[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"aihot-art-91680":3},{"itemId":4,"vertical":5,"category":6,"source":7,"score":8,"title":9,"summary":10,"analysis":11,"url":12,"coverUrl":13,"direction":13,"marketSignal":13,"publishedAt":14},"91680","ai","论文研究","Apple Machine Learning Research",60,"给我看例子：从图像集合中推断视觉概念","苹果提出视觉概念推断任务VICIS，要求模型从少量示例图像中推断共享概念并生成新图像，现有模型表现不佳。","苹果提出视觉概念推断任务VICIS，测试模型从图像集合中推断共享概念并应用于新输入的能力，发现当前最先进VLM表现不佳。\n· VICIS任务：给定少量共享概念的示例图像和一张查询图像，模型需生成保留概念且与查询一致的新图像。\n· 实验表明，现有视觉语言模型（VLM）难以从纯视觉上下文中推理概念，依赖文本指令而忽视视觉示例。\n· 该任务揭示了VLM在视觉推理上的短板，为改进模型提供了明确方向。\n影响\u002F看点：VICIS为评估和提升VLM的视觉概念学习能力设立了新基准，可能推动模型从“文本跟随”向“视觉理解”进化。","https:\u002F\u002Fmachinelearning.apple.com\u002Fresearch\u002Fvisual-concept-inference",null,"2026-07-17 08:00:00"]