Anqi Liu

Jiayu Zheng, Lingxin Hao, Kelun Lu et al.

This study explores how college students interact with generative AI (ChatGPT-4) during educational quizzes, focusing on reliance and predictors of AI adoption. Conducted at the early stages of ChatGPT implementation, when students had limited familiarity with the tool, this field study analyzed 315 student-AI conversations during a brief, quiz-based scenario across various STEM courses. A novel four-stage reliance taxonomy was introduced to capture students' reliance patterns, distinguishing AI competence, relevance, adoption, and students' final answer correctness. Three findings emerged. First, students exhibited overall low reliance on AI and many of them could not effectively use AI for learning. Second, negative reliance patterns often persisted across interactions, highlighting students' difficulty in effectively shifting strategies after unsuccessful initial experiences. Third, certain behavioral metrics strongly predicted AI reliance, highlighting potential behavioral mechanisms to explain AI adoption. The study's findings underline critical implications for ethical AI integration in education and the broader field. It emphasizes the need for enhanced onboarding processes to improve student's familiarity and effective use of AI tools. Furthermore, AI interfaces should be designed with reliance-calibration mechanisms to enhance appropriate reliance. Ultimately, this research advances understanding of AI reliance dynamics, providing foundational insights for ethically sound and cognitively enriching AI practices.

Lei Huang, Hui Shen, Kuan-Jui Su et al.

Genotype-based cis-expression prediction depends on accurately modeling local regulatory architecture. We present block-sparse Bayesian sparse linear mixed model (bsBSLMM), an extension of Bayesian sparse linear mixed model (BSLMM) that incorporates linkage disequilibrium (LD)-block spike-and-slab sparsity and a transcription start site (TSS)-informed SNP inclusion prior. Across 23,098 genes from GEUVADIS European-ancestry lymphoblastoid cell lines, bsBSLMM retained more predictable genes than BSLMM, LASSO, BLUP, TIGAR elastic net, and TIGAR Dirichlet-process regression under matched evaluation criteria. Compared with BSLMM, bsBSLMM improved held-out prediction performance for most shared genes, with gains driven primarily by LD-block sparsity and further enhanced by the TSS-informed prior. Variants selected by bsBSLMM showed stronger enrichment in GM12878 DNase and H3K27ac regulatory regions than variants selected by BSLMM. In transcriptome-wide association study (TWAS) analysis, bsBSLMM recovered established inflammatory bowel disease signals, including IL23R, and identified additional genome-wide significant genes not detected by BSLMM. Independent validation in the Louisiana Osteoporosis Study reproduced the increased prediction yield across ancestries and recovered biologically relevant bone mineral density pathways in downstream TWAS and gene set enrichment analyses. These results demonstrate that incorporating LD-block structure and biologically informed SNP priors improves cis-expression prediction and enhances downstream TWAS discovery.

Zhengping Jiang, Mehran Khodabandeh, Akash Bharadwaj et al.

Aligning large language models (LLMs) to heterogeneous and rapidly evolving safety requirements remains a critical challenge. Existing instruction-tuned LLMs and standalone safety classifiers often fail to generalize to new safety configurations, motivating the need for Reward Models (RMs) that are explicitly configurable to changing specifications. We introduce the Configurable Safety Reward Model (CSRM), which is jointly optimized for calibrated safety compliance and reward modeling. Our approach is supported by configuration-targeted data augmentation that enforces instruction adherence while preserving relative severity structure. The resulting RM is sensitive to fine-grained safety configurations and conversational nuances, substantially improving generalization to previously unseen safety configurations. CSRM achieves state-of-the-art performance on recent configurable safety benchmarks, including CoSApien (94.6% F1) and DynaBench (75.8% F1), without requiring additional human annotation. When used for downstream safety alignment, CSRM yields LLMs with a significantly improved helpfulness-safety tradeoff compared to existing baselines.

Ha Manh Bui, Metod Jazbec, Eric Nalisnick et al.

Offline-to-Online Reinforcement Learning (O2O-RL) leverages an offline, pre-trained policy to minimize costly online interactions. Although data-efficient, O2O-RL is susceptible to shifts between offline and online distributions. Existing work aims to mitigate the harm of this shift by finetuning the policy on trajectory data sampled from a diffusion model. Inspired by this line of work, we propose DUAL: an efficient \textbf{D}iffusion \textbf{U}ncertainty-\textbf{A}ware framework for offline-to-online reinforcement \textbf{L}earning. DUAL utilizes the prior knowledge of the diffusion model to distill a fast-sampling diffusion actor policy and transition model in the offline phase. DUAL also employs a Laplace approximation and distance transition-state-shift detection, thereby using uncertainty quantification to improve exploration versus exploitation in the online phase. We formally show that our actor loss with the Laplace approximation provides a proxy for a principled estimate of epistemic uncertainty. Empirically, DUAL improves the online expected return over O2O-RL baselines across multiple settings and environments.

Sara Kangaslahti, Danny Ebanks, Jean Kossaifi et al.

This paper proposes a topic modeling method that scales linearly to billions of documents. We make three core contributions: i) we present a topic modeling method, Tensor Latent Dirichlet Allocation (TLDA), that has identifiable and recoverable parameter guarantees and sample complexity guarantees for large data; ii) we show that this method is computationally and memory efficient (achieving speeds over 3-4x those of prior parallelized Latent Dirichlet Allocation (LDA) methods), and that it scales linearly to text datasets with over a billion documents; iii) we provide an open-source, GPU-based implementation, of this method. This scaling enables previously prohibitive analyses, and we perform two real-world, large-scale new studies of interest to political scientists: we provide the first thorough analysis of the evolution of the #MeToo movement through the lens of over two years of Twitter conversation and a detailed study of social media conversations about election fraud in the 2020 presidential election. Thus this method provides social scientists with the ability to study very large corpora at scale and to answer important theoretically-relevant questions about salient issues in near real-time.

Xuewei Cao, Joyce H Keyak, Sigurdur Sigurdsson et al.

Hip fracture risk assessment is an important but challenging task. Quantitative CT-based patient specific finite element analysis (FEA) computes the force (fracture load) to break the proximal femur in a particular loading condition. It provides different structural information about the proximal femur that can influence a subject overall fracture risk. To obtain a more robust measure of fracture risk, we used principal component analysis (PCA) to develop a global FEA computed fracture risk index that incorporates the FEA-computed yield and ultimate failure loads and energies to failure in four loading conditions (single-limb stance and impact from a fall onto the posterior, posterolateral, and lateral aspects of the greater trochanter) of 110 hip fracture subjects and 235 age and sex matched control subjects from the AGES-Reykjavik study. We found that the first PC (PC1) of the FE parameters was the only significant predictor of hip fracture. Using a logistic regression model, we determined if prediction performance for hip fracture using PC1 differed from that using FE parameters combined by stratified random resampling with respect to hip fracture status. The results showed that the average of the area under the receive operating characteristic curve (AUC) using PC1 was always higher than that using all FE parameters combined in the male subjects. The AUC of PC1 and AUC of the FE parameters combined were not significantly different than that in the female subjects or in all subjects

Kate Sanders, Reno Kriz, Anqi Liu et al.

