A.1 Motivation For what purpose was the dataset created? As an affiliated dataset, we created MIMIC-CXR-VQA to provide a benchmark for medical visual question answering systems. Who created the dataset (e.g., which team, research group) and on behalf of which Who funded the creation of the dataset? This work was (partially) supported by Microsoft Research Asia, Institute of Information & Communications Technology Planning & Evaluation (IITP) grant (No.2019-0-00075, RS-2022-00155958), National Research Foundation of Korea (NRF) grant (NRF-2020H1D3A2A03100945), and the Korea Health Industry Development Institute (KHIDI) What do the instances that comprise the dataset represent (e.g., documents, photos, EHRXQA contains natural questions and corresponding SQL/NeuralSQL queries (text). How many instances are there in total (of each type, if appropriate)? In EHRXQA, there are about 46.2K instances (16,366 image-related samples, 16,529 table-related samples, and 13,257 image+table-related samples).

Large Language Models (LLMs) have demonstrated remarkable capabilities in many real-world applications. Nonetheless, LLMs are often criticized for their tendencytoproduce hallucinations, wherein themodels fabricate incorrectstatements on tasks beyond their knowledge and perception.

The ability of large language models (LLMs) to mimic human-like intelligence hasledtoasurgeinLLM-based autonomous agents. ThoughrecentLLMsseem capable of planning and reasoning given user instructions, their effectiveness in applying these capabilities for autonomous task solving remains underexplored. This is especially true in enterprise settings, where automated agents hold the promise of a high impact.

To mitigate the dependency on labeled data and enhance the resilience of graph models, selfsupervised learning has been widely adopted in GNN training.

Sparsified Online Newton (SONew) method, a memory-efficient second-order algorithm that yields a sparsified yet effective preconditioner.

Large language models (LLMs) have achieved a remarkable performance on diverse tasks across multiple domains, as reported in recent surveys [Wei et al., 2022, Bubeck et al., 2023].