Quantized Approximately Orthogonal Recurrent Neural Networks
This work addresses the deployment challenge for ORNNs in resource-constrained environments, representing an incremental improvement by applying quantization to an existing method.
The paper tackles the problem of deploying Orthogonal Recurrent Neural Networks (ORNNs) on compact devices by quantizing their weight matrices and activations, achieving results similar to state-of-the-art full-precision models with 4-bit quantization on standard benchmarks.
In recent years, Orthogonal Recurrent Neural Networks (ORNNs) have gained popularity due to their ability to manage tasks involving long-term dependencies, such as the copy-task, and their linear complexity. However, existing ORNNs utilize full precision weights and activations, which prevents their deployment on compact devices.In this paper, we explore the quantization of the weight matrices in ORNNs, leading to Quantized approximately Orthogonal RNNs (QORNNs). The construction of such networks remained an open problem, acknowledged for its inherent instability. We propose and investigate two strategies to learn QORNN by combining quantization-aware training (QAT) and orthogonal projections. We also study post-training quantization of the activations for pure integer computation of the recurrent loop. The most efficient models achieve results similar to state-of-the-art full-precision ORNN, LSTM and FastRNN on a variety of standard benchmarks, even with 4-bits quantization.