Contemporary vision benchmarks predominantly consider tasks on which humans can achieve near-perfect performance. However, humans are frequently presented with visual data that they cannot classify with 100% certainty, and models trained on standard vision benchmarks achieve low performance when evaluated on this data. To address this issue, we introduce a procedure for creating datasets of ambiguous images and use it to produce SQUID-E ("Squidy"), a collection of noisy images extracted from videos. All images are annotated with ground truth values and a test set is annotated with human uncertainty judgments. We use this dataset to characterize human uncertainty in vision tasks and evaluate existing visual event classification models. Experimental results suggest that existing vision models are not sufficiently equipped to provide meaningful outputs for ambiguous images and that datasets of this nature can be used to assess and improve such models through model training and direct evaluation of model calibration. These findings motivate large-scale ambiguous dataset creation and further research focusing on noisy visual data.

Samuel Barham, Orion Weller, Michelle Yuan et al.

To foster the development of new models for collaborative AI-assisted report generation, we introduce MegaWika, consisting of 13 million Wikipedia articles in 50 diverse languages, along with their 71 million referenced source materials. We process this dataset for a myriad of applications, going beyond the initial Wikipedia citation extraction and web scraping of content, including translating non-English articles for cross-lingual applications and providing FrameNet parses for automated semantic analysis. MegaWika is the largest resource for sentence-level report generation and the only report generation dataset that is multilingual. We manually analyze the quality of this resource through a semantically stratified sample. Finally, we provide baseline results and trained models for crucial steps in automated report generation: cross-lingual question answering and citation retrieval.

Drew Prinster, Anqi Liu, Suchi Saria

We propose \textbf{JAWS}, a series of wrapper methods for distribution-free uncertainty quantification tasks under covariate shift, centered on the core method \textbf{JAW}, the \textbf{JA}ckknife+ \textbf{W}eighted with data-dependent likelihood-ratio weights. JAWS also includes computationally efficient \textbf{A}pproximations of JAW using higher-order influence functions: \textbf{JAWA}. Theoretically, we show that JAW relaxes the jackknife+'s assumption of data exchangeability to achieve the same finite-sample coverage guarantee even under covariate shift. JAWA further approaches the JAW guarantee in the limit of the sample size or the influence function order under common regularity assumptions. Moreover, we propose a general approach to repurposing predictive interval-generating methods and their guarantees to the reverse task: estimating the probability that a prediction is erroneous, based on user-specified error criteria such as a safe or acceptable tolerance threshold around the true label. We then propose \textbf{JAW-E} and \textbf{JAWA-E} as the repurposed proposed methods for this \textbf{E}rror assessment task. Practically, JAWS outperform state-of-the-art predictive inference baselines in a variety of biased real world data sets for interval-generation and error-assessment predictive uncertainty auditing tasks.

Xuan Shen, Brian Wingenroth, Zichao Wang et al.

The opioid crisis represents a significant moment in public health that reveals systemic shortcomings across regulatory systems, healthcare practices, corporate governance, and public policy. Analyzing how these interconnected systems simultaneously failed to protect public health requires innovative analytic approaches for exploring the vast amounts of data and documents disclosed in the UCSF-JHU Opioid Industry Documents Archive (OIDA). The complexity, multimodal nature, and specialized characteristics of these healthcare-related legal and corporate documents necessitate more advanced methods and models tailored to specific data types and detailed annotations, ensuring the precision and professionalism in the analysis. In this paper, we tackle this challenge by organizing the original dataset according to document attributes and constructing a benchmark with 400k training documents and 10k for testing. From each document, we extract rich multimodal information-including textual content, visual elements, and layout structures-to capture a comprehensive range of features. Using multiple AI models, we then generate a large-scale dataset comprising 360k training QA pairs and 10k testing QA pairs. Building on this foundation, we develop domain-specific multimodal Large Language Models (LLMs) and explore the impact of multimodal inputs on task performance. To further enhance response accuracy, we incorporate historical QA pairs as contextual grounding for answering current queries. Additionally, we incorporate page references within the answers and introduce an importance-based page classifier, further improving the precision and relevance of the information provided. Preliminary results indicate the improvements with our AI assistant in document information extraction and question-answering tasks. The dataset is available at: https://huggingface.co/datasets/opioidarchive/oida-qa

Zhengping Jiang, Jingyu Zhang, Nathaniel Weir et al.

Hallucinations pose a challenge to the application of large language models (LLMs) thereby motivating the development of metrics to evaluate factual precision. We observe that popular metrics using the Decompose-Then-Verify framework, such as \FActScore, can be manipulated by adding obvious or repetitive subclaims to artificially inflate scores. This observation motivates our new customizable plug-and-play subclaim selection component called Core, which filters down individual subclaims according to their uniqueness and informativeness. We show that many popular factual precision metrics augmented by Core are substantially more robust on a wide range of knowledge domains. We release an evaluation framework supporting easy and modular use of Core and various decomposition strategies, which we recommend adoption by the community. We also release an expansion of the FActScore biography dataset to facilitate further studies of decomposition-based factual precision evaluation.

Chen Zhao, Anqi Liu, Xiao Zhang et al.

Integration of heterogeneous and high-dimensional multi-omics data is becoming increasingly important in understanding genetic data. Each omics technique only provides a limited view of the underlying biological process and integrating heterogeneous omics layers simultaneously would lead to a more comprehensive and detailed understanding of diseases and phenotypes. However, one obstacle faced when performing multi-omics data integration is the existence of unpaired multi-omics data due to instrument sensitivity and cost. Studies may fail if certain aspects of the subjects are missing or incomplete. In this paper, we propose a deep learning method for multi-omics integration with incomplete data by Cross-omics Linked unified embedding with Contrastive Learning and Self Attention (CLCLSA). Utilizing complete multi-omics data as supervision, the model employs cross-omics autoencoders to learn the feature representation across different types of biological data. The multi-omics contrastive learning, which is used to maximize the mutual information between different types of omics, is employed before latent feature concatenation. In addition, the feature-level self-attention and omics-level self-attention are employed to dynamically identify the most informative features for multi-omics data integration. Extensive experiments were conducted on four public multi-omics datasets. The experimental results indicated that the proposed CLCLSA outperformed the state-of-the-art approaches for multi-omics data classification using incomplete multi-omics data.

Drew Prinster, Clara Fannjiang, Ji Won Park et al.

