Published onNovember 17, 2024Test Time Training for Abstract ReasoningAbstraction and Reasoning CorpusLow-Rank AdaptationTest-Time Trainingabstract reasoningaugmented inferencedatadata augmentationfine-tuninggeometric transformationhierarchical votinginvertible transformationlarge language modellosssymmetryThis paper explores the use of Test-Time Training (TTT) to enhance the abstract reasoning capabilities of large language models (LLMs), specifically focusing on the Abstraction and Reasoning Corpus (ARC) benchmark.