Hanan Salam

Abu Sufian, Cosimo Distante, Marco Leo et al.

Text-to-image (T2I) generative models are largely used in AI-powered real-world applications and value creation. However, their strategic deployment raises critical concerns for responsible AI management, particularly regarding the reproduction and amplification of race- and gender-related stereotypes that can undermine organizational ethics. In this work, we investigate whether such societal biases are systematically encoded within the pretrained latent spaces of state-of-the-art T2I models. We conduct an empirical study across the five most popular open-source models, using ten neutral, profession-related prompts to generate 100 images per profession, resulting in a dataset of 5,000 images evaluated by diverse human assessors representing different races and genders. We demonstrate that all five models encode and amplify pronounced societal skew: caregiving and nursing roles are consistently feminized, while high-status professions such as corporate CEO, politician, doctor, and lawyer are overwhelmingly represented by males and mostly White individuals. We further identify model-specific patterns, such as QWEN-Image's near-exclusive focus on East Asian outputs, Kandinsky's dominance of White individuals, and SDXL's comparatively broader but still biased distributions. These results provide critical insights for AI project managers and practitioners, enabling them to select equitable AI models and customized prompts that generate images in alignment with the principles of responsible AI. We conclude by discussing the risks of these biases and proposing actionable strategies for bias mitigation in building responsible GenAI systems. The code and Data Repository: https://github.com/Sufianlab/T2IBias

Hanjun Luo, Zhimu Huang, Haoyu Huang et al.

Text-to-Image (T2I) generative models have revolutionized content creation, yet they inherently risk amplifying societal biases. While sociological research provides systematic classifications of bias, existing T2I benchmarks largely conflate these nuances or focus narrowly on occupational stereotypes, leaving the multi-dimensional nature of generative bias inadequately measured. In this paper, we introduce BiasIG, a unified benchmark that quantifies social biases across a curated dataset of 47,040 prompts. Grounded in sociological and machine ethics frameworks, BiasIG disentangles biases across 4 dimensions to enable fine-grained diagnosis. To facilitate scalable and reliable evaluation, we propose a fully automated pipeline powered by a fine-tuned multi-modal large language model, achieving high alignment accuracy comparable to human experts. Extensive experiments on 8 T2I models and 3 debiasing methods not only validate BiasIG as a robust diagnostic tool, but also reveal critical insights: interventions on protected attributes often trigger unintended confounding effects on unrelated demographics, and debiasing methods exhibit a persistent tendency toward discrimination rather than mere ignorance. Our work advocates for a precise, taxonomy-driven approach to fairness in AIGC, providing a theoretical framework for using BiasIG's metrics as feedback signals in future closed-loop mitigation. The benchmark is openly available at https://github.com/Astarojth/BiasIG.

Hanan Salam, Oya Celiktutan, Hatice Gunes et al.

An integral part of seamless human-human communication is engagement, the process by which two or more participants establish, maintain, and end their perceived connection. Therefore, to develop successful human-centered human-machine interaction applications, automatic engagement inference is one of the tasks required to achieve engaging interactions between humans and machines, and to make machines attuned to their users, hence enhancing user satisfaction and technology acceptance. Several factors contribute to engagement state inference, which include the interaction context and interactants' behaviours and identity. Indeed, engagement is a multi-faceted and multi-modal construct that requires high accuracy in the analysis and interpretation of contextual, verbal and non-verbal cues. Thus, the development of an automated and intelligent system that accomplishes this task has been proven to be challenging so far. This paper presents a comprehensive survey on previous work in engagement inference for human-machine interaction, entailing interdisciplinary definition, engagement components and factors, publicly available datasets, ground truth assessment, and most commonly used features and methods, serving as a guide for the development of future human-machine interaction interfaces with reliable context-aware engagement inference capability. An in-depth review across embodied and disembodied interaction modes, and an emphasis on the interaction context of which engagement perception modules are integrated sets apart the presented survey from existing surveys.

Jialin Li, Maha Elgarf, Alia Waleed et al.