An agent must try new behaviors to explore and improve. In high-stakes environments, an agent that violates safety constraints may cause harm and must be taken offline, curtailing any future interaction. Imitating old behavior is safe, but excessive conservatism discourages exploration. How much behavior change is too much? We show how to use any safe reference policy as a probabilistic regulator for any optimized but untested policy. Conformal calibration on data from the safe policy determines how aggressively the new policy can act, while provably enforcing the user's declared risk tolerance. Unlike conservative optimization methods, we do not assume the user has identified the correct model class nor tuned any hyperparameters. Unlike previous conformal methods, our theory provides finite-sample guarantees even for non-monotonic bounded constraint functions. Our experiments on applications ranging from natural language question answering to biomolecular engineering show that safe exploration is not only possible from the first moment of deployment, but can also improve performance.

Gina Wong, Joshua Gleason, Rama Chellappa et al.

Learning models whose predictions are invariant under multiple environments is a promising approach for out-of-distribution generalization. Such models are trained to extract features $X_{\text{inv}}$ where the conditional distribution $Y \mid X_{\text{inv}}$ of the label given the extracted features does not change across environments. Invariant models are also supposed to generalize to shifts in the marginal distribution $p(X_{\text{inv}})$ of the extracted features $X_{\text{inv}}$, a type of shift we call an $\textit{invariant covariate shift}$. However, we show that proposed methods for learning invariant models underperform under invariant covariate shift, either failing to learn invariant models$\unicode{x2014}$even for data generated from simple and well-studied linear-Gaussian models$\unicode{x2014}$or having poor finite-sample performance. To alleviate these problems, we propose $\textit{weighted risk invariance}$ (WRI). Our framework is based on imposing invariance of the loss across environments subject to appropriate reweightings of the training examples. We show that WRI provably learns invariant models, i.e. discards spurious correlations, in linear-Gaussian settings. We propose a practical algorithm to implement WRI by learning the density $p(X_{\text{inv}})$ and the model parameters simultaneously, and we demonstrate empirically that WRI outperforms previous invariant learning methods under invariant covariate shift.

Yu Yang, Yihong Guo, Anqi Liu et al.

Off-dynamics offline reinforcement learning seeks to learn a target-domain policy from a large source dataset and a limited target dataset under mismatched transition dynamics. Existing approaches such as reward augmentation and data filtering are constrained to the source dataset and cannot synthesize new target behavior to improve coverage beyond the collected source trajectories. While recent model-based methods attempt to address this by learning target-aware dynamics, the generated experience is constructed only at the transition level, which leads to accumulated errors over long horizons. These limitations necessitate a shift toward trajectory-level generation for off-dynamics offline RL. We propose CEDGE, a Cross-domain Energy-guided Diffusion GEneration framework. CEDGE trains a trajectory diffusion model on source-domain trajectories and adapts the generated samples to the target domain through energy guidance. This guidance is derived by minimizing the distribution mismatch between the source and desired target-domain trajectories and is decomposed into return, domain, and behavior energy components. The resulting energy-guided trajectories are useful both for direct planning and as synthetic data for policy learning. Since target adaptation is achieved via energy guidance rather than retraining the diffusion model, CEDGE can be efficiently adapted to new target dynamics compared to previous methods. Experiments on the ODRL benchmark demonstrate that trajectory-level energy-guided generation improves diffusion planning under dynamics shifts and produces synthetic data that improves downstream target policy learning.

Chen Zhao, Kuan-Jui Su, Chong Wu et al.

Background: Missing data is a common challenge in mass spectrometry-based metabolomics, which can lead to biased and incomplete analyses. The integration of whole-genome sequencing (WGS) data with metabolomics data has emerged as a promising approach to enhance the accuracy of data imputation in metabolomics studies. Method: In this study, we propose a novel method that leverages the information from WGS data and reference metabolites to impute unknown metabolites. Our approach utilizes a multi-view variational autoencoder to jointly model the burden score, polygenetic risk score (PGS), and linkage disequilibrium (LD) pruned single nucleotide polymorphisms (SNPs) for feature extraction and missing metabolomics data imputation. By learning the latent representations of both omics data, our method can effectively impute missing metabolomics values based on genomic information. Results: We evaluate the performance of our method on empirical metabolomics datasets with missing values and demonstrate its superiority compared to conventional imputation techniques. Using 35 template metabolites derived burden scores, PGS and LD-pruned SNPs, the proposed methods achieved R^2-scores > 0.01 for 71.55% of metabolites. Conclusion: The integration of WGS data in metabolomics imputation not only improves data completeness but also enhances downstream analyses, paving the way for more comprehensive and accurate investigations of metabolic pathways and disease associations. Our findings offer valuable insights into the potential benefits of utilizing WGS data for metabolomics data imputation and underscore the importance of leveraging multi-modal data integration in precision medicine research.

Zhaoning Li, Qiaoli Jiang, Zhengming Wu et al.

Autonomous cars are indispensable when humans go further down the hands-free route. Although existing literature highlights that the acceptance of the autonomous car will increase if it drives in a human-like manner, sparse research offers the naturalistic experience from a passenger's seat perspective to examine the humanness of current autonomous cars. The present study tested whether the AI driver could create a human-like ride experience for passengers based on 69 participants' feedback in a real-road scenario. We designed a ride experience-based version of the non-verbal Turing test for automated driving. Participants rode in autonomous cars (driven by either human or AI drivers) as a passenger and judged whether the driver was human or AI. The AI driver failed to pass our test because passengers detected the AI driver above chance. In contrast, when the human driver drove the car, the passengers' judgement was around chance. We further investigated how human passengers ascribe humanness in our test. Based on Lewin's field theory, we advanced a computational model combining signal detection theory with pre-trained language models to predict passengers' humanness rating behaviour. We employed affective transition between pre-study baseline emotions and corresponding post-stage emotions as the signal strength of our model. Results showed that the passengers' ascription of humanness would increase with the greater affective transition. Our study suggested an important role of affective transition in passengers' ascription of humanness, which might become a future direction for autonomous driving.

Yihong Guo, Junjie Luo, Guodong Gao et al.

Offline contextual bandits allow one to learn policies from historical/offline data without requiring online interaction. However, offline policy optimization that maximizes overall expected rewards can unintentionally amplify the reward disparities across groups. As a result, some groups might benefit more than others from the learned policy, raising concerns about fairness, especially when the resources are limited. In this paper, we study a group-sensitive fairness constraint in offline contextual bandits, reducing group-wise reward disparities that may arise during policy learning. We tackle the following common-parity requirements: the reward disparity is constrained within some user-defined threshold or the reward disparity should be minimized during policy optimization. We propose a constrained offline policy optimization framework by introducing group-wise reward disparity constraints into an off-policy gradient-based optimization procedure. To improve the estimation of the group-wise reward disparity during training, we employ a doubly robust estimator and further provide a convergence guarantee for policy optimization. Empirical results in synthetic and real-world datasets demonstrate that our method effectively reduces reward disparities while maintaining competitive overall performance.

