In MoE with learned routing, the router learns how to allocate the inputs to different expert network. To increase the computation efficiency, the computation of the router should not significantly slow the entire model. Furthermore, there are two key points to address: sparsity and load balancing.


MoE with Learned Routing

Compared with MoE with learned router, unlearnable routing mechanism simplifies the computation by getting rid of load balancing process. Unlearnable routing can be achieved by hashing function, which hash a token to an expert, or by optimizing the consistency between a pair of randomly selected experts and allocating the tokens randomly.

MoE with Unlearnable Routing

Mixture of Expert (MoE) takes the advantage of increased amount of parameters without increasing the computation load. A layer in such a model consists of several expert network, and only part of them will be activated during inference. One key component of MoE model is the routing mechanism, which can be divided into two types: learned routing and unlearnable routing.

New York University

There are three types of techniques to achieve dynamic neural network: Skimming, MoE, and Early Exit. The common idea is to adjust computation based on inputs.
- Skimming
    - Skipping and Early Stopping
    - Computation Reduction
    - Dynamic Hierarchical RNN
- Mixture of Experts
    - Learned Routing
    - Unlearnable Routing
- Early Exit
    - Confidence-based
    - Ensemble-based
    - Learning-based
    - Cascading

Types of Dynamic Neural Networks Adopted in NLP

Xu, C., & McAuley, J. (2022). A Survey on Dynamic Neural Networks for Natural Language Processing. ArXiv, abs/2202.07101.

A Survey on Dynamic Neural Networks for Natural Language Processing

Skimming is widely used in Recurrent Neural Network (RNN), where inputs are sequentially processed in time steps. Skimming models save computation timewisely by dynamically allocating computations to different time steps. It intuitively matches how human read and extract information: focusing on the important part and ignoring parts with little importance.
There are three types of skimming:
- Skipping and early stopping
- Computation reduction
- Dynamic hierarchical RNN

Skimming

Mixture of Expert in Dynamic Neural Network

Early exit method aims to terminate the computation in earlier layer to reduce computation and achieve a better performance. The key of such method is the exit criterion, which can be categorized as: confidence-based, ensemble-based and learning-based.

Learn Before

Related

Learn After