Deep neural networks are prone to various bias issues, jeopardizing their applications for high-stake decision-making. Existing fairness methods typically offer a fixed accuracy-fairness trade-off, since the weight of the well-trained model is a fixed point (fairness-optimum) in the weight space. Nevertheless, more flexible accuracy-fairness trade-offs at inference time are practically desired since: 1) stakes of the same downstream task can vary for different individuals, and 2) different regions have diverse laws or regularization for fairness. If using the previous fairness methods, we have to train multiple models, each offering a specific level of accuracy-fairness trade-off. This is often computationally expensive, time-consuming, and difficult to deploy, making it less practical for real-world applications. To address this problem, we propose You Only Debias Once (YODO) to achieve in-situ flexible accuracy-fairness trade-offs at inference time, using a single model that trained only once. Instead of pursuing one individual fixed point (fairness-optimum) in the weight space, we aim to find a "line" in the weight space that connects the accuracy-optimum and fairness-optimum points using a single model. Points (models) on this line implement varying levels of accuracy-fairness trade-offs. At inference time, by manually selecting the specific position of the learned "line", our proposed method can achieve arbitrary accuracy-fairness trade-offs for different end-users and scenarios. Experimental results on tabular and image datasets show that YODO achieves flexible trade-offs between model accuracy and fairness, at ultra-low overheads. For example, if we need $100$ levels of trade-off on the \acse dataset, YODO takes $3.53$ seconds while training $100$ fixed models consumes $425$ seconds. The code is available at https://github.com/ahxt/yodo.

Machine unlearning is a process to remove specific data points from a trained model while maintaining the performance on retain data, addressing privacy or legal requirements. Despite its importance, existing unlearning evaluations tend to focus on logit-based metrics (i.e., accuracy) under small-scale scenarios. We observe that this could lead to a false sense of security in unlearning approaches under real-world scenarios. In this paper, we conduct a new comprehensive evaluation that employs representation-based evaluations of the unlearned model under large-scale scenarios to verify whether the unlearning approaches genuinely eliminate the targeted forget data from the model's representation perspective. Our analysis reveals that current state-of-the-art unlearning approaches either completely degrade the representational quality of the unlearned model or merely modify the classifier (i.e., the last layer), thereby achieving superior logit-based evaluation metrics while maintaining significant representational similarity to the original model. Furthermore, we introduce a novel unlearning evaluation setup from a transfer learning perspective, in which the forget set classes exhibit semantic similarity to downstream task classes, necessitating that feature representations diverge significantly from those of the original model. Our comprehensive benchmark not only addresses a critical gap between theoretical machine unlearning and practical scenarios, but also establishes a foundation to inspire future research directions in developing genuinely effective unlearning methodologies.

Video-based dialogue systems, such as education assistants, have compelling application value, thereby garnering growing interest. However, the current video-based dialogue systems are limited by their reliance on a single dialogue type, which hinders their versatility in practical applications across a range of scenarios, including question-answering, emotional dialog, etc. In this paper, we identify this challenge as how to generate video-driven multilingual mixed-type dialogues. To mitigate this challenge, we propose a novel task and create a human-to-human video-driven multilingual mixed-type dialogue corpus, termed KwaiChat, containing a total of 93,209 videos and 246,080 dialogues, across 4 dialogue types, 30 domains, 4 languages, and 13 topics. Additionally, we establish baseline models on KwaiChat. An extensive analysis of 7 distinct LLMs on KwaiChat reveals that GPT-4o achieves the best performance but still cannot perform well in this situation even with the help of in-context learning and fine-tuning, which indicates that the task is not trivial and needs further research.

The ability of Natural Language Processing (NLP) methods to categorize text into multiple classes has motivated their use in online content moderation tasks, such as hate speech and fake news detection. However, there is limited understanding of how or why these methods make such decisions, or why certain content is moderated in the first place. To investigate the hidden mechanisms behind content moderation, we explore multiple directions: 1) training classifiers to reverse-engineer content moderation decisions across countries; 2) explaining content moderation decisions by analyzing Shapley values and LLM-guided explanations. Our primary focus is on content moderation decisions made across countries, using pre-existing corpora sampled from the Twitter Stream Grab. Our experiments reveal interesting patterns in censored posts, both across countries and over time. Through human evaluations of LLM-generated explanations across three LLMs, we assess the effectiveness of using LLMs in content moderation. Finally, we discuss potential future directions, as well as the limitations and ethical considerations of this work. Our code and data are available at https://github.com/causalNLP/censorship

Synthetic data has garnered attention as a Privacy Enhancing Technology (PET) in sectors such as healthcare and finance. When using synthetic data in practical applications, it is important to provide protection guarantees. In the literature, two family of approaches are proposed for tabular data: on the one hand, Similarity-based methods aim at finding the level of similarity between training and synthetic data. Indeed, a privacy breach can occur if the generated data is consistently too similar or even identical to the train data. On the other hand, Attack-based methods conduce deliberate attacks on synthetic datasets. The success rates of these attacks reveal how secure the synthetic datasets are. In this paper, we introduce a contrastive method that improves privacy assessment of synthetic datasets by embedding the data in a more representative space. This overcomes obstacles surrounding the multitude of data types and attributes. It also makes the use of intuitive distance metrics possible for similarity measurements and as an attack vector. In a series of experiments with publicly available datasets, we compare the performances of similarity-based and attack-based methods, both with and without use of the contrastive learning-based embeddings. Our results show that relatively efficient, easy to implement privacy metrics can perform equally well as more advanced metrics explicitly modeling conditions for privacy referred to by the GDPR.