Anton Orlichenko, Grant Daly, Ziyu Zhou et al.

Most packages for the analysis of fMRI-based functional connectivity (FC) and genomic data are used with a programming language interface, lacking an easy-to-navigate GUI frontend. This exacerbates two problems found in these types of data: demographic confounds and quality control in the face of high dimensionality of features. The reason is that it is too slow and cumbersome to use a programming interface to create all the necessary visualizations required to identify all correlations, confounding effects, or quality control problems in a dataset. To remedy this situation, we have developed ImageNomer, a data visualization and analysis tool that allows inspection of both subject-level and cohort-level demographic, genomic, and imaging features. The software is Python-based, runs in a self-contained Docker image, and contains a browser-based GUI frontend. We demonstrate the usefulness of ImageNomer by identifying an unexpected race confound when predicting achievement scores in the Philadelphia Neurodevelopmental Cohort (PNC) dataset. In the past, many studies have attempted to use FC to identify achievement-related features in fMRI. Using ImageNomer, we find a clear potential for confounding effects of race. Using correlation analysis in the ImageNomer software, we show that FCs correlated with Wide Range Achievement Test (WRAT) score are in fact more highly correlated with race. Investigating further, we find that whereas both FC and SNP (genomic) features can account for 10-15\% of WRAT score variation, this predictive ability disappears when controlling for race. In this work, we demonstrate the advantage of our ImageNomer GUI tool in data exploration and confound detection. Additionally, this work identifies race as a strong confound in FC data and casts doubt on the possibility of finding unbiased achievement-related features in fMRI and SNP data of healthy adolescents.

Aayush Mishra, Daniel Khashabi, Anqi Liu

Activation steering is a popular white-box control technique that modifies model activations to elicit an abstract change in output behavior. It has also become a standard tool in interpretability (e.g., probing truthfulness, or translating activations into human-readable explanations and safety research (e.g., studying jailbreakability). However, it is unclear whether steered activation states are realizable by any textual prompt. In this work, we cast this question as a surjectivity problem: for a fixed model, does every steered activation admit a pre-image under the model's natural forward pass? Under practical assumptions, we prove that activation steering pushes the residual stream off the manifold of states reachable from discrete prompts. Almost surely, no prompt can reproduce the same internal behavior induced by steering. We also illustrate this finding empirically across three widely used LLMs. Our results establish a formal separation between white-box steerability and black-box prompting. We therefore caution against interpreting the ease and success of activation steering as evidence of prompt-based interpretability or vulnerability, and argue for evaluation protocols that explicitly decouple white-box and black-box interventions.

Wenda Wang, Anqi Liu, Junqi Yang et al.

Converting hand-drawn sketches into structured 3D geometries remains challenging due to the difficulty of representing non-Euclidean surfaces and maintaining topological consistency. Existing generative models such as GANs, NeRFs, and diffusion architectures often fail to produce editable manifolds directly usable in downstream design workflows. We present Sketch2MinSurf, a hybrid vision-language and geometric optimization framework that integrates vision-language guidance with minimal-surface theory to generate smooth and editable 3D surfaces from hand-drawn sketches. The core of our approach is a spatial-topological encoding that represents geometry as tuples of node coordinates and real/virtual edge skeletons, enabling stable topological control during generation. We further introduce the Sketch2MinSurf Structural Loss (S2MS-Loss), a reward-modulated objective that jointly constrains geometric reconstruction and topological coherence. On a test set of 100 sketches, Sketch2MinSurf achieves a topological similarity score of 0.844, outperforming existing sketch-to-shape baselines. The generated manifolds are directly editable and free from non-manifold artifacts. A public art installation at a university showcases the method's potential for human-intent-driven 3D form generation. The dataset and code are available at https://anonymous.4open.science/r/Sketch2MinSurf/.

Aayush Mishra, Anqi Liu

We study the problem of invariant learning when the environment labels are unknown. We focus on the invariant representation notion when the Bayes optimal conditional label distribution is the same across different environments. Previous work conducts Environment Inference (EI) by maximizing the penalty term from Invariant Risk Minimization (IRM) framework. The EI step uses a reference model which focuses on spurious correlations to efficiently reach a good environment partition. However, it is not clear how to find such a reference model. In this work, we propose to repeat the EI process and retrain an ERM model on the \textit{majority} environment inferred by the previous EI step. Under mild assumptions, we find that this iterative process helps learn a representation capturing the spurious correlation better than the single step. This results in better Environment Inference and better Invariant Learning. We show that this method outperforms baselines on both synthetic and real-world datasets.

Anqi Liu, Shiyi Mu, Shugong Xu

Autonomous driving algorithms usually employ sRGB images as model input due to their compatibility with the human visual system. However, visually pleasing sRGB images are possibly sub-optimal for downstream tasks when compared to RAW images. The availability of RAW images is constrained by the difficulties in collecting real-world driving data and the associated challenges of annotation. To address this limitation and support research in RAW-domain driving perception, we design a novel and ultra-lightweight RAW reconstruction method. The proposed model introduces a learnable color correction matrix (CCM), which uses only a single convolutional layer to approximate the complex inverse image signal processor (ISP). Experimental results demonstrate that simulated RAW (simRAW) images generated by our method provide performance improvements equivalent to those produced by more complex inverse ISP methods when pretraining RAW-domain object detectors, which highlights the effectiveness and practicality of our approach.

Ha Manh Bui, Anqi Liu

Sampling-based methods, e.g., Deep Ensembles and Bayesian Neural Nets have become promising approaches to improve the quality of uncertainty estimation and robust generalization. However, they suffer from a large model size and high latency at test-time, which limits the scalability needed for low-resource devices and real-time applications. To resolve these computational issues, we propose Density-Softmax, a sampling-free deterministic framework via combining a density function built on a Lipschitz-constrained feature extractor with the softmax layer. Theoretically, we show that our model is the solution of minimax uncertainty risk and is distance-aware on feature space, thus reducing the over-confidence of the standard softmax under distribution shifts. Empirically, our method enjoys competitive results with state-of-the-art techniques in terms of uncertainty and robustness, while having a lower number of model parameters and a lower latency at test-time.

Yihong Guo, Dongqiangzi Ye, Sijia Chen et al.

