Science & Mathematics

Robust partially observable Markov decision processes model agents that act under partial observability while adversarial environment resolves uncertainty in the transition probabilities. Reward objectives such as discounted-sum, long-run average and cumulative reward under finite and infinite horizons allow quantifying the utility of an agent. We consider the problem of solving RPOMDPs under all four reward objectives, where the goal is to decide whether the value of a risk measure with respect to a reward objective exceeds a given threshold. We consider different risk measures, such as (1)~expected value, (2)~mean-variance (MV), (3)~conditional value-at-risk (CVaR), and (4)~entropic risk (ERisk). Our results can be summarized as follows. First, we formalize the class of law-invariant risk measures, that subsume all four risk measures above. Second, we propose a unified framework for solving RPOMDPs with discounted-sum, long-run average and cumulative reward objectives under any law-invariant risk measure, by reducing the problem to solving partially observable stochastic games (POSGs). Our unified framework applies to (s,a)-rectangular RPOMDPs with polytopic uncertainty sets, which we show to be {\em equivalent} to POSGs, for each of the four reward objectives and any law-invariant risk measure. We prove equivalence by designing logspace reductions between the RPOMDP and POSG models, that map strategies to strategies and preserve the distribution of the reward objective, up to a constant rescaling factor. Third, as a corollary, we derive new computational complexity results for R(PO)MDPs and POSGs. In particular, we show that solving (s,a)-rectangular RMDPs is as hard as solving simple stochastic games.

The goal of this paper is to distill insight around the trade-offs present on Bayesian equilibria of transaction fee mechanism (TFM) design under $n$ multiple concurrent proposers (MCP). We conceptualize a technical toolkit that unlocks the analysis of hard-to-characterize equilibria of finite Bayesian MCP TFM games by examining sequences of joint user--proposer mixed equilibria as the number of users increases and keeping the relative congestion $ρ$ that they cause fixed in the sequence limit. We first consider the popular mechanisms of "pay-once" (which charges transactions its bid only once regardless of the duplicate copies that end up being included by proposers) and "pay-per-inclusion" (PPI in short, which charges a transaction for each copy that appears). We prove that PPI is efficient in the limit of equilibria sequences, and incentivizes users to route to only one proposer. We characterize $m$ users' asymptotic behavior in PPI even in finite markets varying with $O(\sqrt{m})$, where we show users in a $Θ(n\sqrt{ρ(1-ρ)/m})$ window near an efficient cutoff exhibit fixed inclusion probability akin to a normal distribution's CDF. We additionally show that on a value-dependent basis, pay-once exhibits a higher degree of duplication and lower probability of inclusion for higher-value transactions than PPI. Pay-once's inefficiency persists even in the large market limit and even in regimes with ample blockspace where all transactions fit across all blocks. We study interpolations between these two mechanisms proving that some higher coefficients (towards PPI) reduce duplication and increase inclusion probability. We then use our tools to show that exclusive proposer payments yield inefficient equilibria (in fact, with maximal possible duplication), even if users still pay multiple times when included more than once and their payment is burnt when non-exclusive.

In high-performance computing (HPC), users must submit resource requirements for their jobs, with user-supplied walltime being a critical parameter. However, these estimates are notoriously inaccurate, as users tend to overestimate walltime to prevent the scheduler from terminating their jobs prematurely. While existing work on learning-based walltime prediction primarily focuses on maximizing prediction accuracy, such predictions are inherently risky for walltime recommendations. Specifically, underestimation leads to job termination before completion, resulting in lost progress and wasted resources. In this study, we present SafeWall, a novel recommendation method that adjusts walltime predictions by incorporating an explicit, nonnegative safety buffer to mitigate underestimation risk. The core of our approach is a risk-aware buffer that dynamically calibrates safety margins based on individual user's historical behavior. When user data is limited, the system employs statistical shrinkage to align these margins with a global baseline, ensuring robustness. This yields a lightweight mechanism that optimizes the trade-off between prediction accuracy and termination risk. SafeWall is modular and applicable to various learning-based prediction models. We evaluate SafeWall using five public workload traces, including the full-lifetime job log from the 4,392-node Theta machine at the Argonne Leadership Computing Facility. Across all traces, SafeWall substantially reduces underestimation risk compared to the unbuffered base predictors, while preserving competitive accuracy. The source code is publicly available at https://github.com/SPEAR-UIC/JSSPP26-SafeWall.

