Elamur Research⌕
☰
home / papers / 2016 / Tracking the World State with Recurrent Entity Networks

Tracking the World State with Recurrent Entity Networks

Paper · 2016 · Mikael Henaff, Jason Weston, Arthur Szlam et al.

Summary

The paper addresses how a model can keep an up-to-date internal picture of a changing situation while it reads, such as the state of a short story, instead of reasoning only once a question arrives. The authors propose the Recurrent Entity Network (EntNet). It is a memory-augmented model with a fixed set of slots, each holding a learned key and a content vector. A gated recurrent update, shared across all slots, modifies them in parallel. A slot's gate opens when the encoded sentence matches its key (location addressing) or its current content, and a normalization step lets older information fade. Sentences are encoded with a learned multiplicative mask over word embeddings, and a one-hop attention readout produces answers. On the 20 bAbI tasks, EntNet is the first method to solve all of them with 10k training examples. It also succeeds on a new grid-world task that needs many supporting facts and generalizes to longer sequences than it saw in training. On the Children's Book Test it is competitive while reading in a single pass. It offers a simpler, parallel alternative to NTM/DNC-style memories.

Key points

  • The memory is split into blocks: m ≈ 5–20 slots, each with about 20–100 units. The update matrices U, V, W are shared across slots, and keys change only during training. A "tied" variant binds keys to entity word embeddings.
  • The new world-model task places two agents on a 10 × 10 grid. They turn or move 1–5 steps, and the model must answer where each agent ends up. EntNet beats LSTMs and MemN2N on this task.
  • With plain BoW input encoding on bAbI 10k, EntNet reaches a mean error of 10.9 with 6 failed tasks, compared with MemN2N at 15.6 and 9. Tying keys to entities gave no benefit in this setting.
  • With only 1k training samples, EntNet performs worse than MemN2N, with a mean error of 29.6 versus 13.9 and 15 versus 11 failed tasks.
  • In joint training on all tasks, averaged over seeds, EntNet's mean error is 9.7 ± 2.6 with about 5 failed tasks. The DNC scores 16.7 ± 7.6 with about 11.2 failed tasks. The DNC's best single seed is stronger, but its results vary much more between runs.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🏢MetaMenlo Park · United States214 works · 241,492 citations🎓New York UniversityNew York · United States219 works · 103,487 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksComputer Vision and Pattern Recognition · 3,486 works

Cite

Cite as Mikael Henaff, Jason Weston, Arthur Szlam, Antoine Bordes and Yann LeCun. 2016. Tracking the World State with Recurrent Entity Networks. arXiv:1612.03969 · ICLR 2017.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.1612.03969 · arXiv 1612.03969

@misc{henaff2016tracking,
  title = {{Tracking the World State with Recurrent Entity Networks}},
  author = {Mikael Henaff, Jason Weston, Arthur Szlam, Antoine Bordes and Yann LeCun},
  year = {2016},
  doi = {10.48550/arxiv.1612.03969},
  eprint = {1612.03969},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.1612.03969}
}