Autonomous driving requires safe planning, but most learning-based planners lack explicit self-correction ability: once an unsafe action is proposed, there is no mechanism to correct it. Thus, we propose CorrectionPlanner, an autoregressive planner with self-correction that models planning as motion-token generation within a propose, evaluate, and correct loop. At each planning step, the policy proposes an action, namely a motion token, and a learned collision critic predicts whether it will induce a collision within a short horizon. If the critic predicts a collision, we retain the sequence of historical unsafe motion tokens as a self-correction trace, generate the next motion token conditioned on it, and repeat this process until a safe motion token is proposed or the safety criterion is met. This self-correction trace, consisting of all unsafe motion tokens, represents the planner's correction process in motion-token space, analogous to a reasoning trace in language models. We train the planner with imitation learning followed by model-based reinforcement learning using rollouts from a pretrained world model that realistically models agents' reactive behaviors. Closed-loop evaluations show that CorrectionPlanner reduces collision rate by over 20% on Waymax and achieves state-of-the-art planning scores on nuPlan.

Bin Wang, Anqi Liu, Jiangtao Zhao et al.

Short-term air traffic flow prediction in terminal airspace is essential for proactive air traffic management. Existing approaches predominantly model traffic flow as aggregated time series, despite traffic dynamics being governed by aircraft states and interactions in continuous airspace. Such aggregation obscures fine-grained information including aircraft kinematics, boundary interactions, and control intent. Here we present AeroSense, a state-to-flow modeling framework that predicts future traffic flow directly from instantaneous airspace situations represented as dynamic sets of aircraft states derived from ADS-B trajectories. By establishing an end-to-end mapping from microscopic aircraft states to future regional traffic flow, AeroSense preserves aircraft-level dynamics while naturally accommodating varying traffic density without relying on historical look-back windows. Experiments on a large-scale real-world dataset show that AeroSense consistently improves predictive accuracy over aggregation-based forecasting approaches, particularly during high-density traffic periods. These findings suggest that instantaneous airspace situations provide an effective alternative to conventional time-series-based traffic forecasting paradigms.

Shiyi Mu, Zichong Gu, Zhiqi Ai et al.

Compared to monocular 3D object detection, stereo-based 3D methods offer significantly higher accuracy but still suffer from high computational overhead and latency. The state-of-the-art stereo 3D detection method achieves twice the accuracy of monocular approaches, yet its inference speed is only half as fast. In this paper, we propose StereoDETR, an efficient stereo 3D object detection framework based on DETR. StereoDETR consists of two branches: a monocular DETR branch and a stereo branch. The DETR branch is built upon 2D DETR with additional channels for predicting object scale, orientation, and sampling points. The stereo branch leverages low-cost multi-scale disparity features to predict object-level depth maps. These two branches are coupled solely through a differentiable depth sampling strategy. To handle occlusion, we introduce a constrained supervision strategy for sampling points without requiring extra annotations. StereoDETR achieves real-time inference and is the first stereo-based method to surpass monocular approaches in speed. It also achieves competitive accuracy on the public KITTI benchmark, setting new state-of-the-art results on pedestrian and cyclist subsets. The code is available at https://github.com/shiyi-mu/StereoDETR-OPEN.

Aayush Mishra, Anqi Liu

Reliable and accurate estimation of the error of an ML model in unseen test domains is an important problem for safe intelligent systems. Prior work uses disagreement discrepancy (DIS^2) to derive practical error bounds under distribution shifts. It optimizes for a maximally disagreeing classifier on the target domain to bound the error of a given source classifier. Although this approach offers a reliable and competitively accurate estimate of the target error, we identify a problem in this approach which causes the disagreement discrepancy objective to compete in the overlapping region between source and target domains. With an intuitive assumption that the target disagreement should be no more than the source disagreement in the overlapping region due to high enough support, we devise Overlap-aware Disagreement Discrepancy (ODD). Maximizing ODD only requires disagreement in the non-overlapping target domain, removing the competition. Our ODD-based bound uses domain-classifiers to estimate domain-overlap and better predicts target performance than DIS^2. We conduct experiments on a wide array of benchmarks to show that our method improves the overall performance-estimation error while remaining valid and reliable. Our code and results are available on GitHub.

Zhitong Yang, Xing Ma, Anqi Liu et al.

Task-oriented dialog(TOD) aims to assist users in achieving specific goals through multi-turn conversation. Recently, good results have been obtained based on large pre-trained models. However, the labeled-data scarcity hinders the efficient development of TOD systems at scale. In this work, we constructed a weakly supervised dataset based on a teacher/student paradigm that leverages a large collection of unlabelled dialogues. Furthermore, we built a modular dialogue system and integrated coarse-to-fine grained classification for user intent detection. Experiments show that our method can reach the dialog goal with a higher success rate and generate more coherent responses.

Kartheek Bondugula, Santiago Mazuelas, Aritz Pérez et al.

Loss functions play a central role in supervised classification. Cross-entropy (CE) is widely used, whereas the mean absolute error (MAE) loss can offer robustness but is difficult to optimize. Interpolating between the CE and MAE losses, generalized cross-entropy (GCE) has recently been introduced to provide a trade-off between optimization difficulty and robustness. Existing formulations of GCE result in a non-convex optimization over classification margins that is prone to underfitting, leading to poor performances with complex datasets. In this paper, we propose a minimax formulation of generalized cross-entropy (MGCE) that results in a convex optimization over classification margins. Moreover, we show that MGCEs can provide an upper bound on the classification error. The proposed bilevel convex optimization can be efficiently implemented using stochastic gradient computed via implicit differentiation. Using benchmark datasets, we show that MGCE achieves strong accuracy, faster convergence, and better calibration, especially in the presence of label noise.

Anqi Liu, Jiangtao Zhao, Guiyuan Jiang et al.

Accurate air traffic prediction in the terminal airspace (TA) is pivotal for proactive air traffic management (ATM). However, existing data-driven approaches predominantly rely on time series-based forecasting paradigms, which inherently overlook critical aircraft state information, such as real-time kinematics and proximity to airspace boundaries. To address this limitation, we propose \textit{AeroSense}, a direct state-to-flow modeling framework for air traffic prediction. Unlike classical time series-based methods that first aggregate aircraft trajectories into macroscopic flow sequences before modeling, AeroSense explicitly represents the real-time airspace situation as \textit{a dynamic set of aircraft states}, enabling the direct processing of a variable number of aircraft instead of time series as inputs. Specifically, we introduce a situation-aware state representation that enables AeroSense to sense the instantaneous terminal airspace situation directly from microscopic aircraft states. Furthermore, we design a model architecture that incorporates masked self-attention to capture inter-aircraft interactions, together with two decoupled prediction heads to model heterogeneous flow dynamics across two key functional areas of the TA. Extensive experiments on a large-scale real-world airport dataset demonstrate that AeroSense consistently achieves state-of-the-art performance, validating that direct modeling of microscopic aircraft states yields substantially higher predictive fidelity than time series-based baselines. Moreover, the proposed framework exhibits superior robustness during peak traffic periods, achieves Pareto-optimal performance under dayparting multi-object evaluation, and provides meaningful interpretability through attention-based visualizations.

Drew Prinster, Samuel Stanton, Anqi Liu et al.

