On transfer learning using a MAC model variant
This work addresses transfer learning challenges in visual reasoning, though it is incremental as it builds on an existing model with minor modifications.
The authors tackled the problem of improving transfer learning efficiency by introducing a simplified variant of the MAC model that trains faster while maintaining comparable accuracy, achieving a 15-point increase in accuracy on CLEVR and CoGenT datasets through fine-tuning to match state-of-the-art results.
We introduce a variant of the MAC model (Hudson and Manning, ICLR 2018) with a simplified set of equations that achieves comparable accuracy, while training faster. We evaluate both models on CLEVR and CoGenT, and show that, transfer learning with fine-tuning results in a 15 point increase in accuracy, matching the state of the art. Finally, in contrast, we demonstrate that improper fine-tuning can actually reduce a model's accuracy as well.