing experts across computing devices according to token distribution statistics. Comprehensive experiments demonstrate that LAEP effectively reduces model size and substantially improves pre-training efficiency. When pre-training Yuan3.0 Ultra from scratch original with 1515B parameters, this algorithm delivers a 49% boost in pre-training efficiency and a 33.3% reduction in total parameters, while preserving the model's outstanding multi-domain performance. On enterprise scenario benchmarks including Docmatix, ChatRAG, SummEval and MMTab, Yuan3.0 Ultra achieves leading accuracy. The model and codes are publicly available at https://github.com/Yuan-lab-LLM/Yuan3.0-Ultra.
Read the full abstract

All authors YuanLab.ai: Shawn Wu, Jiangang Luo, Darcy Chen, Sean Wang, Louie Li, Allen Wang, Xudong Zhao, Tong Yu, Bach Li, Joseph Shen, Gawain Ma, Jasper Jia, Marcus Mao, Claire Wang, Hunter He, Carol Wang, Zera Zhang, Jason Wang, Chonly Shen, Leo Zhang, Logan Chen, Qasim Meng, James Gong, Daniel Zhao, Penn Zheng, Owen Zhu

Machine Learning

Jet-RL: Enabling On-Policy FP8 Reinforcement Learning with Unified Training and Rollout Precision Flow

Reinforcement learning (RL) is essential for enhancing the complex reasoning capabilities of large language models (LLMs). However, existing RL training pipelines are computationally inefficient and resource-intensive, with the rollout phase accounting for over 70% of total training time. Quantized RL training, particularly using FP8 precision, offers a promising approach to mitigating this bottleneck. A commonly adopted strategy applies FP8 precision during rollout while retaining BF16 precision for training. In this work, we present the first comprehensive study of FP8 RL training and demonstrate that the widely used BF16-training + FP8-rollout strategy suffers from severe training instability and catastrophic accuracy collapse under long-horizon rollouts and challenging tasks. Our analysis shows that these failures stem from the off-policy nature of the approach, which introduces substantial numerical mismatch between training and inference. Motivated by these observations, we propose Jet-RL, an FP8 RL training framework that enables robust and stable RL optimization. The key idea is to adopt a unified FP8 precision flow for both training and rollout, thereby minimizing numerical discrepancies and eliminating the need for inefficient inter-step calibration. Extensive experiments validate the effectiveness of Jet-RL: our method achieves up to 33% speedup in the rollout phase, up to 41% speedup in the training phase, and a 16% end-to-end speedup over BF16 training, while maintaining stable convergence across all settings and incurring negligible accuracy degradation.
Read the full abstract

All authors Haocheng Xi, Charlie Ruan, Peiyuan Liao, Yujun Lin, Han Cai, Yilong Zhao, Shuo Yang, Kurt Keutzer, Song Han, Ligeng Zhu

Machine Learning · Cross-listed from cs.CL

APEX-Agents

We introduce the AI Productivity Index for Agents (APEX-Agents), a benchmark for assessing whether AI agents can execute long-horizon, cross-application tasks created by investment banking analysts, management consultants, and corporate lawyers. APEX-Agents requires agents to navigate realistic work environments with files and tools. We test eight agents for the leaderboard using Pass@1. Gemini 3 Flash (Thinking=High) achieves the highest score of 24.0%, followed by GPT-5.2 (Thinking=High), Claude Opus 4.5 (Thinking=High), and Gemini 3 Pro (Thinking=High). We open source the APEX-Agents benchmark (n=480) with all prompts, rubrics, gold outputs, files, and metadata. We also open source Archipelago, our infrastructure for agent execution and evaluation.
Read the full abstract

All authors Bertie Vidgen, Austin Mann, Abby Fennelly, John Wright Stanly, Lucas Rothman, Marco Burstein, Julien Benchek, David Ostrofsky, Anirudh Ravichandran, Debnil Sur, Neel Venugopal, Alannah Hsia, Isaac Robinson, Calix Huang, Olivia Varones, Daniyal Khan, Michael Haines, Austin Bridges, Jesse Boyle, Koby Twist, Zach Richards, Chirag Mahapatra, Brendan Foody, Osvald Nitski

Machine Learning

Spatiotemporal Wildfire Prediction and Reinforcement Learning for Helitack Suppression

Wildfires are growing in frequency and intensity, devastating ecosystems and communities while causing billions of dollars in suppression costs and economic damage annually in the U.S. Traditional wildfire management is mostly reactive, addressing fires only after they are detected. We introduce FireCastRL, a proactive artificial intelligence (AI) framework that combines wildfire forecasting with intelligent suppression strategies. Our framework first uses a deep spatiotemporal model to predict wildfire ignition. For high-risk predictions, we deploy a pre-trained reinforcement learning (RL) agent to execute real-time suppression tactics with helitack units inside a physics-informed 3D simulation. The framework generates a threat assessment report to help emergency responders optimize resource allocation and planning. In addition, we are publicly releasing a large-scale, spatiotemporal dataset containing $\mathbf{9.5}$ million samples of environmental variables for wildfire prediction. Our work demonstrates how deep learning and RL can be combined to support both forecasting and tactical wildfire response. More details can be found at https://sites.google.com/view/firecastrl.
Read the full abstract