Small business owners (SBOs) often lack the resources and design experience needed to produce high-quality advertisements. To address this, we developed ACAI (AI Co-Creation for Advertising and Inspiration), an GenAI-powered multimodal advertisement creation tool, and conducted a user study with 16 SBOs in London to explore their perceptions of and interactions with ACAI in advertisement creation. Our findings reveal that structured inputs enhance user agency and control while improving AI outputs by facilitating better brand alignment, enhancing AI transparency, and offering scaffolding that assists novice designers, such as SBOs, in formulating prompts. We also found that ACAI's multimodal interface bridges the design skill gap for SBOs with a clear advertisement vision, but who lack the design jargon necessary for effective prompting. Building on our findings, we propose three capabilities: contextual intelligence, adaptive interactions, and data management, with corresponding design recommendations to advance the co-creative attributes of AI-mediated design tools.

Dynamic Treatment Regimes (DTRs) provide a systematic approach for making sequential treatment decisions that adapt to individual patient characteristics, particularly in clinical contexts where survival outcomes are of interest. Censoring-Aware Tree-Based Reinforcement Learning (CA-TRL) is a novel framework to address the complexities associated with censored data when estimating optimal DTRs. We explore ways to learn effective DTRs, from observational data. By enhancing traditional tree-based reinforcement learning methods with augmented inverse probability weighting (AIPW) and censoring-aware modifications, CA-TRL delivers robust and interpretable treatment strategies. We demonstrate its effectiveness through extensive simulations and real-world applications using the SANAD epilepsy dataset, where it outperformed the recently proposed ASCL method in key metrics such as restricted mean survival time (RMST) and decision-making accuracy. This work represents a step forward in advancing personalized and data-driven treatment strategies across diverse healthcare settings.

Malicious content generated by large language models (LLMs) can pose varying degrees of harm. Although existing LLM-based moderators can detect harmful content, they struggle to assess risk levels and may miss lower-risk outputs. Accurate risk assessment allows platforms with different safety thresholds to tailor content filtering and rejection. In this paper, we introduce per-topic severity rubrics for 11 harmful topics and build BingoGuard, an LLM-based moderation system designed to predict both binary safety labels and severity levels. To address the lack of annotations on levels of severity, we propose a scalable generate-then-filter framework that first generates responses across different severity levels and then filters out low-quality responses. Using this framework, we create BingoGuardTrain, a training dataset with 54,897 examples covering a variety of topics, response severity, styles, and BingoGuardTest, a test set with 988 examples explicitly labeled based on our severity rubrics that enables fine-grained analysis on model behaviors on different severity levels. Our BingoGuard-8B, trained on BingoGuardTrain, achieves the state-of-the-art performance on several moderation benchmarks, including WildGuardTest and HarmBench, as well as BingoGuardTest, outperforming best public models, WildGuard, by 4.3\%. Our analysis demonstrates that incorporating severity levels into training significantly enhances detection performance and enables the model to effectively gauge the severity of harmful responses.

Detecting toxic language including sexism, harassment and abusive behaviour, remains a critical challenge, particularly in its subtle and context-dependent forms. Existing approaches largely focus on isolated message-level classification, overlooking toxicity that emerges across conversational contexts. To promote and enable future research in this direction, we introduce SafeSpeech, a comprehensive platform for toxic content detection and analysis that bridges message-level and conversation-level insights. The platform integrates fine-tuned classifiers and large language models (LLMs) to enable multi-granularity detection, toxic-aware conversation summarization, and persona profiling. SafeSpeech also incorporates explainability mechanisms, such as perplexity gain analysis, to highlight the linguistic elements driving predictions. Evaluations on benchmark datasets, including EDOS, OffensEval, and HatEval, demonstrate the reproduction of state-of-the-art performance across multiple tasks, including fine-grained sexism detection.

Generative AI is frequently portrayed as revolutionary or even apocalyptic, prompting calls for novel regulatory approaches. This essay argues that such views are misguided. Instead, generative AI should be understood as an evolutionary step in the broader algorithmic media landscape, alongside search engines and social media. Like these platforms, generative AI centralizes information control, relies on complex algorithms to shape content, and extensively uses user data, thus perpetuating common problems: unchecked corporate power, echo chambers, and weakened traditional gatekeepers. Regulation should therefore share a consistent objective: ensuring media institutions remain trustworthy. Without trust, public discourse risks fragmenting into isolated communities dominated by comforting, tribal beliefs -- a threat intensified by generative AI's capacity to bypass gatekeepers and personalize truth. Current governance frameworks, such as the EU's AI Act and the US Executive Order 14110, emphasize reactive risk mitigation, addressing measurable threats like national security, public health, and algorithmic bias. While effective for novel technological risks, this reactive approach fails to adequately address broader issues of trust and legitimacy inherent to digital media. Proactive regulation fostering transparency, accountability, and public confidence is essential. Viewing generative AI exclusively as revolutionary risks repeating past regulatory failures that left social media and search engines insufficiently regulated. Instead, regulation must proactively shape an algorithmic media environment serving the public good, supporting quality information and robust civic discourse.