The widespread deployment of Large Language Models (LLMs) has intensified concerns about subtle social biases embedded in their outputs. Existing guardrails often fail when faced with indirect or contextually complex bias-inducing prompts. To address these limitations, we propose a unified framework for both systematic bias evaluation and targeted mitigation. Our approach introduces six novel Metamorphic Relations (MRs) that, based on metamorphic testing principles, transform direct bias-inducing inputs into semantically equivalent yet adversarially challenging variants. These transformations enable an automated method for exposing hidden model biases: when an LLM responds inconsistently or unfairly across MR-generated variants, the underlying bias becomes detectable. We further show that the same MRs can be used to generate diverse bias-inducing samples for fine-tuning, directly linking the testing process to mitigation. Using six state-of-the-art LLMs - spanning open-source and proprietary models - and a representative subset of 385 questions from the 8,978-item BiasAsker benchmark covering seven protected groups, our MRs reveal up to 14% more hidden biases compared to existing tools. Moreover, fine-tuning with both original and MR-mutated samples significantly enhances bias resiliency, increasing safe response rates from 54.7% to over 88.9% across models. These results highlight metamorphic relations as a practical mechanism for improving fairness in conversational AI.

Current evaluation of mathematical reasoning in language models relies primarily on answer accuracy, potentially masking fundamental failures in logical computation. We introduce a diagnostic framework that distinguishes genuine mathematical reasoning from superficial pattern matching through four complementary axes: forward-backward consistency, transitivity coverage, counterfactual sensitivity, and perturbation robustness. Through a case study applying this framework to Qwen3-0.6B on the MenatQA dataset, we reveal a striking disconnect between surface performance and reasoning fidelity--while the model achieves reasonable answer accuracy (70%+), it demonstrates poor backward consistency (15%), limited transitivity coverage (32.2%), and brittle sensitivity to perturbations. Our diagnostics expose reasoning failures invisible to traditional accuracy metrics, suggesting that this small model relies heavily on pattern matching rather than genuine logical computation. While our empirical findings are based on a single 600M-parameter model, the diagnostic framework itself is model-agnostic and generalizable. We release our evaluation protocols to enable the research community to assess reasoning fidelity across different model scales and architectures, moving beyond surface-level accuracy toward verifiable mathematical reasoning.

Heatwaves are intensifying worldwide and are among the deadliest weather disasters. The burden falls disproportionately on marginalized populations and the Global South, where under-resourced health systems, exposure to urban heat islands, and the lack of adaptive infrastructure amplify risks. Yet current numerical weather prediction models often fail to capture micro-scale extremes, leaving the most vulnerable excluded from timely early warnings. We present a Graph Neural Network framework for localized, high-resolution temperature forecasting. By leveraging spatial learning and efficient computation, our approach generates forecasts at multiple horizons, up to 48 hours. For Southwestern Ontario, Canada, the model captures temperature patterns with a mean MAE of 1.93$^{\circ}$C across 1-48h forecasts and MAE@48h of 2.93$^{\circ}$C, evaluated using 24h input windows on the largest region. While demonstrated here in a data-rich context, this work lays the foundation for transfer learning approaches that could enable localized, equitable forecasts in data-limited regions of the Global South.

Abstract--The widespread adoption of complex machine learning models in high-stakes domains has brought the "black-box" problem to the forefront of responsible AI research. This paper aims at addressing this issue by improving the Explainable Boosting Machine (EBM), a state-of-the-art glassbox model that delivers both high accuracy and complete transparency. The paper outlines three distinct enhancement methodologies: targeted hyperparameter optimization with Bayesian methods, the implementation of a custom multi-objective function for fairness for hyperparameter optimization, and a novel self-supervised pre-training pipeline for cold-start scenarios. All three methodologies are evaluated across standard benchmark datasets, including the Adult Income, Credit Card Fraud Detection, and UCI Heart Disease datasets. The analysis indicates that while the tuning process yielded marginal improvements in the primary ROC AUC metric, it led to a subtle but important shift in the model's decision-making behavior, demonstrating the value of a multi-faceted evaluation beyond a single performance score. This work is positioned as a critical step toward developing machine learning systems that are not only accurate but also robust, equitable, and transparent, meeting the growing demands of regulatory and ethical compliance. A. The Black-Box Problem in High-Stakes Domains The remarkable surge in the performance of machine learning models has led to their pervasive adoption across a multitude of domains, from retail and finance to medicine and judicial systems. Complex, high-performing models, such as deep neural networks and ensemble methods like Random Forest and XGBoost, have become the de facto standard for many tasks.

Background: Autonomous agents powered by Large Language Models (LLMs) are driving a paradigm shift toward an "Internet of Agents" (IoA). While offering immense potential, this vision also introduces novel and systemic risks to safety and security. Objectives: Unlike common threat-centric taxonomies, our survey provides a principled, architectural framework for engineering safe and reliable agentic systems. We aim to identify the architectural sources of vulnerabilities to establish a foundation for secure design. Methods: We perform a bottom-up deconstruction of agentic systems, treating each component as a dual-use interface. The analysis spans three levels of complexity: the foundational Single Agent, the collaborative Multi-Agent System (MAS), and the visionary Interoperable Multi-Agent System (IMAS). At each level, we identify core architectural components and their inherent security risks. Results & Conclusions: Our central finding is that agentic safety is an architectural principle, not an add-on. By identifying specific vulnerabilities and deriving mitigation principles at each level of the agentic stack, this survey serves as a foundational guide for building the capable, safe, and trustworthy AI needed to realize a secure Internet of Agents.

