
Reasoning Over Complex Documents with DocLLM with Armineh Nourbakhsh - #672
The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence)
Challenges of Model Performance on Tabular Data
The model struggles with tabular data due to the need to context switch when encountering tables compared to other document segments like bullet lists or paragraphs. Unlike other segments, tables require attention in two directions which changes the reading order. Tables vary in reading order, some being more columnar while others require row-wise processing. The model did not perform well in tabular reasoning tasks but excelled in key information extraction tasks. In visual question answering, the model performed on par with the state of the art, including GPT four.
00:00
Transcript
Play full episode
Remember Everything You Learn from Podcasts
Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.