In personalized machine learning, the aim of personalization is to train a model that caters to a specific individual or group of individuals by optimizing one or more performance metrics and adhering to specific constraints. In this paper, we discuss the need for personalization in affective computing and present the first survey of existing approaches for personalization in affective computing. Our review spans training techniques and objectives towards the personalization of affective computing models across various interaction modes and contexts. We develop a taxonomy that clusters existing approaches into Data-level and Model-level approaches. Across the Data-Level and Model-Level broad categories, we group existing approaches into seven sub-categories: (1) User-Specific Models, (2) Group-Specific Models, (3) Weighting-Based Approaches, (4) Feature Augmentation, (5) Generative-Based Models which fall into the Data-Level approaches, (6) Fine-Tuning Approaches, and (7) Multitask Learning Approaches falling under the model-level approaches. We provide a problem formulation for personalized affective computing, and to each of the identified sub-categories. Additionally, we provide a statistical analysis of the surveyed literature, analyzing the prevalence of different affective computing tasks, interaction modes (i.e. Human-Computer Interaction (HCI), Human-Human interaction (HHI), Human-Robot Interaction (HRI)), interaction contexts (e.g. educative, social, gaming, etc.), and the level of personalization among the surveyed works. Based on our analysis, we provide a road-map for researchers interested in exploring this direction.

Hanjun Luo, Zhimu Huang, Sylvia Chung et al.

Text-to-image (T2I) systems increasingly rely on upstream prompters, either humans or multimodal large language models (MLLMs), to translate user intent into detailed prompts. Yet current benchmarks fix the prompt and only evaluate T2I models, leaving the prompting proficiency of this upstream component entirely unmeasured. We introduce AtelierEval, the first unified benchmark that quantifies prompting proficiency across 360 expert-crafted tasks. Grounded in a cognitive view, it spans three task categories and instantiates tasks using a taxonomy of real-world challenges, with a dual interface for both humans and MLLMs. To enable scalable and reliable evaluation, we propose AtelierJudge, a skill-based, memory-augmented agentic evaluator. It produces subjective and objective scores for prompt-image pairs, achieving a Spearman correlation of 0.79 with human experts, approaching human performance. Extensive experiments benchmark 8 MLLMs against 48 human users across 4 T2I backends, validate AtelierEval as a robust diagnostic tool, and reveal the superiority of mimicry over planning, advocating for an image-augmented direction for future prompters. Our work is released to support future research.

Hanjun Luo, Shenyu Dai, Chiming Ni et al.

Despite the rapid advancement of LLM-based agents, the reliable evaluation of their safety and security remains a significant challenge. Existing rule-based or LLM-based evaluators often miss dangers in agents' step-by-step actions, overlook subtle meanings, fail to see how small issues compound, and get confused by unclear safety or security rules. To overcome this evaluation crisis, we introduce AgentAuditor, a universal, training-free, memory-augmented reasoning framework that empowers LLM evaluators to emulate human expert evaluators. AgentAuditor constructs an experiential memory by having an LLM adaptively extract structured semantic features (e.g., scenario, risk, behavior) and generate associated chain-of-thought reasoning traces for past interactions. A multi-stage, context-aware retrieval-augmented generation process then dynamically retrieves the most relevant reasoning experiences to guide the LLM evaluator's assessment of new cases. Moreover, we developed ASSEBench, the first benchmark designed to check how well LLM-based evaluators can spot both safety risks and security threats. ASSEBench comprises 2293 meticulously annotated interaction records, covering 15 risk types across 29 application scenarios. A key feature of ASSEBench is its nuanced approach to ambiguous risk situations, employing "Strict" and "Lenient" judgment standards. Experiments demonstrate that AgentAuditor not only consistently improves the evaluation performance of LLMs across all benchmarks but also sets a new state-of-the-art in LLM-as-a-judge for agent safety and security, achieving human-level accuracy. Our work is openly accessible at https://github.com/Astarojth/AgentAuditor.

Hanjun Luo, Chiming Ni, Jiaheng Wen et al.

LLM-powered coding agents are reshaping the development paradigm. However, existing evaluation systems, neither traditional tests for humans nor benchmarks for LLMs, fail to capture this shift. They remain focused on well-defined algorithmic problems, which excludes problems where success depends on human-AI collaboration. Such collaborative problems not only require human reasoning to interpret complex contexts and guide solution strategies, but also demand AI efficiency for implementation. To bridge this gap, we introduce HAI-Eval, a unified benchmark designed to measure the synergy of human-AI partnership in coding. HAI-Eval's core innovation is its "Collaboration-Necessary" problem templates, which are intractable for both standalone LLMs and unaided humans, but solvable through effective collaboration. Specifically, HAI-Eval uses 45 templates to dynamically create tasks. It also provides a standardized IDE for human participants and a reproducible toolkit with 450 task instances for LLMs, ensuring an ecologically valid evaluation. We conduct a within-subject study with 45 participants and benchmark their performance against 5 state-of-the-art LLMs under 4 different levels of human intervention. Results show that standalone LLMs and unaided participants achieve poor pass rates (0.67% and 18.89%), human-AI collaboration significantly improves performance to 31.11%. Our analysis reveals an emerging co-reasoning partnership. This finding challenges the traditional human-tool hierarchy by showing that strategic breakthroughs can originate from either humans or AI. HAI-Eval establishes not only a challenging benchmark for next-generation coding agents but also a grounded, scalable framework for assessing core developer competencies in the AI era. Our benchmark and interactive demo will be openly accessible.

