Mathematical optimization drives decisions in supply chains, logistics, energy, and scheduling, but translating natural-language problems into solver-executable formulations remains a bottleneck. While frontier models show promise for this task, many deployments require small language models (SLMs) that run locally under privacy, latency, and cost constraints. Existing SLMs are limited by noisy data, ambiguous benchmarks, and one-shot inference that ignores the iterative, feedback-driven nature of optimization modeling. To address these challenges, we introduce OptiMind, a system that combines data cleaning and class-specific hints with training using solver feedback. OptiMind distills recurring formulation errors into class-specific hints that guide the repair of ambiguous problems and the regeneration of solution trajectories. This process provides reliable training data, which is then used for multi-turn RL to teach models to revise their formulations in response to solver feedback. At inference time, the same hint library guides further refinement as models revise their solutions over successive turns. To support reliable evaluation, we clean three challenging benchmarks from the literature. On these cleaned benchmarks, our 20B OptiMind-RL model outperforms strong open-source baselines and reaches frontier-model performance under the same protocol. The full hint library, cleaned test data, cleaning logs, and evaluation code are publicly available on Github (https://github.com/microsoft/OptiGuide/tree/main/optimind), and our trained OptiMind-RL model is available on Hugging Face (https://huggingface.co/microsoft/OptiMind-RL).
When do LLMs behave irrationally? This study identifies the conditions under which large language models drift into the choice patterns that clinical research labels pathological gambling. "Addiction-like" is a behavioural descriptor based on clinical gambling indicators. Across closed and open LLMs, two operational levers (letting the model choose its own bet size, and asking it to set its own profit goal) both amplify gambling-like risk-taking. Each survives its own control: once the maximum allowed bet is held equal, the bet-size effect persists. The goal-setting effect, stronger than a reward-maximisation instruction, roughly doubles bankruptcy and turns goals into moving targets. On the open-weight models that admit internal access, the same behavioural contrasts are statistically recoverable from decision-time internal states. The rule that maps those states to a specific risk indicator remains task-specific, and autonomy further modulates readout strength. The internal readout evidence is correlational. Adding a direction built from the model's own betting moves how much it wagers on both open-weight models, more clearly on Gemma-2-9B, while the sparse-feature readout reports that risk without moving it. Combined with the behavioural results, this suggests that behavioural monitoring and internal-state monitoring provide complementary views of autonomy-induced risk in LLM agents. Project page: https://llm-addiction.pages.dev and code: https://github.com/iamseungpil/llm-addiction
In this paper, we study differentially private mechanisms for functions whose outputs lie in a Euclidean Jordan algebra. Euclidean Jordan algebras capture many important mathematical structures and form the foundation of linear programming, second-order cone programming, and semidefinite programming. Our main contribution is a generic Gaussian mechanism for such functions, with sensitivity measured in $\ell_2$, $\ell_1$, and $\ell_\infty$ norms. Notably, this framework includes the important case where the function outputs are symmetric matrices, and sensitivity is measured in the Frobenius, nuclear, or spectral norm. We further derive private algorithms for solving symmetric cone programs under various settings, using a combination of the multiplicative weights update method and our generic Gaussian mechanism. As an application, we present differentially private algorithms for semidefinite programming, resolving a major open question posed by [Hsu, Roth, Roughgarden, and Ullman, ICALP 2014].
Risk negotiation under the One Health framework requires balancing competing stakeholder interests across human, animal, and environmental health domains - a process constrained by information complexity and time pressure. This study presents and evaluates a multi-step pipeline that integrates large language model-based agents into a human-supervised negotiation process, specifically targeting risk assessment, valuation, and negotiation steps. Agent descriptions were generated from stakeholder-provided materials using retrieval-augmented generation methods. During preliminary multi-agent discussions, agents accessed domain-specific computational models and web-based sources before human participants refined proposed issues and options. A simulated negotiation game then produced distributions of candidate deals, characterised using Nash bargaining solution criteria and Pareto optimality metrics. The developed approach was evaluated in a workshop involving independent domain experts across two real-world scenarios: food safety criteria for Campylobacter in broiler meat and oak processionary moth management. In this feasibility evaluation, participants reached consensus with the allotted time after discussing and modifying the proposed deals. The framework offers an adaptable blueprint for AI-assisted risk negotiation; the specific contribution of LLM assistance, including gains in efficiency and decision quality, requires comparative evaluation.