Recent time series foundation models (TSFMs) can consume covariates in-context, extending zero-shot forecasting to problems driven by external factors. This enables scenario analysis: forecasting how a target responds to events that set covariates to chosen future values. Whether they can do this reliably is an open question. We introduce WhatIf-Bench, a benchmark measuring how well forecasters use covariates to forecast a target's response to future events. WhatIf-Bench is semi-synthetic: time series come from mechanistic simulators of data centers, power grids, road traffic, and water networks, with controlled covariate-target relationships grounded in each system. Each task presents a history of events intervening on covariates, then asks to forecast a future event that either replays seen values (Replay) or sets them to unseen ones (Novel). Varying the number of covariates, historical events, and event length yields 37,665 tasks. Evaluating TSFMs, frontier LLMs, and simple regression baselines, we find the LLMs reach the highest skill (improvement over a naive baseline), ahead of the best TSFMs. Every model forecasts well on Replay futures (48 to 85% skill) but drops to -7 to 50% on Novel futures, with the regression baselines falling the most. Skill on Novel futures also falls as events set more covariates jointly. More historical events raise skill on Novel futures across the board, but large differences remain. WhatIf-Bench reveals generalization to Novel futures as the core challenge, and offers a testbed toward arbitrary scenario analysis.

As dexterous robotic manipulation advances, scalable annotations of human hand-object interactions are becoming increasingly important. Furthermore, previous hand contact estimation methods suffered from issues such as dorsal-side misclassifications and insufficient precision, while lacking a unified framework for integrating multiple hand to environment interaction modalities, including distance fields, contact thresholds, and pressure. To address that, we present DenseHap, a unified end-to-end approach to estimate hand distance field recognition, comprehensive hand contact detection, hand pressure analysis, and hand reconstruction from a single monocular image. By incorporating three distinctly defined heads, along with a pretrained 3D reconstructed skeleton and an enhanced dataset, DenseHap outperforms previous state of the art models on four evaluation datasets, improving the F1 score by 8.5 percentage points (a 16% relative improvement) on untrained datasets and by 3 percentage points (5% relative improvement) on in-distribution datasets. Furthermore, DenseHap has effectively improved short-range hand estimate and pressure evaluation. Code and models will be publicly available upon the paper's acceptance.

A poker agent can make a legal move and still make a strategically poor one. We ask whether compact AI agents can exhibit strategic behavior under hidden information and whether preference optimization can improve beyond static and zero-shot policies. We first evaluate a static baseline and three frozen language models: SmallThinker, Qwen3, and Gemma-3 whose performance varies substantially between exact-game and six-player evaluation. We therefore propose Strategic Preference Optimization with KL-anchored Equilibrium (SPOKE), combining teacher-relative value penalties, policy anchoring, risk constraints, and structured rationale supervision, and compare it with SFT, DPO, GRPO, and GSPO. Across three runs of a SmallThinker-4B-A0.6B policy, SPOKE achieves the lowest mean exploitability in Kuhn (0.213) and Leduc (1.081) and the highest six-max top-1 agreement (0.775), while maintaining teacher KL comparable to SFT (0.522 vs 0.523). These results provide preliminary evidence that strategic preference optimization can improve strategic quality while preserving teacher-consistent behavior.