In this thesis, we developed a comprehensive framework for sentiment analysis that takes its many aspects into account mainly for Turkish. We have also proposed several approaches specific to sentiment analysis in English only. We have accordingly made five major and three minor contributions. We generated a novel and effective feature set by combining unsupervised, semi-supervised, and supervised metrics. We then fed them as input into classical machine learning methods, and outperformed neural network models for datasets of different genres in both Turkish and English. We created a polarity lexicon with a semi-supervised domain-specific method, which has been the first approach applied for corpora in Turkish. We performed a fine morphological analysis for the sentiment classification task in Turkish by determining the polarities of morphemes. This can be adapted to other morphologically-rich or agglutinative languages as well. We have built a novel neural network architecture, which combines recurrent and recursive neural network models for English. We built novel word embeddings that exploit sentiment, syntactic, semantic, and lexical characteristics for both Turkish and English. We also redefined context windows as subclauses in modelling word representations in English. This can also be applied to other linguistic fields and natural language processing tasks. We have achieved state-of-the-art and significant results for all these original approaches. Our minor contributions include methods related to aspect-based sentiment in Turkish, parameter redefinition in the semi-supervised approach, and aspect term extraction techniques for English. This thesis can be considered the most detailed and comprehensive study made on sentiment analysis in Turkish as of July, 2020. Our work has also contributed to the opinion classification problem in English.

As deep learning models evolve, new applications and challenges are rapidly emerging. Tasks that once relied on a single modality, such as text, images, or audio, are now enriched by seamless interactions between multimodal data. These connections bridge information gaps: an image can visually materialize a text, while audio can add context to an image. Researchers have developed numerous multimodal models, but most rely on resource-intensive training across multiple modalities. Similarly, extending these models to new languages often follows the same resource-heavy training strategy. In this work, we propose a multimodal and multilingual architecture, CACARA, trained through emergent alignment learning, enabling the seamless integration of new modalities into an existing bimodal/multimodal model without requiring full retraining. This work breaks new ground by demonstrating that this emergent alignment paradigm can unlock multilingual capabilities from monolingual training. By fine-tuning the newly incorporated modality only on data aligned with the English language, our model develops support for over 100 languages without explicit multilingual pretraining or tuning of the text encoder. Such emergent multimodal and multilingual properties are gained efficiently, preserving previously learned knowledge at a training cost comparable to that of a monolingual model. Our strategy achieves up to a 14.24 percentage points improvement in R@1 audio-to-text retrieval, outperforming state-of-the-art multimodal models -- all without the heavy computational cost of retraining across every modality and language.

Data collection often reflects human decisions. In healthcare, for instance, a referral for a diagnostic test is influenced by the patient's health, their preferences, available resources, and the practitioner's recommendations. Despite the extensive literature on the informativeness of missingness, its implications on the performance of Large Language Models (LLMs) have not been studied. Through a series of experiments on data from Columbia University Medical Center, a large urban academic medical center, and MIMIC-IV, we demonstrate that patterns of missingness significantly impact zero-shot predictive performance. Notably, the explicit inclusion of missingness indicators at prompting benefits some while hurting other LLMs' zero-shot predictive performance and calibration, suggesting an inconsistent impact. The proposed aggregated analysis and theoretical insights suggest that larger models benefit from these interventions, while smaller models can be negatively impacted. The LLM paradigm risks obscuring the impact of missingness, often neglected even in conventional ML, even further. We conclude that there is a need for more transparent accounting and systematic evaluation of the impact of representing (informative) missingness on downstream performance.

With the continuous advancement of image generation technology, advanced models such as GPT-Image-1 and Qwen-Image have achieved remarkable text-to-image consistency and world knowledge However, these models still fall short in photorealistic image generation. Even on simple T2I tasks, they tend to produce " fake" images with distinct AI artifacts, often characterized by "overly smooth skin" and "oily facial sheens". To recapture the original goal of "indistinguishable-from-reality" generation, we propose RealGen, a photorealistic text-to-image framework. RealGen integrates an LLM component for prompt optimization and a diffusion model for realistic image generation. Inspired by adversarial generation, RealGen introduces a "Detector Reward" mechanism, which quantifies artifacts and assesses realism using both semantic-level and feature-level synthetic image detectors. We leverage this reward signal with the GRPO algorithm to optimize the entire generation pipeline, significantly enhancing image realism and detail. Furthermore, we propose RealBench, an automated evaluation benchmark employing Detector-Scoring and Arena-Scoring. It enables human-free photorealism assessment, yielding results that are more accurate and aligned with real user experience. Experiments demonstrate that RealGen significantly outperforms general models like GPT-Image-1 and Qwen-Image, as well as specialized photorealistic models like FLUX-Krea, in terms of realism, detail, and aesthetics. The code is available at https://github.com/yejy53/RealGen.

Synthetic personae experiments have become a prominent method in Large Language Model alignment research, yet the representativeness and ecological validity of these personae vary considerably between studies. Through a review of 63 peer-reviewed studies published between 2023 and 2025 in leading NLP and AI venues, we reveal a critical gap: task and population of interest are often underspecified in persona-based experiments, despite personalization being fundamentally dependent on these criteria. Our analysis shows substantial differences in user representation, with most studies focusing on limited sociodemographic attributes and only 35% discussing the representativeness of their LLM personae. Based on our findings, we introduce a persona transparency checklist that emphasizes representative sampling, explicit grounding in empirical data, and enhanced ecological validity. Our work provides both a comprehensive assessment of current practices and practical guidelines to improve the rigor and ecological validity of persona-based evaluations in language model alignment research.