As artificial intelligence (AI) / machine learning (ML) gain widespread adoption, practitioners are increasingly seeking means to quantify and control the risk these systems incur. This challenge is especially salient when such systems have autonomy to collect their own data, such as in black-box optimization and active learning, where their actions induce sequential feedback-loop shifts in the data distribution. Conformal prediction is a promising approach to uncertainty and risk quantification, but prior variants' validity guarantees have assumed some form of ``quasi-exchangeability'' on the data distribution, thereby excluding many types of sequential shifts. In this paper we prove that conformal prediction can theoretically be extended to \textit{any} joint data distribution, not just exchangeable or quasi-exchangeable ones. Although the most general case is exceedingly impractical to compute, for concrete practical applications we outline a procedure for deriving specific conformal algorithms for any data distribution, and we use this procedure to derive tractable algorithms for a series of AI/ML-agent-induced covariate shifts. We evaluate the proposed algorithms empirically on synthetic black-box optimization and active learning tasks.

Yihong Guo, Yixuan Wang, Yuanyuan Shi et al.

Training a policy in a source domain for deployment in the target domain under a dynamics shift can be challenging, often resulting in performance degradation. Previous work tackles this challenge by training on the source domain with modified rewards derived by matching distributions between the source and the target optimal trajectories. However, pure modified rewards only ensure the behavior of the learned policy in the source domain resembles trajectories produced by the target optimal policies, which does not guarantee optimal performance when the learned policy is actually deployed to the target domain. In this work, we propose to utilize imitation learning to transfer the policy learned from the reward modification to the target domain so that the new policy can generate the same trajectories in the target domain. Our approach, Domain Adaptation and Reward Augmented Imitation Learning (DARAIL), utilizes the reward modification for domain adaptation and follows the general framework of generative adversarial imitation learning from observation (GAIfO) by applying a reward augmented estimator for the policy optimization step. Theoretically, we present an error bound for our method under a mild assumption regarding the dynamics shift to justify the motivation of our method. Empirically, our method outperforms the pure modified reward method without imitation learning and also outperforms other baselines in benchmark off-dynamics environments.

Liaoyaqi Wang, Chunsheng Zuo, William Jurayj et al.

Scaling test-time computation with reinforcement learning (RL) has emerged as a reliable path to improve large language models (LLM) reasoning ability. Yet, outcome-based reward often incentivizes models to be overconfident, leading to hallucinations, unreliable confidence-based control, and unnecessary compute allocation. We introduce Reinforcement Learning with Confidence Margin (\textbf{RLCM}), a calibration-aware RL framework that jointly optimizes correctness and confidence reliability via a margin-enhanced process reward over intermediate-budget completions. Rather than aligning confidence to correctness likelihoods, RLCM encourages to widen the confidence margin between correct and incorrect steps within a single reasoning trajectory. Across mathematical, code, logic and science benchmarks, our method substantially improves calibration while maintaining or improving accuracy. We further show that, with calibrated confidence signals, the resulting models enable more efficient conformal risk control and effective confidence-weighted aggregation.

Zhengping Jiang, Anqi Liu, Benjamin Van Durme

Language model outputs are not always reliable, thus prompting research into how to adapt model responses based on uncertainty. Common approaches include: \emph{abstention}, where models refrain from generating responses when uncertain; and \emph{linguistic calibration}, where models hedge their statements using uncertainty quantifiers. However, abstention can withhold valuable information, while linguistically calibrated responses are often challenging to leverage in downstream tasks. We propose a unified view, Conformal Linguistic Calibration (CLC), which reinterprets linguistic calibration as \emph{answer set prediction}. First we present a framework connecting abstention and linguistic calibration through the lens of linguistic pragmatics. We then describe an implementation of CLC that allows for controlling the level of imprecision in model responses. Results demonstrate our method produces calibrated outputs with conformal guarantees on factual accuracy. Further, our approach enables fine-tuning models to perform uncertainty-aware adaptive claim rewriting, offering a controllable balance between factuality and specificity.

Ha Manh Bui, Anqi Liu

Morden deep ensembles technique achieves strong uncertainty estimation performance by going through multiple forward passes with different models. This is at the price of a high storage space and a slow speed in the inference (test) time. To address this issue, we propose Density-Regression, a method that leverages the density function in uncertainty estimation and achieves fast inference by a single forward pass. We prove it is distance aware on the feature space, which is a necessary condition for a neural network to produce high-quality uncertainty estimation under distribution shifts. Empirically, we conduct experiments on regression tasks with the cubic toy dataset, benchmark UCI, weather forecast with time series, and depth estimation under real-world shifted applications. We show that Density-Regression has competitive uncertainty estimation performance under distribution shifts with modern deep regressors while using a lower model size and a faster inference speed.

Zhengping Jiang, Yining Lu, Hanjie Chen et al.

Free-text rationales play a pivotal role in explainable NLP, bridging the knowledge and reasoning gaps behind a model's decision-making. However, due to the diversity of potential reasoning paths and a corresponding lack of definitive ground truth, their evaluation remains a challenge. Existing evaluation metrics rely on the degree to which a rationale supports a target label, but we find these fall short in evaluating rationales that inadvertently leak the labels. To address this problem, we propose RORA, a Robust free-text Rationale evaluation against label leakage. RORA quantifies the new information supplied by a rationale to justify the label. This is achieved by assessing the conditional V-information \citep{hewitt-etal-2021-conditional} with a predictive family robust against leaky features that can be exploited by a small model. RORA consistently outperforms existing approaches in evaluating human-written, synthetic, or model-generated rationales, particularly demonstrating robustness against label leakage. We also show that RORA aligns well with human judgment, providing a more reliable and accurate measurement across diverse free-text rationales.

Liaoyaqi Wang, Zhengping Jiang, Anqi Liu et al.

We present a state-of-the-art model for fine-grained probability estimation of propositions conditioned on context. Recent advances in large language models (LLMs) have significantly enhanced their reasoning capabilities, particularly on well-defined tasks with complete information. However, LLMs continue to struggle with making accurate and well-calibrated probabilistic predictions under uncertainty or partial information. While incorporating uncertainty into model predictions often boosts performance, obtaining reliable estimates of that uncertainty remains understudied. In particular, LLM probability estimates tend to be coarse and biased towards more frequent numbers. Through a combination of human and synthetic data creation and assessment, scaling to larger models, and better supervision, we propose a set of strong and precise probability estimation models. We conduct systematic evaluations across tasks that rely on conditional probability estimation and show that our approach consistently outperforms existing fine-tuned and prompting-based methods by a large margin.

Zhouxiang Fang, Aayush Mishra, Muhan Gao et al.

