@MASTERSTHESIS{ 2024:125589402,
 	title = {SDXL debiasing mechanism : a fairness improvement to image generation through diffusion kernels},
 	year = {2024},
 	url = "https://tede2.pucrs.br/tede2/handle/tede/11655",
 	abstract = "This thesis explores the pressing issue of unwanted biases in text-to-image synthesis models, focusing particularly on debiasing mechanisms within Stable Diffusion models, notably the Stable Diffusoin eXtra Large (SDXL) release. By proposing an approach centered on mod ifying Classifier-Free Guidance, the study aims to steer image generation away from harmful societal biases while preserving image fidelity and full capabilities of SDXL model. Leveraging insights from previous debiasing attempts, this research pioneers the usage of inner workings of Diffusion Models to debias harmful concepts at it?s core. Thereby, it contributes to a more
 equitable and responsible usage of generative AI systems. Results demonstrate that the proposed method effectively mitigates biases without compromising image quality, at the cost of an increase in inference time. Despite current limitations, this work represents a crucial step towards fairer image generation and underscores the importance of ethical considerations in AI development.",
 	publisher = {Pontif?cia Universidade Cat?lica do Rio Grande do Sul},
 	scholl = {Programa de P?s-Gradua??o em Ci?ncia da Computa??o},
 	note = {Escola Polit?cnica}
}