AAAI 2026 Papers

Large language models (LLMs) have achieved remarkable success in many domains, but concerns about data quality and privacy are growing. Federated Learning (FL) offers a privacy-preserving solution by training a model on local clients without sharing data. However, the impact of biased private data on LLMs fine-tuned through FL remains understudied. This work investigates how client-side biased data affects the global model during federated fine-tuning of LLMs. We simulate realistic scenarios where some clients possess datasets containing social biases (stereotypes, discriminatory language) while others have clean data through extensive experiments with popular FL algorithms (FedAvg, FedAdam and FedProx) and popular LLMs (LLaMA, Mistral, Phi-3 and Gemma) across datasets with varying bias proportions (33%, 66%, 100%). Our findings reveal that 1) FedAdam consistently shows the lowest bias propagation, reducing CrowS-Pairs scores by up to 15% compared to FedAvg; 2) Even small amounts of biased data (33%) can significantly influence global model bias; 3) Mixed biased and neutral data distributions lead to 5%-7% higher bias scores than segregated distributions. Additionally, we propose Bias-Aware Model Aggregation (BAMA), a novel debiasing method for federated fine-tuning that consistently reduces bias across various models and algorithms.

Early warning systems for disease outbreaks play a crucial role in public health for management and contingency planning. However, most predictive modeling works focus on flat models that incorporate exogenous inputs (e.g. climate, demographics) to predict future outbreaks at different locations, but do not jointly model multiple spatial aggregation levels. In this paper, we introduce HierarNet, a unique independent-interactive hierarchical forecasting framework that aims to predict disease outbreaks at different levels of spatial resolution, such as provinces, regions, and nations. HierarNet consists of two main phases. In the local phase, we train independent forecasting models for all locations at all levels. In the global phase, all models iteratively interact with others across different levels via their hierarchical relationships under an ensemble fashion to maximize their agreements. This global local hierarchical interactive scheme makes HierarNet a highly effective and flexible method (i.e. it can work with an arbitrary base prediction model and available exogenous data for each location independently). Extensive experiments are conducted on various disease datasets (e.g., Dengue fever, flu, diarrhea, and Bluetongue) in different countries (e.g., France, Vietnam, and USA) to show the performance of HierarNet compared to 19 state-of-the-art (SOTA) methods such as MinT, DYCHEM, WITRAN, SegRNN, TSMixer, PatchTST, or iTransformer. We also illustrate the generability of HierarNet in other domains, e.g., web traffic forecasting.

Crowdsourced delivery platforms (e.g., Meituan, Uber Eats, DoorDash) have become vital infrastructure in urban logistics, yet their competitive order-grabbing mechanisms often lead to strategy homogenization, inefficiency, and income inequality. This paper presents ARDE (Adaptive Regulation via Dual-layer Evolution), an evolutionary governance framework that integrates individual reinforcement learning with adaptive platform-level regulation. The outer agent dynamically generates governance signals based on system diagnostics (strategy entropy, Gini coefficient, completion rate), while inner agents employ Diffusion Q-Learning guided by a language-model-driven reward shaping module to promote fairness and strategy diversity. Experiments on real-world datasets show that ARDE achieves stable diversity (0.997 ± 0.184), reduces inequality (Gini change 1.3%), and maintains high efficiency. Further comparison (ARDE-PPO vs. MAPPO) confirms that its advantages stem from explicit hierarchical governance rather than algorithmic coincidence. Overall, ARDE offers a scalable and interpretable paradigm for reconciling individual rationality with collective welfare in gig economies and other multi-agent socio-technical systems.

Sustainability is becoming increasingly critical in the maritime transport, encompassing both environmental and social impacts, such as Greenhouse Gas (GHG) emissions and navigational safety. Traditional vessel navigation heavily relies on human experience, often lacking autonomy and emission awareness, and is prone to human errors that may compromise safety. In this paper, we propose a Curriculum Reinforcement Learning (CRL) framework integrated with a realistic, data-driven marine simulation environment and a machine learning-based fuel consumption prediction module. The simulation environment is constructed using real-world vessel movement data and enhanced with a Diffusion Model to simulate dynamic maritime conditions. Vessel fuel consumption is estimated using historical operational data and learning-based regression. The surrounding environment is represented as image-based inputs to capture spatial complexity. We design a lightweight, policy-based CRL agent with a comprehensive reward mechanism that considers safety, emissions, timeliness, and goal completion. This framework effectively handles complex tasks progressively while ensuring stable and efficient learning in continuous action spaces. We validate the proposed approach in a sea area of the Indian Ocean, demonstrating its efficacy in enabling sustainable and safe vessel navigation.

