草庐IT

Empowering

全部标签

Empowering Low-Light Image Enhancer through Customized Learnable Priors 论文阅读笔记

中科大、西安交大、南开大学发表在ICCV2023的论文,作者里有李重仪老师和中科大的JieHuang(ECCV2022的FECCVPR2022的ENC和CVPR2023的ERL的一作)喔,看来可能是和JieHuang同一个课题组的,而且同样代码是开源的,我很喜欢。文章利用了MAE的encoder来做一些事情,提出了一个叫customizedunfoldingenhancer(CUE)的方法。从MAE中学了illuminationprior和noiseprior两个先验,用到了retinex模型中。流程如下图所示:文章用的是如下的常规retinex公式:目标是最小化如下表达式:把限制项(2b)放

【开源AI大模型】WizardCoder: Empowering Code Large Language Models with Evol-Instruct

文章目录WizardCoder:EmpoweringCodeLargeLanguageModelswithEvol-InstructNewsComparingWizardCoderwiththeClosed-SourceModels.ComparingWizardCoderwiththeOpen-SourceModels.CallforFeedbacksUnofficialVideoIntroductionsContentsOnlineDemoFine-tuningInferenceEvaluationHumanEval

Empowering Long-tail Item Recommendation through Cross Decoupling Network (CDN)

EmpoweringLong-tailItemRecommendationthroughCrossDecouplingNetwork(CDN)来源:KDD'2023GoogleResearch目录EmpoweringLong-tailItemRecommendationthroughCrossDecouplingNetwork(CDN)长尾问题分析CDNItemMemorizationandGeneralizationDecoupling记忆特征(Memorizationfeatures)泛化特征(Generalizationfeatures)物品表征学习UserSampleDecouplin