MARVEL: Multidimensional Abstraction and Reasoning through Visual Evaluation and Learning

arXiv:2404.13591v1 Announce Type: cross
Abstract: While multi-modal large language models (MLLMs) have shown significant progress on many popular visual reasoning benchmarks, whether they possess abstract visual reasoning abilities remains an open question. Similar to the Sudoku puzzles, abstract visual reasoning (AVR) problems require finding high-level patterns (e.g., repetition constraints) that control the input shapes (e.g., digits) in a specific task configuration (e.g., matrix). However, existing AVR benchmarks only considered a limited set of patterns (addition, conjunction), input shapes (rectangle, square), …

abstract abstraction arxiv benchmarks constraints cs.cv cs.lg evaluation language language models large language large language models marvel mllms modal multidimensional multi-modal patterns popular progress question reasoning sudoku through type visual