Understanding opinion evolution in complex social networks is crucial for modeling social influence and predicting collective behavior. Yet, most models overlook how community structures shape opinion updates, often assuming homogeneous influence. This abstraction neglects individuals’ stronger responsiveness to intra-community peers—an empirically observed driver of localized consensus and inter-group polarization. We propose GCAOFP, a co-evolutionary framework that jointly models opinion dynamics and community formation as an integrated process. In GCAOFP, agents strategically alternate between two coupled modules: (1) a Community Dynamics Module, where agents play a non-cooperative game to optimize community memberships based on opinion alignment and structural cohesion; and (2) an Opinion Adjustment Module, where agents revise opinions via a bounded-confidence mechanism modulated by community-induced influence weights. This dual-stage process captures the feedback loop between structure and opinion. We prove that GCAOFP converges to stable equilibria, ensuring intra-community consensus and inter-community diversity—dynamics that standard models fail to replicate. Experiments on real-world networks show that GCAOFP better reproduces localized opinion clusters, while offering strong scalability and interpretability, illuminating the strategic foundations of polarization.

Formative feedback is widely recognized as one of the most effective drivers of student learning, yet it remains difficult to implement equitably at scale. In large or low-resource courses, instructors often lack the time, staffing, and bandwidth required to review and respond to every student reflection, creating gaps in support precisely where learners would benefit most. This paper presents a theory-grounded system that uses five coordinated role-based LLM agents (Evaluator, Equity Monitor, Metacognitive Coach, Aggregator, and Reflexion Reviewer) to score learner reflections with a shared rubric and to generate short, bias-aware, learner-facing comments. The agents first produce structured rubric scores, then check for potentially biased or exclusionary language, add metacognitive prompts that invite students to think about their own thinking, and finally compose a concise feedback message of at most 120 words. The system includes simple fairness checks that compare scoring error across lower and higher scoring learners, enabling instructors to monitor and bound disparities in accuracy. We evaluate the pipeline in a 12-session AI literacy program with adult learners. In this setting, the system produces rubric scores that approach expert-level agreement, and trained graders rate the AI-generated comments as helpful, empathetic, and well aligned with instructional goals. Taken together, these results show that multi-agent LLM systems can deliver equitable, high-quality formative feedback at a scale and speed that would be impossible for human graders alone. The approach demonstrates how structured agent roles, fairness checks, and learning-science principles can work together to support instructors while preserving pedagogical intent. More broadly, the work points toward a future where feedback-rich learning becomes feasible for any course size or context, advancing long-standing goals of equity, access, and instructional capacity in education.

Accurate forecasting of the grid carbon intensity factor (CIF) is critical for enabling demand-side management and reducing emissions in modern electricity systems. Leveraging multiple interrelated time series, CIF prediction is typically formulated as a multivariate time series forecasting problem. Despite advances in deep learning-based methods, it remains challenging to capture the fine-grained local-temporal dependencies, dynamic higher-order cross-variable dependencies, and complex multi-frequency patterns for CIF forecasting. To address these issues, we propose a novel model that integrates two parallel modules: 1) one enhances the extraction of local-temporal dependencies under multi-frequency by applying multiple wavelet-based convolutional kernels to overlapping patches of varying lengths; 2) the other captures dynamic cross-variable dependencies under multi-frequency to model how inter-variable relationships evolve across the time-frequency domain. Evaluations on four representative electricity markets from Australia, featuring varying levels of renewable penetration, demonstrate that the proposed method outperforms the state-of-the-art models. An ablation study further validates the complementary benefits of the two proposed modules. Designed with built-in interpretability, the proposed model also enables better understanding of its predictive behavior, as shown in a case study where it adaptively shifts attention to relevant variables and time intervals during a disruptive event.

Effective crime linkage analysis is crucial for identifying serial offenders and enhancing public safety. To address the limitations of traditional crime linkage methods when handling high-dimensional, sparse, and heterogeneous data, this paper proposes a Siamese Autoencoder framework to learn meaningful latent representations and uncover correlations in highly complex data. Using a dataset from the Violent Crime Linkage Analysis System—a database maintained by the Serious Crime Analysis Section of the UK’s National Crime Agency—our approach mitigates signal dilution in high-dimensional sparse data through decoder-stage integration of geographic-temporal features. This integration amplifies learned behavioral representations rather than allowing them to be overwhelmed at the input stage, leading to consistent improvements over baseline methods across multiple metrics. We further examine how different data reduction strategies based on domain-expert can impact model performance, offering practical insights into preprocessing for crime linkage. Our solution shows that advanced machine learning approaches can enhance linkage accuracy, improving AUC by up to 9% over traditional methods and providing insights to support human decision-making in crime investigation.

