Binary-decomposed DCNN for accelerating computation and compressing model without retraining
This addresses the need for efficient DCNN deployment on resource-constrained devices, offering a practical solution with incremental improvements over existing methods.
The paper tackles the problem of large model sizes and slow computation in Deep Convolutional Neural Networks (DCNN) for mobile and low-performance devices by proposing Binary-decomposed DCNN, which accelerates inference and compresses models without retraining. Results show speed increases of 1.79x for AlexNet and 2.07x for VGG-16, with model compression of about 80% and error rate increases limited to 1.20% and 2.16%, respectively.
Recent trends show recognition accuracy increasing even more profoundly. Inference process of Deep Convolutional Neural Networks (DCNN) has a large number of parameters, requires a large amount of computation, and can be very slow. The large number of parameters also require large amounts of memory. This is resulting in increasingly long computation times and large model sizes. To implement mobile and other low performance devices incorporating DCNN, model sizes must be compressed and computation must be accelerated. To that end, this paper proposes Binary-decomposed DCNN, which resolves these issues without the need for retraining. Our method replaces real-valued inner-product computations with binary inner-product computations in existing network models to accelerate computation of inference and decrease model size without the need for retraining. Binary computations can be done at high speed using logical operators such as XOR and AND, together with bit counting. In tests using AlexNet with the ImageNet classification task, speed increased by a factor of 1.79, models were compressed by approximately 80%, and increase in error rate was limited to 1.20%. With VGG-16, speed increased by a factor of 2.07, model sizes decreased by 81%, and error increased by only 2.16%.