FAM-Bench: New Benchmark Tests AI's Food-as-Medicine Reasoning
FAM-Bench has been developed by researchers as a multimodal benchmark aimed at assessing AI models' reasoning capabilities regarding food selections in relation to specific health issues. In contrast to current food AI benchmarks that emphasize dish identification, recipe comprehension, nutrient analysis, or general nutrition inquiries, FAM-Bench focuses on the decision-making aspect of whether a particular food option is suitable for a specific health condition. This benchmark features 2,500 instances validated by nutrition experts across 13 diet-related health issues. It includes two related tasks: evaluating dish suitability based on images and ingredient lists, and ranking four dishes according to their appropriateness for particular conditions. Both tasks necessitate the integration of ingredient data, visual cues, and clinical nutrition guidelines. This research was published on arXiv under ID 2605.31410.
Key facts
- FAM-Bench is a multimodal benchmark for Food-as-Medicine reasoning.
- It contains 2,500 nutrition-expert-verified instances.
- It covers 13 diet-related health conditions.
- It includes two tasks: dish-level suitability assessment and comparative dish analysis.
- Tasks require integrating ingredient evidence, visual preparation cues, and clinical nutrition constraints.
- Existing food AI benchmarks do not test health-aware decision making.
- The benchmark was announced on arXiv with ID 2605.31410.
- The work is categorized as a new announcement type.
Entities
Institutions
- arXiv