Recent works have suggested that In-Context Learning (ICL) operates in dual modes, i.e. task retrieval (remember learned patterns from pre-training) and task learning (inference-time ''learning'' from demonstrations). However, disentangling these the two modes remains a challenging goal. We introduce ICL CIPHERS, a class of task reformulations based on substitution ciphers borrowed from classic cryptography. In this approach, a subset of tokens in the in-context inputs are substituted with other (irrelevant) tokens, rendering English sentences less comprehensible to human eye. However, by design, there is a latent, fixed pattern to this substitution, making it reversible. This bijective (reversible) cipher ensures that the task remains a well-defined task in some abstract sense, despite the transformations. It is a curious question if LLMs can solve tasks reformulated by ICL CIPHERS with a BIJECTIVE mapping, which requires ''deciphering'' the latent cipher. We show that LLMs are better at solving tasks reformulated by ICL CIPHERS with BIJECTIVE mappings than the NON-BIJECTIVE (irreversible) baseline, providing a novel approach to quantify ''learning'' in ICL. While this gap is small, it is consistent across the board on four datasets and six models. Finally, we examine LLMs' internal representations and identify evidence in their ability to decode the ciphered inputs.

Kevin He, David Adam, Sarah Han-Oh et al.

Measurement quality assurance (QA) practices play a key role in the safe use of Intensity Modulated Radiation Therapies (IMRT) for cancer treatment. These practices have reduced measurement-based IMRT QA failure below 1%. However, these practices are time and labor intensive which can lead to delays in patient care. In this study, we examine how conformal prediction methodologies can be used to robustly triage plans. We propose a new training-aware conformal risk control method by combining the benefit of conformal risk control and conformal training. We incorporate the decision making thresholds based on the gamma passing rate, along with the risk functions used in clinical evaluation, into the design of the risk control framework. Our method achieves high sensitivity and specificity and significantly reduces the number of plans needing measurement without generating a huge confidence interval. Our results demonstrate the validity and applicability of conformal prediction methods for improving efficiency and reducing the workload of the IMRT QA process.

Ha Manh Bui, Enrique Mallada, Anqi Liu

By leveraging the representation power of deep neural networks, neural upper confidence bound (UCB) algorithms have shown success in contextual bandits. To further balance the exploration and exploitation, we propose Neural-$σ^2$-LinearUCB, a variance-aware algorithm that utilizes $σ^2_t$, i.e., an upper bound of the reward noise variance at round $t$, to enhance the uncertainty quantification quality of the UCB, resulting in a regret performance improvement. We provide an oracle version for our algorithm characterized by an oracle variance upper bound $σ^2_t$ and a practical version with a novel estimation for this variance bound. Theoretically, we provide rigorous regret analysis for both versions and prove that our oracle algorithm achieves a better regret guarantee than other neural-UCB algorithms in the neural contextual bandits setting. Empirically, our practical method enjoys a similar computational efficiency, while outperforming state-of-the-art techniques by having a better calibration and lower regret across multiple standard settings, including on the synthetic, UCI, MNIST, and CIFAR-10 datasets.

Aayush Mishra, Daniel Khashabi, Anqi Liu

Supervised Fine-Tuning (SFT) is used to specialize model behavior by training weights to produce intended target responses for queries. In contrast, In-Context Learning (ICL) adapts models during inference with instructions or demonstrations in the prompt. ICL can offer better generalizability and more calibrated responses compared to SFT in data scarce settings, at the cost of more inference compute. In this work, we ask the question: Can ICL's internal computations be used to improve the qualities of SFT? We first show that ICL and SFT produce distinct activation patterns, indicating that the two methods achieve adaptation through different functional mechanisms. Motivated by this observation and to use ICL's rich functionality, we introduce ICL Activation Alignment (IA2), a self-distillation technique which aims to replicate ICL's activation patterns in SFT models and incentivizes ICL-like internal reasoning. Performing IA2 as a priming step before SFT significantly improves the accuracy and calibration of model outputs, as shown by our extensive empirical results on 12 popular benchmarks and 2 model families. This finding is not only practically useful, but also offers a conceptual window into the inner mechanics of model adaptation.

Drew Prinster, Xing Han, Anqi Liu et al.

Responsibly deploying artificial intelligence (AI) / machine learning (ML) systems in high-stakes settings arguably requires not only proof of system reliability, but also continual, post-deployment monitoring to quickly detect and address any unsafe behavior. Methods for nonparametric sequential testing -- especially conformal test martingales (CTMs) and anytime-valid inference -- offer promising tools for this monitoring task. However, existing approaches are restricted to monitoring limited hypothesis classes or ``alarm criteria'' (e.g., detecting data shifts that violate certain exchangeability or IID assumptions), do not allow for online adaptation in response to shifts, and/or cannot diagnose the cause of degradation or alarm. In this paper, we address these limitations by proposing a weighted generalization of conformal test martingales (WCTMs), which lay a theoretical foundation for online monitoring for any unexpected changepoints in the data distribution while controlling false-alarms. For practical applications, we propose specific WCTM algorithms that adapt online to mild covariate shifts (in the marginal input distribution), quickly detect harmful shifts, and diagnose those harmful shifts as concept shifts (in the conditional label distribution) or extreme (out-of-support) covariate shifts that cannot be easily adapted to. On real-world datasets, we demonstrate improved performance relative to state-of-the-art baselines.

Lei Huang, Chuan Qiu, Kuan-Jui Su et al.

Genotype imputation enables dense variant coverage for genome-wide association and risk-prediction studies, yet conventional reference-panel methods remain limited by ancestry bias and reduced rare-variant accuracy. We present Genotype Bidirectional Encoder Representations from Transformers (GenoBERT), a transformer-based, reference-free framework that tokenizes phased genotypes and uses a self-attention mechanism to capture both short- and long-range linkage disequilibrium (LD) dependencies. Benchmarking on two independent datasets including the Louisiana Osteoporosis Study (LOS) and the 1000 Genomes Project (1KGP) across ancestry groups and multiple genotype missingness levels (5-50%) shows that GenoBERT achieves the highest overall accuracy compared to four baseline methods (Beagle5.4, SCDA, BiU-Net, and STICI). At practical sparsity levels (up to 25% missing), GenoBERT attains high overall imputation accuracy ($r^2 approx 0.98$) across datasets, and maintains robust performance ($r^2 > 0.90$) even at 50% missingness. Experimental results across different ancestries confirm consistent gains across datasets, with resilience to small sample sizes and weak LD. A 128-SNP (single-nucleotide polymorphism) context window (approximately 100 Kb) is validated through LD-decay analyses as sufficient to capture local correlation structures. By eliminating reference-panel dependence while preserving high accuracy, GenoBERT provides a scalable and robust solution for genotype imputation and a foundation for downstream genomic modeling.

Nathan Breslow, Aayush Mishra, Mahler Revsine et al.