Ensuring fairness in machine learning requires understanding how sensitive attributes like race or gender causally influence outcomes. Existing causal discovery (CD) methods often struggle to recover fairness-relevant pathways in the presence of noise, confounding, or data corruption. Large language models (LLMs) offer a complementary signal by leveraging semantic priors from variable metadata. We propose a hybrid LLM-guided CD framework that extends a breadth-first search strategy with active learning and dynamic scoring. Variable pairs are prioritized for querying using a composite score combining mutual information, partial correlation, and LLM confidence, enabling more efficient and robust structure discovery. To evaluate fairness sensitivity, we introduce a semi-synthetic benchmark based on the UCI Adult dataset, embedding domain-informed bias pathways alongside noise and latent confounders. We assess how well CD methods recover both global graph structure and fairness-critical paths (e.g., sex→education→income). Our results demonstrate that LLM-guided methods, including our active, dynamically scored variant, outperform baselines in recovering fairness-relevant structure under noisy conditions. We analyze when LLM-driven insights complement statistical dependencies and discuss implications for fairness auditing in high-stakes domains.

Energy usage prediction is important for various real-world applications, including grid management, infrastructure planning, and disaster response. Although a plethora of deep learning approaches have been proposed to perform this task, most of them either overlook the essential spatial correlations across households or fail to scale to individualized prediction, making them less effective for accurate fine-grained user-level prediction. In addition, due to the dynamic and uncertain nature of energy usage caused by various factors such as extreme weather events, quantifying uncertainty for reliable prediction is also significant, but it has not been fully explored in existing work. In this paper, we propose a unified framework called TrustEnergy for accurate and reliable user-level energy usage prediction. There are two key technical components in TrustEnergy, (i) a Hierarchical Spatiotemporal Representation module to efficiently capture both macro and micro energy usage patterns with a novel memory-augmented spatiotemporal graph neural network, and (ii) an innovative Sequential Conformalized Quantile Regression module to dynamically adjust uncertainty bounds to ensure valid prediction intervals over time, without making strong assumptions about the underlying data distribution. We implement and evaluate our TrustEnergy framework by working with an electricity provider in Florida, and the results show our TrustEnergy can achieve a 5.4% increase in prediction accuracy and 5.7% improvement in uncertainty quantification compared to state-of-the-art baselines.

Invasive mechanical ventilation (MV) is a life-sustaining therapy commonly used in the intensive care unit (ICU) for patients with severe and acute conditions. These patients frequently rely on MV for breathing. Given the high risk of death in such cases, optimal MV settings can reduce mortality, minimize ventilator-induced lung injury, shorten ICU stays, and ease the strain on healthcare resources. However, optimizing MV settings remains a complex and error-prone process due to patient-specific variability. While Offline Reinforcement Learning (RL) shows promise for optimizing MV settings, current methods struggle with the hybrid (continuous and discrete) nature of MV settings. Discretizing continuous settings leads to exponential growth in the action space, which limits the number of optimizable settings. Converting the predictions back to continuous can cause a distribution shift, compromising safety and performance. To address this challenge, in the IntelliLung project, we are developing an AI-based approach where we constrain the action space and employ factored action critics. This approach allows us to scale to six optimizable settings compared to 2-3 in previous studies. We adapt SOTA offline RL algorithms to operate directly on hybrid action spaces, avoiding the pitfalls of discretization. We also introduce a clinically grounded reward function based on ventilator-free days and physiological targets. Using multi-objective optimization for reward selection, we show that this leads to a more equitable consideration of all clinically relevant objectives. Notably, we develop a system in close collaboration with healthcare professionals that is aligned with real-world clinical objectives and designed with future deployment in mind.

Real-world adoption of closed-loop insulin delivery systems (CLIDS) in type 1 diabetes remains low, driven not by technical failure, but by diverse behavioral, psychosocial, and social barriers. We introduce ChatCLIDS, the first benchmark to rigorously evaluate LLM–driven persuasive dialogue for health behavior change. Our framework features a library of expert-validated virtual patients, each with clinically grounded, heterogeneous profiles and realistic adoption barriers, and simulates multi-turn interactions with nurse agents equipped with a diverse set of evidence-based persuasive strategies. ChatCLIDS uniquely supports longitudinal counseling and adversarial social influence scenarios, enabling robust, multi-dimensional evaluation. Our findings reveal that while larger and more reflective LLMs adapt strategies over time, all models struggle to overcome resistance, especially under realistic social pressure. These results highlight critical limitations of current LLMs for behavior change, and offer a high-fidelity, scalable testbed for advancing trustworthy persuasive AI in healthcare and beyond.