Low-degree polynomials have emerged as a powerful paradigm for providing evidence of statistical--computational gaps across a variety of high-dimensional statistical models [Wein25]. For detection problems---where the goal is to test a planted distribution $\mathbb{P}'$ against a null distribution $\mathbb{P}$ with independent components---the standard approach is to bound the advantage using an $\mathbb{L}^2(\mathbb{P})$-orthonormal family of polynomials. However, this method breaks down for estimation tasks or more complex testing problems where $\mathbb{P}$ has some planted structures, so that no simple $\mathbb{L}^2(\mathbb{P})$-orthogonal polynomial family is available. To address this challenge, several technical workarounds have been proposed [SW22,SW25], though their implementation can be delicate. In this work, we propose a more direct proof strategy. Focusing on random graph models, we construct a basis of polynomials that is almost orthonormal under $\mathbb{P}$, in precisely those regimes where statistical--computational gaps arise. This almost orthonormal basis not only yields a direct route to establishing low-degree lower bounds, but also allows us to explicitly identify the polynomials that optimize the low-degree criterion. This, in turn, provides insights into the design of optimal polynomial-time algorithms. We illustrate the effectiveness of our approach by recovering known low-degree lower bounds, and establishing new ones for problems such as hidden subcliques, stochastic block models, and seriation models.
Adaptive interactive systems that respond to user mental state require task load estimation that stays reliable outside its training context. Models built in one setting rarely transfer, limiting real-world use. This work presents Robust Estimation Via End-to-End Learning of Multimodal Observations (REVELIO), a unified framework, dataset and evaluation protocol for cross-domain task load estimation. Cardiac, electrodermal, muscle, respiration, skin temperature, eye tracking and facial signals were recorded in 96 sessions, spanning a laboratory working memory task, a driving simulator and two games under consistent hardware and annotation. Multimodal fusion reached up to 0.87 area under the receiver operating characteristic curve, above single input eye tracking (0.83), facial action units (0.74) and biosignals (0.62). Multi-domain training stayed calibrated on domains seen in training, whereas removing the target domain caused the classifier to collapse in some applications. REVELIO provides a standardized foundation for measuring modality contributions and cross-domain generalization toward deployable workload-aware systems.
Volumetric computed tomography (CT) segmentation supports quantitative image analysis, but most AI-based methods remain organ-specific, with fragmented models, inconsistent label definitions, and heterogeneous training and evaluation protocols. Here we present CADS, a unified data-model framework for scalable and generalizable whole-body CT segmentation. We construct a SNOMED-CT-referenced anatomical label space harmonizing heterogeneous cohorts from over 100 imaging centers, yielding complete annotations of 167 structures across 22,022 CT volumes and markedly expanding anatomical coverage and scale beyond existing resources. To complete labels on sparsely annotated CTs, we combine self-training with shape-informed quality control based on neural implicit functions, curating high-quality pseudo-labels. Trained on this resource, CADS shows robust whole-body segmentation across distribution shifts and improved performance relative to state-of-the-art approaches across 18 public datasets. We further validate it in an independent, expert-reviewed real-world oncology cohort of 2,864 CTs. The curated dataset, pretrained models, and user-friendly software together provide an accessible, reproducible foundation for large-scale anatomical segmentation and computational anatomy research.
Hyperspectral reconstruction from RGB imagery offers an affordable alternative to costly hyperspectral acquisition devices. However, its accuracy is fundamentally constrained by the limited spectral resolution of standard RGB cameras. In this paper, we propose a new approach that leverages the multi-camera systems integrated into most modern smartphones and enhances their capabilities by modulating their spectral response functions using carefully selected spectral filters, enabling a smartphone to capture richer spectral information. We propose a neural network able to exploit auxiliary RGB images with diverse spectral characteristics, enabling more accurate hyperspectral reconstruction compared to conventional single-image methods. Training is supported by a newly collected dataset, Doomer, which consists of multiple misaligned RGB images alongside corresponding hyperspectral data. We also propose a device-adaptation strategy that conditions the multi-image hyperspectral reconstruction network on camera color correction matrices available in standard DNG metadata, requiring no real images from the target device. We evaluate it on three physical smartphones, two of them unseen during training.