Himanshi Lalwani, Hanan Salam

College students often face academic challenges that hamper their productivity and well-being. Although self-help books and productivity apps are popular, they often fall short. Books provide generalized, non-interactive guidance, and apps are not inherently educational and can hinder the development of key organizational skills. Traditional productivity coaching offers personalized support, but is resource-intensive and difficult to scale. In this study, we present a proof-of-concept for a socially assistive robot (SAR) as an educational coach and a potential solution to the limitations of existing productivity tools and coaching approaches. The SAR delivers six different lessons on time management and task prioritization. Users interact via a chat interface, while the SAR responds through speech (with a toggle option). An integrated dashboard monitors progress, mood, engagement, confidence per lesson, and time spent per lesson. It also offers personalized productivity insights to foster reflection and self-awareness. We evaluated the system with 15 college students, achieving a System Usability Score of 79.2 and high ratings for overall experience and engagement. Our findings suggest that SAR-based productivity coaching can offer an effective and scalable solution to improve productivity among college students.

Sneheel Sarangi, Maha Elgarf, Hanan Salam

Theory of Mind (ToM) is the ability to understand and reflect on the mental states of others. Although this capability is crucial for human interaction, testing on Large Language Models (LLMs) reveals that they possess only a rudimentary understanding of it. Although the most capable closed-source LLMs have come close to human performance on some ToM tasks, they still perform poorly on complex variations of the task that involve more structured reasoning. In this work, we utilize the concept of "pretend-play", or ``Simulation Theory'' from cognitive psychology to propose ``Decompose-ToM'': an LLM-based inference algorithm that improves model performance on complex ToM tasks. We recursively simulate user perspectives and decompose the ToM task into a simpler set of functions: subject identification, question-reframing, world model updation, and knowledge availability. We test the algorithm on higher-order ToM tasks and a task testing for ToM capabilities in a conversational setting, demonstrating that our approach shows significant improvement across models compared to baseline methods while requiring minimal prompt tuning across tasks and no additional model training.

Keya Shah, Himanshi Lalwani, Hanan Salam

College students face well-being challenges driven by academic pressure, financial strain, and social expectations. While campus counseling and student-success programs offer support, access is often limited by stigma, waitlists, and scheduling constraints. Existing digital tools focus on emotional check-ins or chatbots and may overlook structured goal setting and aligning goals with personal values. We present GROW, a goal-centered well-being coaching system that puts values-aligned goals at the center of the student experience. GROW combines the SMART framework with principles from Acceptance and Commitment Therapy in a conversational AI coach that helps students clarify aspirations, break them into concrete steps, and reflect on progress. The system links action plans with Google Calendar, sends reminders, and provides a dashboard that shows progress and engagement. We evaluated GROW through interviews with clinical psychologists, student-success staff, and faculty, followed by a one-week deployment with 30 undergraduates. Findings offer design implications for interactive systems that support engagement, accountability, and sense of purpose in higher education.