Artificial Intelligence Generated Content (AIGC) assisting image production triggers controversy in journalism while attracting attention from media agencies. Key issues involve misinformation, authenticity, semantic fidelity, and interpretability. Most AIGC tools are opaque “black boxes,” hindering the dual demands of content accuracy and semantic alignment and creating ethical, sociotechnical, and trust dilemmas. This paper explores pathways for controllable image production in journalism’s special coverage and conducts two experiments with projects from China’s media agency: (1) Experiment 1 tests cross-platform adaptability via standardized prompts across three scenes, revealing disparities in semantic alignment, cultural specificity, and visual realism driven by training-corpus bias and platform-level filtering. (2) Experiment 2 builds a human-in-the-loop modular pipeline combining high-precision segmentation (SAM, GroundingDINO), semantic alignment (BrushNet), and style regulating (Style-LoRA, Prompt-to-Prompt), ensuring editorial fidelity through CLIP-based semantic scoring, NSFW/OCR/YOLO filtering, and verifiable content credentials. Traceable deployment preserves semantic representation. Consequently, we propose a human-AI collaboration mechanism for AIGC assisted image production in special coverage and recommend evaluating Character Identity Stability (CIS), Cultural Expression Accuracy (CEA), and User-Public Appropriateness (U-PA).

Language acquisition is vital to revealing the nature of human language intelligence and has recently emerged as a promising perspective for improving the interpretability of large language models (LLMs). However, it is ethically and practically infeasible to conduct experiments that require controlling human learners' language inputs. This poses challenges for the verifiability and scalability of language acquisition modeling, particularly in Chinese second language acquisition (SLA). While LLMs provide a controllable and reproducible alternative, a systematic benchmark to support phase-wise modeling and assessment is still lacking. To address these issues, we propose HSKBenchmark, the first benchmark for staged modeling and writing assessment of LLMs in Chinese SLA. The benchmark covers HSK levels 3 to 6, comprising authentic textbooks with 6.76M tokens, 16K synthetic instruction data, 30 test topics and a linguistically-grounded evaluation system. To simulate human acquisition trajectories, a curriculum-tuning framework is introduced, which trains LLMs in a progression from beginner to advanced proficiency levels. Since language production in writing is a key perspective for observing SLA development, an evaluation system is established to probe LLMs in writing, including the coverage of level-based grammar items, writing errors, lexical complexity, syntactic complexity, and holistic scoring. We also develop an HSKAgent fine-tuned on 10K compositions from Chinese second language learners to automate this evaluation system. Extensive experimental results demonstrate that HSKBenchmark not only models Chinese SLA effectively, but also serves as a reliable benchmark for dynamic writing assessment in LLMs. Our fine-tuned LLMs have writing performance on par with advanced human learners and exhibit human-like acquisition characteristics. The HSKBenchmark, HSKAgent, and checkpoints serve as foundational tools and resources, with the potential to pave the way for future research on language acquisition modeling and LLMs interpretability.

Fine-grained action evaluation in medical vision faces unique challenges due to the unavailability of comprehensive datasets, stringent precision requirements, and insufficient spatiotemporal dynamic modeling of very rapid actions. To support development and evaluation, we introduce CPREval-6k, a multi-view, multi-label medical action benchmark containing 6,372 expert-annotated videos with 22 clinical labels. Using this dataset, we present GaussMedAct, a multivariate Gaussian encoding framework, to advance medical motion analysis through adaptive spatiotemporal representation learning. Multivariate Gaussian Representation projects the joint motions to a temporally scaled multi-dimensional space, and decomposes actions into adaptive 3D Gaussians that serve as tokens. These tokens preserve motion semantics through anisotropic covariance modeling while maintaining robustness to spatiotemporal noise. Hybrid Spatial Encoding, employing a Cartesian and Vector dual-stream strategy, effectively utilizes skeletal information in the form of joint and bone features. The proposed method achieves 92.1% Top-1 accuracy with real-time inference on the benchmark, outperforming the baseline by +5.9% accuracy with only 10% FLOPs. Cross-dataset experiments confirm the superiority of our method in robustness.