Humanoid robots hold great promise as general-purpose agents in human-centered environments, yet adapting generalist vision-language-action (VLA) policies to real-world humanoid loco-manipulation remains challenging. Unlike fixed-base or wheeled manipulators, humanoids must coordinate locomotion, posture, gaze, and dual-arm manipulation over a high-dimensional whole-body action space, where inaccurate base or waist motions can easily induce unstable and aggressive upper-body compensation. We propose HAF-VLA, a hierarchical action-flow framework for humanoid VLA control. Built upon a pretrained flow-matching VLA, HAF-VLA inherits the vision-language prefix encoder and action expert weights, and adapts them to humanoid whole-body loco-manipulation with learnable stage embeddings and cross-stage KV-cache conditioning. Instead of generating the full action chunk in a single pass, HAF-VLA progressively denoises a shared whole-body action space: first locomotion and head commands, then waist refinement, and finally full-body manipulation. To preserve structural dependencies across stages, we introduce a cross-stage KV cache conditioning mechanism, enabling subsequent action generation stages to refine and augment their outputs based on the action representations from preceding stages. Experimental results across six real-world tasks demonstrate that our framework significantly enhances the performance of existing models in humanoid loco-manipulation tasks, demonstrating an effective pathway for adapting strong mobile-manipulation VLAs to humanoid whole-body loco-manipulation.

Generating reusable driving environments requires both controllable future observations and an explicit representation that supports rendering from different viewpoints, yet existing methods still recover explicit scenes from generated RGB videos through optimization for each sequence. To overcome these limitations, we present Drive2Gauss, a feedforward framework that converts jointly generated RGB, metric depth, and dynamic flow into dynamic 4D Gaussian scenes. A Multimodal Video Generator predicts the three modalities from reference multiview images and future control signals. Furthermore, we introduce a Gaussian Query Decoder that uses depth to initialize sparse 3D queries and flow to associate the queries across time, and integrates appearance information through spatiotemporal feature aggregation and iterative query refinement. The reconstruction stage operates in a single feedforward pass without optimization for each scene. Extensive results on nuScenes demonstrate that our approach attains an FID of 6.38 and an FVD of 12.43 for future RGB generation, a PSNR of 29.91 dB when reconstructing its generated observations, and supports both direct rendering under lateral camera shifts and downstream video fixing. Code and more visualizations are available at https://zttt04. github.io/Drive2Gauss_page/.

Here, we present TOX-FM, a family of non-clinical PFMs trained on large-scale diverse histopathology datasets of over one million whole slide images consisting of over 60 organs across six animal species from over 14,000 toxicology studies across six decades. We introduce a comprehensive benchmark of 18 non-clinical downstream tasks, incorporate an established benchmark of 9 clinical downstream tasks, and evaluate TOX-FM against state-of-the-art clinical PFMs across both non-clinical and clinical domains. Notably, we explore open questions surrounding domain-specific training and cross-domain generalization, offering insights into the opportunities and challenges of applying PFMs in the non-clinical domain.

The feature-extraction quality of pre-trained models sets the ceiling of downstream transfer learning, yet few-shot model selection remains difficult: supervised evaluation on scarce labels is unstable and yields fluctuating rankings. We propose Entropy-Guided Model Evaluation (EGiME), a semi-supervised criterion that assesses a candidate model from the clustering structure of its feature space without any classifier training. EGiME performs $K$-means clustering over all labeled and unlabeled samples, then combines two entropy terms: an inter-cluster entropy measuring global separability, and an intra-cluster entropy measuring the compactness of labeled anchors, fused through an inverse-frequency weight that emphasizes small clusters. On five few-shot benchmarks and nine pre-trained models, EGiME achieves a correlation score of 0.966 with the full-pool ground-truth accuracy obtained by averaging the accuracies of K-means, KNN, and SVM. At labeling ratios of 4\%--20\%, it produces markedly more stable model rankings than cross-validated supervised baselines, establishing it as a practical alternative to conventional supervised evaluation for pre-trained model selection under data scarcity.

Attack success rate (ASR) evaluates jailbreaks against large language models without accounting for repeated calls to the victim model. Yet each additional call gives the deployed system another opportunity to detect the attack and intervene. We introduce $\mathrm{ASR}_{\mathrm{STEP}}$, which penalizes additional victim-model calls while retaining binary success judgments. Our evaluation considers GCG, PAIR with a step budget, and a single-step PAP baseline on JailBreakBench across three model families. GCG's scores decrease by 7.0-8.7 percentage points after discounting, while PAP's remain unchanged. While for PAIR attack, we see a wider discounting range between 3.4-12 percentage points, showing that the "step" quantity differs per iterative strategy. Reporting both metrics makes the discount due to repeated victim-model interaction explicit. Code is available here: https://github.com/SuELS-Group/beyond_asr.