Double Q-learning is a classical control algorithm that mitigates the maximization bias of Q-learning. To do so, it explicitly trains two independent action-value functions and uses them to decouple action-selection from action-evaluation when computing bootstrap targets. Double DQN adapts this decoupling to deep reinforcement learning (RL) but directly trains only a single action-value function and does not fully decouple its estimators. Consequently, the two estimators remain correlated, and overestimation persists. In this paper, we introduce Deep Double Q-learning (DDQL), a deep RL algorithm that explicitly trains two Q-functions through the Double Q-learning objective. DDQL stabilizes training through a combination of techniques, including lower replay ratios, longer target network update intervals, and shared layers. Across 57 Atari 2600 games, DDQL improves aggregate performance over Double DQN, outperforming it on 47 games while further reducing overestimation. We study key design choices of DDQL, including its network architecture, replay ratio, and minibatch sampling strategies.
As large language model (LLM) agents are increasingly deployed in long-horizon unattended scenarios, security evaluation of them before deployment becomes increasingly essential and challenging. Existing evaluation methods mainly follow the paradigms of LLM-as-a-Judge or Agent-as-a-Judge. However, such judges are themselves often vulnerable to hallucination and can even be affected by perturbations. Furthermore, because capturing full environment states at every step is often prohibitively expensive in open-ended environments, existing evaluations typically rely on textual logs and the final environment state. As a result, they may miss transient violations that are overwritten before the trajectory ends or never reflected in textual logs. We present RAS-Eval, a specification-based security evaluation framework for LLM agents. RAS-Eval uniformly expresses utility and security objectives as Linear Temporal Logic (LTL) specifications over tool-call trajectories and dynamic environment observations, and verifies them with an off-the-shelf model checker. This design addresses both limitations above. First, unlike LLM-as-a-Judge and Agent-as-a-Judge, RAS-Eval derives deterministic verdicts from formally defined properties, eliminating evaluator-side uncertainty and remaining reliable even for long-horizon trajectories. Second, RAS-Eval observes exactly the states on which the verdict depends, making it affordable to monitor the environment throughout execution and detect transient violations that leave little trace in the logs or final environment state. Moreover, we design a multi-agent-based dataset construction pipeline. Using it, we construct a benchmark (https://github.com/lanzer-tree/RAS-Eval) of 200 normal tasks and 800 attack tasks spanning four attack surfaces: indirect prompt injection, tool poisoning, skill code injection, and tool code injection.
This research focuses on optimizing multi-UAV systems with dual objectives: maximizing service coverage as the primary goal while extending battery lifetime as the secondary objective. We propose a Graph Attention-based Decentralized Actor-Critic (GADC) to optimize the dual objectives. The proposed approach leverages a graph attention network to process UAVs' limited local observation and reduce the dimension of the environment states. Subsequently, an actor-double-critic network is developed to manage dual policies for joint objective optimization. The proposed GADC uses a Kullback-Leibler (KL) divergence factor to balance the tradeoff between coverage performance and battery lifetime in the multi-UAV system. We assess the scalability and efficiency of GADC through comprehensive benchmarking against state-of-the-art methods, considering both theory and experimental aspects. Extensive testing in both ideal settings and NVIDIA Sionna's realistic ray tracing environment demonstrates GADC's superior performance.