We introduce PandemIQ Llama, a domain-adapted large language model (LLM) designed specifically for pandemic intelligence applications. Building on the pre-trained Llama-3.1-8B model, we conducted continuous training using our curated Pandemic Corpus. This dataset was assembled from authoritative public health sources, scientific literature, and specialized knowledge repositories, comprising 508,924 documents totaling 5.8 billion tokens, which is the largest pandemic domain specific data cohort for LLM training. Benefited from our curated large data cohorts and through continuous training leveraging extensive computational resources, the developed PandemIQ Llama model can extract critical domain knowledge on pandemic, which is typically underrepresented in general-purpose language models, To evaluate its performance, we conducted comprehensive comparison of PandemIQ Llama with both prompt-engineered and task-specific fine-tuned baseline models using two tasks: the Biomedical Alert News Question Answering task (1,508 reports with 30 expert-generated questions each) and the Disease Event Type Classification benchmark (4,500 news snippets across eight disease categories). PandemIQ Llama demonstrated substantial improvements over strong baseline models, achieving performance gains ranging from 3.8% to 10.97%. These results suggest that PandemIQ Llama could significantly enhance public health surveillance and analysis capabilities. In addition, our result also suggests that the LLMs can perform better with continuous training than fine-tuning on domain specific tasks. Social Impact: The BEACON platform, powered by our model, launched and now serves over 100 government and multilateral public health organizations and users across 154 countries. Analytics from the platform is being integrated into the Epidemic Intelligence from Open Sources system run by the World Health Organization. This integration will provide public health decision-makers with a powerful LLM-based tool for pandemic surveillance.

People who stutter (PWS) face systemic exclusion in today’s voice-driven society, where access to voice assistants, authentication systems, and remote work tools increasingly depends on fluent speech. Current automatic speech recognition (ASR) systems, trained predominantly on fluent speech, fail to serve millions of PWS worldwide. We present STEAMROLLER, a real time system that transforms stuttered speech into fluent output through a novel multi-stage, multi-agent AI pipeline. Our approach addresses three critical technical challenges: (1) the difficulty of direct speech to speech conversion for disfluent input, (2) semantic distortions introduced during ASR transcription of stuttered speech, and (3) latency constraints for real time communication. STEAMROLLER employs a three stage architecture comprising ASR transcription, multi-agent text repair, and speech synthesis, where our core innovation lies in a collaborative multi-agent framework that iteratively refines transcripts while preserving semantic intent. Experiments on the FluencyBank dataset and a user study demonstrates clear word error rate (WER) reduction and strong user satisfaction. Beyond immediate accessibility benefits, fine tuning ASR on STEAMROLLER repaired speech further yields additional WER improvements, creating a pathway toward inclusive AI ecosystems.

Wildfire risk prediction remains a critical yet challenging task due to the complex interactions among fuel conditions, meteorology, topography, and human activity. Despite growing interest in data-driven approaches, publicly available benchmark datasets that support long-term temporal modeling, large-scale spatial coverage, and multimodal drivers remain scarce. To address this gap, we present a 25-year, daily-resolution wildfire dataset covering 240 million hectares across British Columbia and surrounding regions. The dataset includes 38 covariates, encompassing active fire detections, weather variables, fuel conditions, terrain features, and anthropogenic factors. Using this benchmark, we evaluate a diverse set of time-series forecasting models, including CNN-based, linear-based, Transformer-based, and Mamba-based architectures. We also investigate effectiveness of position embedding and the relative importance of different fire-driving factors.

Modern slavery affects millions of people worldwide, and regulatory frameworks such as Modern Slavery Acts now require companies to publish detailed disclosures. However, these statements are often vague and inconsistent, making manual review time-consuming and difficult to scale. While NLP offers a promising path forward, high-stakes compliance tasks require more than accurate classification: they demand transparent, rule-aligned outputs that legal experts can verify. Existing applications of large language models (LLMs) often reduce complex regulatory assessments to binary decisions, lacking the necessary structure for robust legal scrutiny. We argue that compliance verification is fundamentally a rule-matching problem: it requires evaluating whether textual statements adhere to well-defined regulatory rules. To this end, we propose a novel framework that harnesses AI for rule-level compliance verification while preserving expert oversight. At its core is the Compliance Alignment Judge (CA-Judge), which evaluates model-generated justifications based on their fidelity to statutory requirements. Using this feedback, we train the Compliance Alignment LLM (CALLM), a model that produces rule-consistent, human-verifiable outputs. CALLM improves predictive performance and generates outputs that are both transparent and legally grounded, offering a more verifiable and actionable solution for real-world compliance analysis.