References

26 references, 6 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Bhuwan Dhingra, Hanxiao Liu, William Cohen and Salakhutdinov, Ruslan. attention readers for text comprehension. CoRR, abs/1606.01549, 2016. http://arxiv.org/abs/1606.01549. ↗
  2. Chandar, Sarath, Ahn, Sungjin, Larochelle, Hugo, Vincent, Pascal, Tesauro, Gerald, and Bengio, Yoshua. Hierarchical memory networks. arXiv preprint arXiv:1605.07427, 2016. — [arXiv:1605.07427](https://arxiv.org/abs/1605.07427) find
  3. Cho, Kyunghyun, van Merrienboer, Bart, Bahdanau, Dzmitry, and Bengio, Yoshua. On the properties of neural machine translation: Encoder-decoder approaches. In Proceedings of SSST@EMNLP 2014, Eighth Workshop on Syntax, Semantics and Structure in Statistical Translation, Doha, Qatar, 25 October 2014, p…
  4. Collobert, Ronan, Kavukcuoglu, Koray, and Farabet, Clment. Torch7: A matlab-like environment for machine learning, 2011. find
  5. Cui, Yiming, Chen, Zhipeng, Wei, Si, Wang, Shijin, Liu, Ting, and Hu, Guoping. Attention-over-attention neural networks for reading comprehension. CoRR, abs/1607.04423, 2016. URL http://arxiv.org/abs/1607.04423. ↗
  6. Graves, Alex, Wayne, Greg, and Dnihelka, Ivo. Neural Turing Machines, September 2014. URL http://arxiv.org/abs/1410.5401. ↗
  7. Graves, Alex, Wayne, Greg, Reynolds, Malcolm, Harley, Tim, Danihelka, Ivo, Grabska-Barwińska, Agnieszka, Colmenarejo, Sergio Gómez, Grefenstette, Edward, Ramalho, Tiago, Agapiou, John, et al. Hybrid computing using a neural network with dynamic external memory. Nature, 2016. find
  8. Grefenstette, Edward, Hermann, Karl Moritz, Suleyman, Mustafa, and Blunsom, Phil. Learning to transduce with unbounded memory. In Advances in Neural Information Processing Systems, pp. 1828–1836, 2015. find
  9. Gulcehre, Caglar, Chandar, Sarath, Cho, Kyunghyun, and Bengio, Yoshua. Dynamic neural turing machines with soft and hard addressing schemes. CoRR, abs/1607.00036, 2016. URL http://arxiv.org/abs/1607.00036. ↗
  10. He, Kaiming, Zhang, Xiangyu, Ren, Shaoqing, and Sun, Jian. Delving deep into rectifiers: Surpassing human-level performance on imagenet classification. CoRR, abs/1502.01852, 2015. find
  11. Hill, Felix, Bordes, Antoine, Chopra, Sumit, and Weston, Jason. The goldilocks principle: Reading children’s books with explicit memory representations. In Proceedings of the International Conference on Learning Representations. 2016. — The Goldilocks Principle: Reading Children's Books with Explicit Memory Representations
  12. Hochreiter, Sepp and Schmidhuber, Jürgen. Long short-term memory. Neural Comput., 9(8): 1735–1780, November 1997. ISSN 0899-7667. doi: 10.1162/neco.1997.9.8.1735. URL http://dx.doi.org/10.1162/neco.1997.9.8.1735. ↗
  13. Joulin, Armand and Mikolov, Tomas. Inferring algorithmic patterns with stack-augmented recurrent nets. arXiv preprint arXiv:1503.01007, 2015. find
  14. Kadlec, Rudolf, Schmid, Martin, Bajgar, Ondrej, and Kleindienst, Jan. Text understanding with the attention sum reader network. CoRR, abs/1603.01547, 2016. URL http://arxiv.org/abs/1603.01547. ↗
  15. Kingma, Diederik P. and Ba, Jimmy. Adam: A method for stochastic optimization. CoRR, abs/1412.6980, 2014. URL http://arxiv.org/abs/1412.6980. ↗
  16. Gated-URL Lerer, Adam, Gross, Sam, and Fergus, Rob. Learning physical intuition of block towers by example. In Proceedings of the 33nd International Conference on Machine Learning, ICML 2016, New York City, NY, USA, June 19-24, 2016, pp. 430–438, 2016. URL http://jmlr.org/proceedings/papers/v48/lere… ↗
  17. Li, Yujia, Tarlow, Daniel, Brockschmidt, Marc, and Zemel, Richard S. Gated graph sequence neural networks. CoRR, abs/1511.05493, 2015. URL http://arxiv.org/abs/1511.05493. ↗
  18. Mathieu, Michaël, Couprie, Camille, and LeCun, Yann. Deep multi-scale video prediction beyond mean square error. CoRR, abs/1511.05440, 2015. URL http://arxiv.org/abs/1511.05440. — [arXiv:1511.05440](https://arxiv.org/abs/1511.05440) ↗
  19. Miller, Alexander, Fisch, Adam, Dodge, Jesse, Karimi, Amir-Hossein, Bordes, Antoine, and Weston, Jason. Key-value memory networks for directly reading documents. arXiv preprint arXiv:1606.03126, 2016. find
  20. Munkhdalai, Tsendsuren and Yu, Hong. Reasoning with memory augmented neural networks for language comprehension. CoRR, abs/1610.06454, 2016. URL https://arxiv.org/abs/1610.06454. — [arXiv:1610.06454](https://arxiv.org/abs/1610.06454) ↗
  21. Sukhbaatar, Sainbayar, szlam, arthur, Weston, Jason, and Fergus, Rob. End-to-end memory networks. In Cortes, C., Lawrence, N. D., Lee, D. D., Sugiyama, M., and Garnett, R. (eds.), Advances in Neural Information Processing Systems 28, pp. 2440–2448. Curran Associates, Inc., 2015. URL http://papers.ni… ↗
  22. Sukhbaatar, Sainbayar, Szlam, Arthur, and Fergus, Rob. Learning multiagent communication with backpropagation. CoRR, abs/1605.07736, 2016. URL http://arxiv.org/abs/1605.07736. — [arXiv:1605.07736](https://arxiv.org/abs/1605.07736) ↗
  23. Trischler, Adam, Ye, Zheng, Yuan, Xingdi, and Suleman, Kaheer. Natural language comprehension with the epireader. CoRR, abs/1606.02270, 2016. URL http://arxiv.org/abs/1606.02270. ↗
  24. Weston, Jason. Dialog-based language learning. http://arxiv.org/abs/1604.06045. Weston, Jason, Chopra, Sumit, and Bordes, Antoine. Memory networks. CoRR, abs/1410.3916, 2014. URL http://arxiv.org/abs/1410.3916. ↗
  25. Weston, Jason, Bordes, Antoine, Chopra, Sumit, and Mikolov, Tomas. Towards ai-complete question answering: A set of prerequisite toy tasks. CoRR, abs/1502.05698, 2015. URL http://arxiv.org/abs/1502.05698. ↗
  26. Xiong, Caiming, Merity, Stephen, and Socher, Richard. Dynamic memory networks for visual and textual question answering. In ICML, 2016. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/henaff-2016-tracking-world-state-recurrent-entity/