A limited understanding of how evaporation couples to crystallization pathways hampers control over the formation of solution-processed perovskite photoactive layers. Identifying these mechanisms-and learning how to select and tune them-is essential for achieving the desired crystal morphology and, ultimately, improving solar-cell performance and stability. To investigate crystallization pathways in evaporating thin films, we develop a simple and general zero-dimensional population-balance model that couples evaporation, chemical reactions, and crystallization. We use the model to explore a range of parameters for a solvent-solute system that follows a two-step crystallization pathway: solvent evaporation first induces formation of a crystalline precursor, which subsequently recrystallizes into the final phase. The model predicts the evolution of the film composition during drying and provides quantitative LaMer curves. Based on these results, we introduce a vocabulary of "supply limitations" and "rate limitations" that enables the interpretation of crystallization kinetics in terms of the governing process conditions.

Jailbreak methods are commonly compared using Attack Success Rate (ASR), but these comparisons depend on how responses are generated and judged. In this paper, we examine eight attacks against three open-weight language models using three evaluation procedures. Low-resource language attacks achieve a mean ASR of 96.0% under keyword matching but 2.7% under a Qwen3 evaluator following StrongREJECT scoring instructions. Changing the number and arrangement of attempts reverses the ranking of DeepInception and rewriting. In a separate experiment, sampling 100 responses for each unchanged request at Ollama's default temperature of 0.8 yields 100% ASR under keyword matching but 50% under the Qwen3 evaluator. Beyond evaluator sensitivity, these findings raise a safety concern: repeated sampling can expose residual unsafe behavior despite an initial refusal, suggesting probabilistic suppression rather than robust exclusion. We recommend explicit success definitions, matched response budgets, validated evaluators, and reporting cumulative failure under repeated interaction rather than relying on single-generation tests.

Living plants, like animals, use electrical signals to carry information about their physiological state across their body, from roots to branches. These signals can be recorded, as they are in animals, from a pair of differential extracellular electrodes inserted into signal-conducting tissue. While machine-learning models have been trained to predict specific plant stress states from these signals, each of these models has been trained for only a single task and none have been designed to transfer across tasks, species, or environments. These approaches do not allow easy scaling across the wide range of species and environments that are relevant to agriculture, forestry, and horticulture for both commercial and consumer plant care. Here we present Bombadil, a foundation model for plant electrophysiology, pretrained without labels on over 4 million hours of electrical signals recorded from 100 plant species in ten distinct environments, including laboratories, greenhouses, commercial orchards, and natural forests. In this way, Bombadil is able to learn the fundamental features of plant electrical signals across a wide array of species and environments. Then, from those same electrical signals, Bombadil tracks changes in seven physiological indicators of plant stress, even among species and stressors with no labels in training.

Current Virtual Reality systems typically use cameras to capture user input from controllers or free-hand mid-air interaction. In this paper, we argue that this is a key impediment to productivity scenarios in VR, which require continued interaction over prolonged periods of time---a requirement that controller or free-hand input in mid-air does not satisfy. To address this challenge, we bring rapid touch interaction on surfaces to Virtual Reality---the input modality that users have grown used to on phones and tablets for continued use. We present TapID, a wrist-based inertial sensing system that complements headset-tracked hand poses to trigger input in VR. TapID embeds a pair of inertial sensors in a flexible strap, one at either side of the wrist; from the combination of registered signals, TapID reliably detects surface touch events and, more importantly, identifies the finger used for touch. We evaluated TapID in a series of userstudies on event-detection accuracy (F1 = 0.997) and hand-agnostic finger-identification accuracy (within-user: F1 = 0.93; across users: F1 = 0.91 after 10 refinement taps and F1 = 0.87 without refinement) in a seated table scenario. We conclude with a series of applications that complement hand tracking with touch input and that are uniquely enabled by TapID, including UI control, rapid keyboard typing and piano playing, as well as surface gestures.