Transformer language models (LMs) have been shown to represent concepts as directions in the latent space of hidden activations. However, for any human-interpretable concept, how can we find its direction in the latent space? We present a technique called linear relational concepts (LRC) for finding concept directions corresponding to human-interpretable concepts by first modeling the relation between subject and object as a linear relational embedding (LRE). We find that inverting the LRE and using earlier object layers results in a powerful technique for finding concept directions that outperforms standard black-box probing classifiers. We evaluate LRCs on their performance as concept classifiers as well as their ability to causally change model output.
Relational Semantic Grounding for Test-Time Discovery
School of Electronics and Information Engineering, Suzhou University of Science and Technology(苏州科技大学电子与信息工程学院);School of Intelligence Science and Technology, Nanjing University(南京大学智能科学与技术学院);School of Computer Science Engineering, North Minzu University(北方民族大学计算机科学与工程学院);Computer Vision Center, Universitat Autònoma de Barcelona(巴塞罗那自治大学计算机视觉中心)
ReGraP-LLaVA: Reasoning enabled Graph-based Personalized Large Language and Vision Assistant
Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences(中国科学院深圳先进技术研究所);School of Engineering, Westlake University(西湖大学工程学院);University of Minnesota – Twin Cities(明尼苏达大学双城分校);University of Toronto(多伦多大学)
PTR: A Benchmark for Part-based Conceptual, Relational, and Physical Reasoning
UCLA(加州大学洛杉矶分校);Stanford University(斯坦福大学);MIT BCS, CBMM, CSAIL(麻省理工学院脑与认知科学系、心智/大脑/机器研究中心、计算机科学与人工智能实验室);MIT CSAIL(麻省理工学院计算机科学与人工智能实验室);MIT-IBM Watson AI Lab(麻省理工学院-IBM沃森人工智能实验室)
人类视觉感知的一个关键方面是将视觉场景解析为单个对象并进一步解析为对象部分,形成部分-整体层次结构的能力。这种复合结构可以归纳出丰富的语义概念和语义关系,从而在视觉信号的解释和组织以及视觉感知和推理的概括中发挥重要作用。然而,现有的视觉推理基准大多集中在对象,而不是部分。基于完整部分-整体层次结构的视觉推理比以对象为中心的推理更具挑战性,因为更细粒度的概念,更丰富的几何关系和更复杂的物理。因此,为了更好地服务于基于部分的概念,关系和物理推理,我们引入了一个新的大规模诊断视觉推理数据集命名PTR。PTR包含大约70 k RGBD合成图像,其中包含关于语义实例分割、颜色属性、空间和几何关系以及某些物理属性(如稳定性)的地面实况对象和部件级注释。这些图像与70万个机器生成的问题配对,这些问题涵盖了各种类型的推理类型,使它们成为视觉推理模型的良好测试平台。我们在这个数据集上检查了几个最先进的视觉推理模型,并观察到它们在人类可以很容易地推断出正确答案的情况下仍然会犯许多令人惊讶的错误。我们相信这个数据集将为基于部件的推理开辟新的机会。