Hasibur Rahman, Hanan Salam

Although large language models (LLMs) are increasingly implicated in interpersonal and societal decision-making, their ability to navigate explicit conflicts between legitimately different cultural value systems remains largely unexamined. Existing benchmarks predominantly target cultural knowledge (CulturalBench), value prediction (WorldValuesBench), or single-axis bias diagnostics (CDEval); none evaluate how LLMs adjudicate when multiple culturally grounded values directly clash. We address this gap with CCD-Bench, a benchmark that assesses LLM decision-making under cross-cultural value conflict. CCD-Bench comprises 2,182 open-ended dilemmas spanning seven domains, each paired with ten anonymized response options corresponding to the ten GLOBE cultural clusters. These dilemmas are presented using a stratified Latin square to mitigate ordering effects. We evaluate 17 non-reasoning LLMs. Models disproportionately prefer Nordic Europe (mean 20.2 percent) and Germanic Europe (12.4 percent), while options for Eastern Europe and the Middle East and North Africa are underrepresented (5.6 to 5.8 percent). Although 87.9 percent of rationales reference multiple GLOBE dimensions, this pluralism is superficial: models recombine Future Orientation and Performance Orientation, and rarely ground choices in Assertiveness or Gender Egalitarianism (both under 3 percent). Ordering effects are negligible (Cramer's V less than 0.10), and symmetrized KL divergence shows clustering by developer lineage rather than geography. These patterns suggest that current alignment pipelines promote a consensus-oriented worldview that underserves scenarios demanding power negotiation, rights-based reasoning, or gender-aware analysis. CCD-Bench shifts evaluation beyond isolated bias detection toward pluralistic decision making and highlights the need for alignment strategies that substantively engage diverse worldviews.

Sneheel Sarangi, Hanan Salam

Recent advancements in large language models (LLMs) have demonstrated emergent capabilities in complex reasoning, largely spurred by rule-based Reinforcement Learning (RL) techniques applied during the post-training. This has raised the question of whether similar methods can instill more nuanced, human-like social intelligence, such as a Theory of Mind (ToM), in LLMs. This paper investigates whether small-scale LLMs can acquire a robust and generalizable ToM capability through RL with verifiable rewards (RLVR). We conduct a systematic evaluation by training models on various combinations of prominent ToM datasets (HiToM, ExploreToM, FANToM) and testing for generalization on held-out datasets (e.g., OpenToM). Our findings indicate that small LLMs struggle to develop a generic ToM capability. While performance on in-distribution tasks improves, this capability fails to transfer to unseen ToM tasks with different characteristics. Furthermore, we demonstrate that prolonged RL training leads to models ``hacking'' the statistical patterns of the training datasets, resulting in significant performance gains on in-domain data but no change, or degradation of performance on out-of-distribution tasks. This suggests the learned behavior is a form of narrow overfitting rather than the acquisition of a true, abstract ToM capability.

Benjamin Dadoun, Soufiane Hayou, Hanan Salam et al.

Deep neural networks are known to suffer from exploding or vanishing gradients as depth increases, a phenomenon closely tied to the spectral behavior of the input-output Jacobian. Prior work has identified critical initialization schemes that ensure Jacobian stability, but these analyses are typically restricted to fully connected networks with i.i.d. weights. In this work, we go significantly beyond these limitations: we establish a general stability theorem for deep neural networks that accommodates sparsity (such as that introduced by pruning) and non-i.i.d., weakly correlated weights (e.g. induced by training). Our results rely on recent advances in random matrix theory, and provide rigorous guarantees for spectral stability in a much broader class of network models. This extends the theoretical foundation for initialization schemes in modern neural networks with structured and dependent randomness.

Munachiso Nwadike, Jialin Li, Hanan Salam

In the field of emotion recognition and Human-Machine Interaction (HMI), personalised approaches have exhibited their efficacy in capturing individual-specific characteristics and enhancing affective prediction accuracy. However, personalisation techniques often face the challenge of limited data for target individuals. This paper presents our work on an enhanced personalisation strategy, that leverages data augmentation to develop tailored models for continuous valence and arousal prediction. Our proposed approach, Distance Weighting Augmentation (DWA), employs a weighting-based augmentation method that expands a target individual's dataset, leveraging distance metrics to identify similar samples at the segment-level. Experimental results on the MuSe-Personalisation 2023 Challenge dataset demonstrate that our method significantly improves the performance of features sets which have low baseline performance, on the test set. This improvement in poor-performing features comes without sacrificing performance on high-performing features. In particular, our method achieves a maximum combined testing CCC of 0.78, compared to the reported baseline score of 0.76 (reproduced at 0.72). It also achieved a peak arousal and valence scores of 0.81 and 0.76, compared to reproduced baseline scores of 0.76 and 0.67 respectively. Through this work, we make significant contributions to the advancement of personalised affective computing models, enhancing the practicality and adaptability of data-level personalisation in real world contexts.

