Alexander Kovalenko

Uladzislau Yorsh, Alexander Kovalenko

In pursuit of faster computation, Efficient Transformers demonstrate an impressive variety of approaches -- models attaining sub-quadratic attention complexity can utilize a notion of sparsity or a low-rank approximation of inputs to reduce the number of attended keys; other ways to reduce complexity include locality-sensitive hashing, key pooling, additional memory to store information in compacted or hybridization with other architectures, such as CNN. Often based on a strong mathematical basis, kernelized approaches allow for the approximation of attention with linear complexity while retaining high accuracy. Therefore, in the present paper, we aim to expand the idea of trainable kernel methods to approximate the self-attention mechanism of the Transformer architecture.

Alexander Kovalenko, Pavel Kordík, Magda Friedjungová

Small neural networks with a constrained number of trainable parameters, can be suitable resource-efficient candidates for many simple tasks, where now excessively large models are used. However, such models face several problems during the learning process, mainly due to the redundancy of the individual neurons, which results in sub-optimal accuracy or the need for additional training steps. Here, we explore the diversity of the neurons within the hidden layer during the learning process, and analyze how the diversity of the neurons affects predictions of the model. As following, we introduce several techniques to dynamically reinforce diversity between neurons during the training. These decorrelation techniques improve learning at early stages and occasionally help to overcome local minima faster. Additionally, we describe novel weight initialization method to obtain decorrelated, yet stochastic weight initialization for a fast and efficient neural network training. Decorrelated weight initialization in our case shows about 40% relative increase in test accuracy during the first 5 epochs.