Stochastic Shared Embeddings: Data-driven Regularization of Embedding Layers
This addresses overfitting in large-scale neural networks, particularly for embedding layers, but is incremental as it builds on existing regularization techniques.
The authors tackled the problem of overfitting in deep neural networks by proposing stochastically shared embeddings (SSE), a data-driven regularization method for embedding layers, which improved generalization across six tasks, including recommender systems and natural language processing with transformers and BERT.
In deep neural nets, lower level embedding layers account for a large portion of the total number of parameters. Tikhonov regularization, graph-based regularization, and hard parameter sharing are approaches that introduce explicit biases into training in a hope to reduce statistical complexity. Alternatively, we propose stochastically shared embeddings (SSE), a data-driven approach to regularizing embedding layers, which stochastically transitions between embeddings during stochastic gradient descent (SGD). Because SSE integrates seamlessly with existing SGD algorithms, it can be used with only minor modifications when training large scale neural networks. We develop two versions of SSE: SSE-Graph using knowledge graphs of embeddings; SSE-SE using no prior information. We provide theoretical guarantees for our method and show its empirical effectiveness on 6 distinct tasks, from simple neural networks with one hidden layer in recommender systems, to the transformer and BERT in natural languages. We find that when used along with widely-used regularization methods such as weight decay and dropout, our proposed SSE can further reduce overfitting, which often leads to more favorable generalization results.