In-context learning (ICL) -- the capacity of a model to infer and apply abstract patterns from examples provided within its input -- has been extensively studied in large language models trained for next-token prediction on human text. In fact, prior work often attributes this emergent behavior to distinctive statistical properties in human language. This raises a fundamental question: can ICL arise organically in other sequence domains purely through large-scale predictive training? To explore this, we turn to genomic sequences, an alternative symbolic domain rich in statistical structure. Specifically, we study the Evo2 genomic model, trained predominantly on next-nucleotide (A/T/C/G) prediction, at a scale comparable to mid-sized LLMs. We develop a controlled experimental framework comprising symbolic reasoning tasks instantiated in both linguistic and genomic forms, enabling direct comparison of ICL across genomic and linguistic models. Our results show that genomic models, like their linguistic counterparts, exhibit log-linear gains in pattern induction as the number of in-context demonstrations increases. To the best of our knowledge, this is the first evidence of organically emergent ICL in genomic sequences, supporting the hypothesis that ICL arises as a consequence of large-scale predictive modeling over rich data. These findings extend emergent meta-learning beyond language, pointing toward a unified, modality-agnostic view of in-context learning.

Ha Manh Bui, Felix Parker, Kimia Ghobadi et al.

We study the Non-Stationary Reinforcement Learning (RL) under distribution shifts in both finite-horizon episodic and infinite-horizon discounted Markov Decision Processes (MDPs). In the finite-horizon case, the transition functions may suddenly change at a particular episode. In the infinite-horizon setting, such changes can occur at an arbitrary time step during the agent's interaction with the environment. While the Q-learning Upper Confidence Bound algorithm (QUCB) can discover a proper policy during learning, due to the distribution shifts, this policy can exploit sub-optimal rewards after the shift happens. To address this issue, we propose Density-QUCB (DQUCB), a shift-aware Q-learning~UCB algorithm, which uses a transition density function to detect distribution shifts, then leverages its likelihood to enhance the uncertainty estimation quality of Q-learning~UCB, resulting in a balance between exploration and exploitation. Theoretically, we prove that our oracle DQUCB achieves a better regret guarantee than QUCB. Empirically, our DQUCB enjoys the computational efficiency of model-free RL and outperforms QUCB baselines by having a lower regret across RL tasks, as well as a real-world COVID-19 patient hospital allocation task using a Deep-Q-learning architecture.

Ha Manh Bui, Iliana Maifeld-Carucci, Anqi Liu

We study the problem of actively learning a classifier with a low calibration error. One of the most popular Acquisition Functions (AFs) in pool-based Active Learning (AL) is querying by the model's uncertainty. However, we recognize that an uncalibrated uncertainty model on the unlabeled pool may significantly affect the AF effectiveness, leading to sub-optimal generalization and high calibration error on unseen data. Deep Neural Networks (DNNs) make it even worse as the model uncertainty from DNN is usually uncalibrated. Therefore, we propose a new AF by estimating calibration errors and query samples with the highest calibration error before leveraging DNN uncertainty. Specifically, we utilize a kernel calibration error estimator under the covariate shift and formally show that AL with this AF eventually leads to a bounded calibration error on the unlabeled pool and unseen test data. Empirically, our proposed method surpasses other AF baselines by having a lower calibration and generalization error across pool-based AL settings.

Andrea Wynn, Metod Jazbec, Charith Peris et al. · amazon-science

Large language models (LLMs) demonstrate a remarkable ability to learn new tasks from a few in-context examples. However, this flexibility introduces safety concerns: LLMs can be influenced by incorrect or malicious demonstrations -- for example, if an adversary tampers with or injects harmful examples without a human supervisor noticing. This motivates principled designs in which the system itself includes built-in mechanisms to guard against such attacks. We propose a novel approach to limit the degree to which harmful demonstrations can degrade model performance. First, we define a baseline ``safe'' behavior for the model -- the model's performance given no in-context demonstrations (zero-shot). Next, we apply distribution-free risk control (DFRC) to control the extent to which in-context samples can decay performance below zero-shot. We achieve this by leveraging dynamic early exit prediction, ignoring later attention heads that attend the most to the unsafe inputs. Finally, we propose modifications to DFRC that allow it to both control risk for harmful inputs \textit{and} leverage performance and efficiency gains on helpful inputs. We present both theoretical and empirical results showing that our approach can effectively control risk for harmful in-context demonstrations while simultaneously achieving substantial computational efficiency gains with helpful demonstrations.

Junjie Luo, Yihong Guo, Anqi Liu et al.

Messaging patients is a critical part of healthcare communication, helping to improve things like medication adherence and healthy behaviors. However, traditional mobile message design has significant limitations due to its inability to explore the high-dimensional design space. We develop PAME-AI, a novel approach for Patient Messaging Creation and Optimization using Agentic AI. Built on the Data-Information-Knowledge-Wisdom (DIKW) hierarchy, PAME-AI offers a structured framework to move from raw data to actionable insights for high-performance messaging design. PAME-AI is composed of a system of specialized computational agents that progressively transform raw experimental data into actionable message design strategies. We demonstrate our approach's effectiveness through a two-stage experiment, comprising of 444,691 patient encounters in Stage 1 and 74,908 in Stage 2. The best-performing generated message achieved 68.76% engagement compared to the 61.27% baseline, representing a 12.2% relative improvement in click-through rates. This agentic architecture enables parallel processing, hypothesis validation, and continuous learning, making it particularly suitable for large-scale healthcare communication optimization.

Yihong Guo, Yu Yang, Pan Xu et al.

We study off-dynamics offline reinforcement learning, where the goal is to learn a policy from offline source and limited target datasets with mismatched dynamics. Existing methods either penalize the reward or discard source transitions occurring in parts of the transition space with high dynamics shift. As a result, they optimize the policy using data from low-shift regions, limiting exploration of high-reward states in the target domain that do not fall within these regions. Consequently, such methods often fail when the dynamics shift is significant or the optimal trajectories lie outside the low-shift regions. To overcome this limitation, we propose MOBODY, a Model-Based Off-Dynamics Offline RL algorithm that optimizes a policy using learned target dynamics transitions to explore the target domain, rather than only being trained with the low dynamics-shift transitions. For the dynamics learning, built on the observation that achieving the same next state requires taking different actions in different domains, MOBODY employs separate action encoders for each domain to encode different actions to the shared latent space while sharing a unified representation of states and a common transition function. We further introduce a target Q-weighted behavior cloning loss in policy optimization to avoid out-of-distribution actions, which push the policy toward actions with high target-domain Q-values, rather than high source domain Q-values or uniformly imitating all actions in the offline dataset. We evaluate MOBODY on a wide range of MuJoCo and Adroit benchmarks, demonstrating that it outperforms state-of-the-art off-dynamics RL baselines as well as policy learning methods based on different dynamics learning baselines, with especially pronounced improvements in challenging scenarios where existing methods struggle.