Pengyu Qi

Kunsheng Tang, Wenbo Zhou, Jie Zhang et al.

Large language models (LLMs) have exhibited remarkable capabilities in natural language generation, but they have also been observed to magnify societal biases, particularly those related to gender. In response to this issue, several benchmarks have been proposed to assess gender bias in LLMs. However, these benchmarks often lack practical flexibility or inadvertently introduce biases. To address these shortcomings, we introduce GenderCARE, a comprehensive framework that encompasses innovative Criteria, bias Assessment, Reduction techniques, and Evaluation metrics for quantifying and mitigating gender bias in LLMs. To begin, we establish pioneering criteria for gender equality benchmarks, spanning dimensions such as inclusivity, diversity, explainability, objectivity, robustness, and realisticity. Guided by these criteria, we construct GenderPair, a novel pair-based benchmark designed to assess gender bias in LLMs comprehensively. Our benchmark provides standardized and realistic evaluations, including previously overlooked gender groups such as transgender and non-binary individuals. Furthermore, we develop effective debiasing techniques that incorporate counterfactual data augmentation and specialized fine-tuning strategies to reduce gender bias in LLMs without compromising their overall performance. Extensive experiments demonstrate a significant reduction in various gender bias benchmarks, with reductions peaking at over 90% and averaging above 35% across 17 different LLMs. Importantly, these reductions come with minimal variability in mainstream language tasks, remaining below 2%. By offering a realistic assessment and tailored reduction of gender biases, we hope that our GenderCARE can represent a significant step towards achieving fairness and equity in LLMs. More details are available at https://github.com/kstanghere/GenderCARE-ccs24.

Shixiong Yu, Yuxiang Liu, Zongsheng Hu et al. · allen-ai

Dose verification based on proton-induced positron emitters is a promising quality assurance tool and may leverage the strength of artificial intelligence. To move a step closer towards practical application, the sensitivity analysis of two factors needs to be performed: biological washout and depth selection. selection. A bi-directional recurrent neural network (RNN) model was developed. The training dataset was generated based upon a CT image-based phantom (abdomen region) and multiple beam energies/pathways, using Monte-Carlo simulation (1 mm spatial resolution, no biological washout). For the modeling of biological washout, a simplified analytical model was applied to change raw activity profiles over a period of 5 minutes, incorporating both physical decay and biological washout. For the study of depth selection (a challenge linked to multi field/angle irradiation), truncations were applied at different window lengths (100, 125, 150 mm) to raw activity profiles. Finally, the performance of a worst-case scenario was examined by combining both factors (depth selection: 125 mm, biological washout: 5 mins). The accuracy was quantitatively evaluated in terms of range uncertainty, mean absolute error (MAE) and mean relative errors (MRE). Our proposed AI framework shows good immunity to the perturbation associated with two factors. The detection of proton-induced positron emitters, combined with machine learning, has great potential to implement online patient-specific verification in proton therapy.

Peigui Qi, Kunsheng Tang, Wenbo Zhou et al.

Text-to-image models have shown remarkable capabilities in generating high-quality images from natural language descriptions. However, these models are highly vulnerable to adversarial prompts, which can bypass safety measures and produce harmful content. Despite various defensive strategies, achieving robustness against attacks while maintaining practical utility in real-world applications remains a significant challenge. To address this issue, we first conduct an empirical study of the text encoder in the Stable Diffusion (SD) model, which is a widely used and representative text-to-image model. Our findings reveal that the [EOS] token acts as a semantic aggregator, exhibiting distinct distributional patterns between benign and adversarial prompts in its embedding space. Building on this insight, we introduce SafeGuider, a two-step framework designed for robust safety control without compromising generation quality. SafeGuider combines an embedding-level recognition model with a safety-aware feature erasure beam search algorithm. This integration enables the framework to maintain high-quality image generation for benign prompts while ensuring robust defense against both in-domain and out-of-domain attacks. SafeGuider demonstrates exceptional effectiveness in minimizing attack success rates, achieving a maximum rate of only 5.48\% across various attack scenarios. Moreover, instead of refusing to generate or producing black images for unsafe prompts, SafeGuider generates safe and meaningful images, enhancing its practical utility. In addition, SafeGuider is not limited to the SD model and can be effectively applied to other text-to-image models, such as the Flux model, demonstrating its versatility and adaptability across different architectures. We hope that SafeGuider can shed some light on the practical deployment of secure text-to-image systems.