All authors Shaurya Mathur, Shreyas Bellary Manjunath, Nitin Kulkarni, Alina Vereshchaka

Machine Learning · Cross-listed from astro-ph.IM

Opportunities in AI/ML for the Rubin LSST Dark Energy Science Collaboration

The Vera C. Rubin Observatory's Legacy Survey of Space and Time (LSST) will produce unprecedented volumes of heterogeneous astronomical data (images, catalogs, and alerts) that challenge traditional analysis pipelines. The LSST Dark Energy Science Collaboration (DESC) aims to derive robust constraints on dark energy and dark matter from these data, requiring methods that are statistically powerful, scalable, and operationally reliable. Artificial intelligence and machine learning (AI/ML) are already embedded across DESC science workflows, from photometric redshifts and transient classification to weak lensing inference and cosmological simulations. Yet their utility for precision cosmology hinges on trustworthy uncertainty quantification, robustness to covariate shift and model misspecification, and reproducible integration within scientific pipelines. This white paper surveys the current landscape of AI/ML across DESC's primary cosmological probes and cross-cutting analyses, revealing that the same core methodologies and fundamental challenges recur across disparate science cases. Since progress on these cross-cutting challenges would benefit multiple probes simultaneously, we identify key methodological research priorities, including Bayesian inference at scale, physics-informed methods, validation frameworks, and active learning for discovery. With an eye on emerging techniques, we also explore the potential of the latest foundation model methodologies and LLM-driven agentic AI systems to reshape DESC workflows, provided their deployment is coupled with rigorous evaluation and governance. Finally, we discuss critical software, computing, data infrastructure, and human capital requirements for the successful deployment of these new methodologies, and consider associated risks and opportunities for broader coordination with external actors.
Read the full abstract

All authors LSST Dark Energy Science Collaboration, Eric Aubourg, Camille Avestruz, Matthew R. Becker, Biswajit Biswas, Rahul Biswas, Boris Bolliet, Adam S. Bolton, Clecio R. Bom, Raphaël Bonnet-Guerrini, Alexandre Boucaud, Jean-Eric Campagne, Chihway Chang, Aleksandra Ćiprijanović, Johann Cohen-Tanugi, Michael W. Coughlin, John Franklin Crenshaw, Juan C. Cuevas-Tello, Juan de Vicente, Seth W. Digel, Steven Dillmann, Mariano Javier de León Dominguez Romero, Alex Drlica-Wagner, Sydney Erickson, Alexander T. Gagliano, Christos Georgiou, Aritra Ghosh, Matthew Grayling, Kirill A. Grishin, Alan Heavens, Lindsay R. House, Mustapha Ishak, Wassim Kabalan, Arun Kannawadi, François Lanusse, C. Danielle Leonard, Pierre-François Léget, Michelle Lochner, Yao-Yuan Mao, Peter Melchior, Grant Merz, Martin Millon, Anais Möller, Gautham Narayan, Yuuki Omori, Hiranya Peiris, Laurence Perreault-Levasseur, Andrés A. Plazas Malagón, Nesar Ramachandra, Benjamin Remy, Cécile Roucelle, Jaime Ruiz-Zapatero, Stefan Schuldt, Ignacio Sevilla-Noarbe, Ved G. Shah, Tjitske Starkenburg, Stephen Thorp, Laura Toribio San Cipriano, Tilman Tröster, Roberto Trotta, Padma Venkatraman, Amanda Wasserman, Tim White, Justine Zeghal, Tianqing Zhang, Yuanyuan Zhang

Machine Learning

Q-learning with Adjoint Matching

We propose Q-learning with Adjoint Matching (QAM), a novel TD-based reinforcement learning (RL) algorithm that tackles a long-standing challenge in continuous-action RL: efficient optimization of an expressive diffusion or flow-matching policy with respect to a parameterized Q-function. Effective optimization requires exploiting the first-order information of the critic, but it is challenging to do so for flow or diffusion policies because direct gradient-based optimization via backpropagation through their multi-step denoising process is numerically unstable. Existing methods work around this either by only using the value and discarding the gradient information, or by relying on approximations that sacrifice policy expressivity or bias the learned policy. QAM sidesteps both of these challenges by leveraging adjoint matching, a recently proposed technique in generative modeling, which transforms the critic's action gradient to form a step-wise objective function that is free from unstable backpropagation, while providing an unbiased, expressive policy at the optimum. Combined with temporal-difference backup for critic learning, QAM consistently outperforms prior approaches on hard, sparse reward tasks in both offline and offline-to-online RL.
Read the full abstract