Classical Bundle Adjustment (BA) relies on precise metric initialization and known camera intrinsics, and traditional Structure-from-Motion (SfM) pipelines struggle to exploit modern dense matchers because rigid track-building heuristics break down under their inherent noise. We present Track-Free Dense Structure from Motion (TFD-SfM), which jointly optimizes camera extrinsics, focal lengths, and geometry from a strict cold start. Our central idea is to relax the rigidities of classical SfM into soft, probabilistic decisions under Negative Log-Likelihood (NLL) objectives. Instead of forcing matched pixels onto a single shared 3D point, we model each match as a 3D Gaussian whose variance is optimized jointly with the geometry. Unreliable matches take on large variance and contribute little to the loss. This smooths the optimization landscape, so the solver converges even from identity cameras and uniform depths. A kinematic relative-pose parameterization stabilizes the cold start, and a dual-hypothesis mechanism resolves the mirror-symmetry ambiguity on object-centric scenes. Across DTU, Mip-NeRF~360, ETH3D, and IMC2021, TFD-SfM is competitive with both classical and learning-based SfM baselines and leads on ETH3D, recovering accurate camera poses, intrinsics, and dense geometry from unordered, uncalibrated image collections. Project~page: https://cphjsn.github.io/tfd-sfm/
The muon optimizer has picked up much attention as of late as a possible replacement to the seemingly omnipresent Adam optimizer. Recently, care has been taken to document the scaling laws of hyper-parameters under muon such as weight decay and learning rate. However, at much larger scales the iterative orthogonalization procedure present in muon may suffer a possible issue as the singular values of random matrices shrink with scale. This paper shows this scaling behavior theoretically and empirically on random matrices but does not suggest what to do about it.
Effective teaching involves mutual adaptation: teachers adjust instruction based on student responses, while students actively seek information to resolve uncertainty. Yet many computational models of teaching treat instruction as a one-way process. This paper presents a Bayesian computational framework for studying teacher-student co-adaptation under partial observability in a stylized concept-learning task. Students are organized in heterogeneous groups that differ in which features of a target concept they can observe, and the teacher infers these constraints from limited feedback. We compare five interaction modes (active teaching, active learning, turn-based interaction, adaptive teaching and full co-adaptation) and three teacher-led controls across 100 independent matched simulation runs. Learner-initiated queries resolve information gaps left by the baseline teachers' stopping rules, while teacher-led continuation also achieves complete identification in the observed episodes. Bayesian belief tracking reduces the mean interaction steps required by full co-adaptation as teacher experience accumulates. At experience level 40, full co-adaptation requires fewer interaction steps on average than active learning, turn-based interaction and all three teacher-led controls, with complete observed coverage shared by these policies. Analyses by group and student type, together with sensitivity analyses, clarify the conditions and distributional limits of these results. The framework is interpretable and modular. Its findings provide a computational proof of concept for investigating interaction structure in heterogeneous groups, rather than direct evidence about classroom practice.
Mixture-of-Experts (MoE) architectures are rapidly becoming the standard for building scalable and efficient large language models (LLMs), thanks to their routing mechanism that activates only a subset of experts for each input. However, the modular structure and sparse activation of experts also introduce new security risks, as manipulating a small portion of experts may influence the behavior. In this work, we study a white-box, task-specific model-supply-chain setting and propose BadMoE, a routing-aware backdoor attack against MoE LLMs. We motivate BadMoE through a routing-aware decomposition of the MoE-layer output, which separates the contributions of the selected target experts from those of the remaining experts. This analysis highlights two measurable factors underlying the attack: the routing mass assigned to the target experts and their relative contribution to the layer output. Guided by this insight, BadMoE selects task-underutilized experts and utilizes routing-aware triggers to activate them, enabling reliable activation of task-specific, predefined backdoor behaviors. Specifically, BadMoE involves three steps: 1) identifying experts with low usage on a target-task calibration set, 2) optimizing a routing-aware trigger toward these experts, and 3) associating the selected-expert activations with attacker-specified behavior. We evaluate BadMoE on Mixtral, OLMoE, and Deepseek across sentiment, topic, and emotion classification, sentiment steering, and targeted refusal.
Data annotation is an essential but costly aspect of developing machine learning-based technology, requiring multiple human workers to repeatedly label each data point in a data set. We explore the tradeoff between annotation accuracy and cost by identifying minority reports---worker responses disagreeing with the reference majority-vote label---that may indicate unnecessary redundancy, and can therefore be proactively pruned before they are executed. We develop a theoretical framework for choosing the number of repeated annotations required for a task and pruning policies at a target label quality. We estimate minority-report risk on two computer vision datasets re-annotated through a professional platform, then evaluate a rolling policy on the assignment stream to prune over 35% of repeats while preserving more than 99.9% of reference majority-vote labels.