Valentin Roche, Jean-Philippe Robert, Hanan Salam

Social media can be an important source of information facilitating the detection of new safety signals in pharmacovigilance. Various approaches have investigated the analysis of social media data using AI such as NLP techniques for detecting adverse drug events. Existing approaches have focused on the extraction and identification of Adverse Drug Reactions, Drug-Drug Interactions and drug misuse. However, non of the works tackled the detection of potential safety signals by taking into account the evolution in time of relevant indicators. Moreover, despite the success of deep learning in various healthcare applications, it was not explored for this task. We propose an AI-based approach for the detection of potential pharmaceutical safety signals from patients' reviews that can be used as part of the pharmacovigilance surveillance process to flag the necessity of an in-depth pharmacovigilance investigation. We focus on the Levothyrox case in France which triggered huge attention from the media following the change of the medication formula, leading to an increase in the frequency of adverse drug reactions normally reported by patients. Our approach is two-fold. (1) We investigate various NLP-based indicators extracted from patients' reviews including words and n-grams frequency, semantic similarity, Adverse Drug Reactions mentions, and sentiment analysis. (2) We propose a deep learning architecture, named Word Cloud Convolutional Neural Network (WC-CNN) which trains a CNN on word clouds extracted from the patients comments. We study the effect of different time resolutions and different NLP pre-processing techniques on the model performance. Our results show that the proposed indicators could be used in the future to effectively detect new safety signals. The WC-CNN model trained on word clouds extracted at monthly resolution outperforms the others with an accuracy of 75%.

Hanan Salam

Engagement in Human-Machine Interaction is the process by which entities participating in the interaction establish, maintain, and end their perceived connection. It is essential to monitor the engagement state of patients in various AI-based interactive healthcare paradigms. This includes medical conditions that alter social behavior such as Autism Spectrum Disorder (ASD) or Attention-Deficit/Hyperactivity Disorder (ADHD). Engagement is a multi-faceted construct which is composed of behavioral, emotional, and mental components. Previous research has neglected this multi-faceted nature of engagement and focused on the detection of engagement level or binary engagement label. In this paper, a system is presented to distinguish these facets using contextual and relational features. This can facilitate further fine-grained analysis. Several machine learning classifiers including traditional and deep learning models are compared for this task. An F-Score of 0.74 was obtained on a balanced dataset of 22242 instances with neural network-based classification. The proposed framework shall serve as a baseline for further research on engagement facets recognition, and its integration is socially assistive robotic applications.

Muhammad Muzammel, Hanan Salam, Yann Hoffmann et al.

Depression is a common and serious mood disorder that negatively affects the patient's capacity of functioning normally in daily tasks. Speech is proven to be a vigorous tool in depression diagnosis. Research in psychiatry concentrated on performing fine-grained analysis on word-level speech components contributing to the manifestation of depression in speech and revealed significant variations at the phoneme-level in depressed speech. On the other hand, research in Machine Learning-based automatic recognition of depression from speech focused on the exploration of various acoustic features for the detection of depression and its severity level. Few have focused on incorporating phoneme-level speech components in automatic assessment systems. In this paper, we propose an Artificial Intelligence (AI) based application for clinical depression recognition and assessment from speech. We investigate the acoustic characteristics of phoneme units, specifically vowels and consonants for depression recognition via Deep Learning. We present and compare three spectrogram-based Deep Neural Network architectures, trained on phoneme consonant and vowel units and their fusion respectively. Our experiments show that the deep learned consonant-based acoustic characteristics lead to better recognition results than vowel-based ones. The fusion of vowel and consonant speech characteristics through a deep network significantly outperforms the single space networks as well as the state-of-art deep learning approaches on the DAIC-WOZ database.

Ahmed Rachid Hazourli, Amine Djeghri, Hanan Salam et al.

In this paper, we propose an approach for Facial Expressions Recognition (FER) based on a deep multi-facial patches aggregation network. Deep features are learned from facial patches using deep sub-networks and aggregated within one deep architecture for expression classification . Several problems may affect the performance of deep-learning based FER approaches, in particular, the small size of existing FER datasets which might not be sufficient to train large deep learning networks. Moreover, it is extremely time-consuming to collect and annotate a large number of facial images. To account for this, we propose two data augmentation techniques for facial expression generation to expand FER labeled training datasets. We evaluate the proposed framework on three FER datasets. Results show that the proposed approach achieves state-of-art FER deep learning approaches performance when the model is trained and tested on images from the same dataset. Moreover, the proposed data augmentation techniques improve the expression recognition rate, and thus can be a solution for training deep learning FER models using small datasets. The accuracy degrades significantly when testing for dataset bias.