The field of multimodal question answering and table reasoning is moving towards more efficient and effective methods for retrieving and ranking relevant documents and tables. Recent research has focused on developing novel approaches to multimodal learning, such as joint supervision and multimodal document ranking, to improve the accuracy of question answering systems. Additionally, there is a growing interest in exploring the effectiveness of different input representations and models for table question answering, including the use of images and texts. Noteworthy papers in this area include:

• A paper proposing a novel approach to multimodal textbook question answering using a mechanism for enhancing semantic representations through multi-objective joint training, achieving a 2.4% gain in accuracy on the validation set and 11.1% on the test set.
• A paper presenting a method for dynamically selecting table representations, resulting in a 10% average performance improvement compared to using both representations indiscriminately.
• A paper introducing a cascaded retrieval approach that uses a sparse retrieval model to filter candidate tables before applying dense models, achieving better retrieval performance than state-of-the-art retrievers.
• A paper proposing a unified chart-metadata generation framework for multi-task chart understanding, enabling a single chart to support multiple downstream tasks and resulting in an average performance improvement of 5% across all tasks.
• A paper presenting a method for enhancing table reasoning with iterative row-wise traversals, outperforming reasoning large language models by an average of 4.3% and achieving state-of-the-art results on WikiTableQuestions and TableBench.