# Training Recurrent Answering Units with Joint Loss Minimization for VQA

arXiv: https://arxiv.org/abs/1606.03647 · HTML: https://mlsanity.com/paper/1606.03647/

## Key facts

- Cited by 9 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 12 Jun 2016.

## Built on by (most cited first, 9 of 9)

- [Cognitive Visual Commonsense Reasoning Using Dynamic Working Memory](https://mlsanity.com/paper/2107.01671/) (2021)
- [Interpretable Visual Understanding with Cognitive Attention Network](https://mlsanity.com/paper/2108.02924/) (2021)
- [A Comprehensive Survey on Visual Question Answering Datasets and Algorithms](https://mlsanity.com/paper/2411.11150/) (2024)
- [Using Large Language Models for education managements in Vietnamese with low resources](https://mlsanity.com/paper/2501.15022/) (2025)
- [Survey of Natural Language Processing for Education: Taxonomy, Systematic Review, and Future Trends](https://mlsanity.com/paper/2401.07518/) (2024)
- [An Empirical Study on the Generalization Power of Neural Representations Learned via Visual Guessing Games](https://mlsanity.com/paper/2102.00424/) (2021)
- [Chop Chop BERT: Visual Question Answering by Chopping VisualBERT's Heads](https://mlsanity.com/paper/2104.14741/) (2021)
- [Improved RAMEN: Towards Domain Generalization for Visual Question Answering](https://mlsanity.com/paper/2109.02370/) (2021)
- [Attention Mechanism based Cognition-level Scene Understanding](https://mlsanity.com/paper/2204.08027/) (2022)

## Abstract (arXiv)

We propose a novel algorithm for visual question answering based on a recurrent deep neural network, where every module in the network corresponds to a complete answering unit with attention mechanism by itself. The network is optimized by minimizing loss aggregated from all the units, which share model parameters while receiving different information to compute attention probability. For training, our model attends to a region within image feature map, updates its memory based on the question and attended image feature, and answers the question based on its memory state. This procedure is performed to compute loss in each step. The motivation of this approach is our observation that multi-step inferences are often required to answer questions while each problem may have a unique desirable number of steps, which is difficult to identify in practice. Hence, we always make the first unit in the network solve problems, but allow it to learn the knowledge from the rest of units by backpropagation unless it degrades the model. To implement this idea, we early-stop training each unit as soon as it starts to overfit. Note that, since more complex models tend to overfit on easier questions quickly, the last answering unit in the unfolded recurrent neural network is typically killed first while the first one remains last. We make a single-step prediction for a new question using the shared model. This strategy works better than the other options within our framework since the selected model is trained effectively from all units without overfitting. The proposed algorithm outperforms other multi-step attention based approaches using a single step prediction in VQA dataset.