Temporal sequence processing is fundamental in brain cognitive functions. Experimental data has indicated that the representations of ordinal information and contents of temporal sequences are disentangled in the brain, but the neural mechanism underlying this disentanglement remains largely unclear. Here, we investigate how recurrent neural circuits learn to represent the abstract order structure of temporal sequences, and how this disentangled representation of order structure from that of contents facilitates the processing of temporal sequences. We show that with an appropriate learn protocol, a recurrent neural circuit can learn a set of tree-structured attractor states to encode the corresponding tree-structured orders of given temporal sequences. This abstract temporal order template can then be bound with different contents, allowing for flexible and robust temporal sequence processing. Using a transfer learning task, we demonstrate that the reuse of a temporal order template facilitates the acquisition of new temporal sequences of the same or similar ordinal structure. Using a key-word spotting task, we demonstrate that the attractor representation of order structure improves the robustness of temporal sequence discrimination, if the ordinal information is the key to differentiate different sequences. We hope this study gives us insights into the neural mechanism of representing the ordinal information of temporal sequences in the brain, and helps us to develop brain-inspired temporal sequence processing algorithms.

A key requirement in sequence to sequence processing is the modeling of long range dependencies. To this end, a vast majority of the state-of-the-art models use attention mechanism which is of O(n^2) complexity that leads to slow execution for long sequences. We introduce a new Shuffle-Exchange neural network model for sequence to sequence tasks which have O(log n) depth and O(n log n) total complexity. We show that this model is powerful enough to infer efficient algorithms for common algorithmic benchmarks including sorting, addition and multiplication. We evaluate our architecture on the challenging LAMBADA question answering dataset and compare it with the state-of-the-art models which use attention. Our model achieves competitive accuracy and scales to sequences with more than a hundred thousand of elements. We are confident that the proposed model has the potential for building more efficient architectures for processing large interrelated data in language modeling, music generation and other application domains.

This yields continuous-time counterparts of Fast Weight Programmers and linear Transformers.

Also, perfect shuffles are well defined for non-powers-of-two, but the provided description only makes sense for powers of two. Perhaps this would be a good place to mention that you are restricting your attention to sequences of length 2 k, and padding inputs to allow this.

The paper presents a way to incorporate sparse routing networks into the transformer architecture to reduce the computation cost of attention for long sequences. The reviewers acknowledge that the idea is novel and the experiments suggest that the proposed architecture is potentially useful. However, the experiments do not demonstrate improved efficiency or accuracy on real world tasks with long sequences. Comparison with Transformer architectures that make use of sparse attention is lacking. Hence, I recommend acceptance as a poster.

Temporal sequence processing is fundamental in brain cognitive functions. Experimental data has indicated that the representations of ordinal information and contents of temporal sequences are disentangled in the brain, but the neural mechanism underlying this disentanglement remains largely unclear. Here, we investigate how recurrent neural circuits learn to represent the abstract order structure of temporal sequences, and how this disentangled representation of order structure from that of contents facilitates the processing of temporal sequences. We show that with an appropriate learn protocol, a recurrent neural circuit can learn a set of tree-structured attractor states to encode the corresponding tree-structured orders of given temporal sequences. This abstract temporal order template can then be bound with different contents, allowing for flexible and robust temporal sequence processing.

A key requirement in sequence to sequence processing is the modeling of long range dependencies. To this end, a vast majority of the state-of-the-art models use attention mechanism which is of O(n 2) complexity that leads to slow execution for long sequences. We introduce a new Shuffle-Exchange neural network model for sequence to sequence tasks which have O(log n) depth and O(n log n) total complexity. We show that this model is powerful enough to infer efficient algorithms for common algorithmic benchmarks including sorting, addition and multiplication. We evaluate our architecture on the challenging LAMBADA question answering dataset and compare it with the state-of-the-art models which use attention.

A key requirement in sequence to sequence processing is the modeling of long range dependencies. To this end, a vast majority of the state-of-the-art models use attention mechanism which is of O(n 2) complexity that leads to slow execution for long sequences. We introduce a new Shuffle-Exchange neural network model for sequence to sequence tasks which have O(log n) depth and O(n log n) total complexity. We show that this model is powerful enough to infer efficient algorithms for common algorithmic benchmarks including sorting, addition and multiplication. We evaluate our architecture on the challenging LAMBADA question answering dataset and compare it with the state-of-the-art models which use attention.

Learn Artificial Neural Networks (ANN) in R. Build predictive deep learning models using Keras and Tensorflow R Studio R interface to Keras Keras is a high-level neural networks API developed with a focus on enabling fast experimentation. Being able to go from idea to result with the least possible delay is key to doing good research. Keras has the following key features: Allows the same code to run on CPU or on GPU, seamlessly. User-friendly API which makes it easy to quickly prototype deep learning models. This means that Keras is appropriate for building essentially any deep learning model, from a memory network to a neural Turing machine.