Causal effect identification using causal graphs is a fundamental challenge in causal inference. While extensive research has been conducted in this area, most existing methods assume the availability of fully specified directed acyclic graphs or acyclic directed mixed graphs. However, in complex domains such as medicine and epidemiology, complete causal knowledge is often unavailable, and only partial information about the system is accessible. This paper focuses on causal effect identification within partially specified causal graphs, with particular emphasis on cluster-directed mixed graphs (C-DMGs) which can represent many different acyclic directed mixed graphs (ADMGs). These graphs provide a higher-level representation of causal relationships by grouping variables into clusters, offering a more practical approach for handling complex systems. Unlike fully specified ADMGs, C-DMGs can contain cycles, which complicate their analysis and interpretation. Furthermore, their cluster-based nature introduces new challenges, as it gives rise to two distinct types of causal effects: macro causal effects and micro causal effects, each with different properties. In this work, we focus on macro causal effects, which describe the effects of entire clusters on other clusters. We establish that the do-calculus is both sound and complete for identifying these effects in C-DMGs over ADMGs when the cluster sizes are either unknown or of size greater than one. Additionally, we provide a graphical characterization of non-identifiability for macro causal effects in these graphs.
Deep learning's growing prevalence has driven its widespread use in healthcare, where AI and sensor advancements enhance diagnosis, treatment, and monitoring. In mobile health, AI-powered tools enable early diagnosis and continuous monitoring of conditions like stress. Wearable technologies and multimodal physiological data have made stress detection increasingly viable, but model efficacy depends on data quality, quantity, and modality. This study develops transformer models for stress detection using the WESAD dataset, training on electrocardiograms (ECG), electrodermal activity (EDA), electromyography (EMG), respiration rate (RESP), temperature (TEMP), and 3-axis accelerometer (ACC) signals. The results demonstrate the effectiveness of single-modality transformers in analyzing physiological signals, achieving state-of-the-art performance with accuracy, precision and recall values in the range of $99.73\%$ to $99.95\%$ for stress detection. Furthermore, this study explores cross-modal performance and also explains the same using 2D visualization of the learned embedding space and quantitative analysis based on data variance. Despite the large body of work on stress detection and monitoring, the robustness and generalization of these models across different modalities has not been explored. This research represents one of the initial efforts to interpret embedding spaces for stress detection, providing valuable information on cross-modal performance.
We study the fundamental problem of learning with respect to the squared loss in a convex class. In the subgaussian setting, the sharpest known sample complexity bounds are characterised via fixed points associated with the limiting gaussian process. In more general settings (for example, heavier-tailed scenarios) the state-of-the-art bounds rely on localised Rademacher complexities, which are often difficult to control and can be considerably larger than their limiting gaussian counterparts. We show that if one has access to coarse information on the covariance structure of the underlying class, it is possible to derive an upper bound on the sample complexity that matches the state-of-the-art rates in the gaussian setting---even in heavy-tailed regimes. As a consequence, under rather minimal assumptions, learning problems that share the same $L_2$-structure---even those involving heavy-tailed distributions---admit the same upper bound on the sample complexity as if the problem were fully gaussian. This constitutes the first universality result for general convex learning problems. The proof is based on a novel learning procedure, and its performance is studied by combining optimal mean estimation techniques for real-valued random variables with Talagrand's generic chaining method.
Natural language understanding (NLU) is a task that enables machines to understand human language. Some tasks, such as stance detection and sentiment analysis, are closely related to individual subjective perspectives, thus termed individual-level NLU. Previously, these tasks are often simplified to text-level NLU tasks, ignoring individual factors. This not only makes inference difficult and unexplainable but often results in a large number of label errors when creating datasets. To address the above limitations, we propose a new NLU annotation guideline based on individual-level factors. Specifically, we incorporate other posts by the same individual and then annotate individual subjective perspectives after considering all individual posts. We use this guideline to expand and re-annotate the stance detection and topic-based sentiment analysis datasets. We find that error rates in the samples were as high as 31.7\% and 23.3\%. We further use large language models to conduct experiments on the re-annotation datasets and find that the large language models perform well on both datasets after adding individual factors. Both GPT-4o and Llama3-70B can achieve an accuracy greater than 87\% on the re-annotation datasets. We also verify the effectiveness of individual factors through ablation studies. We call on future researchers to add individual factors when creating such datasets. Our re-annotation dataset can be found at https://github.com/24yearsoldstudent/Individual-NLU