Post-training Quantization with Multiple Points: Mixed Precision without Mixed Precision
This addresses the challenge of efficient neural network deployment on standard hardware for applications like image classification and object detection, though it is an incremental improvement over existing quantization methods.
The paper tackles the problem of quantizing pre-trained neural networks without retraining by proposing multipoint quantization, which approximates weight vectors using linear combinations of multiple low-bit vectors rather than single low-bit numbers. The method achieves state-of-the-art results on ImageNet classification and generalizes to object detection tasks like PASCAL VOC, with minimal memory and computation overhead.
We consider the post-training quantization problem, which discretizes the weights of pre-trained deep neural networks without re-training the model. We propose multipoint quantization, a quantization method that approximates a full-precision weight vector using a linear combination of multiple vectors of low-bit numbers; this is in contrast to typical quantization methods that approximate each weight using a single low precision number. Computationally, we construct the multipoint quantization with an efficient greedy selection procedure, and adaptively decides the number of low precision points on each quantized weight vector based on the error of its output. This allows us to achieve higher precision levels for important weights that greatly influence the outputs, yielding an 'effect of mixed precision' but without physical mixed precision implementations (which requires specialized hardware accelerators). Empirically, our method can be implemented by common operands, bringing almost no memory and computation overhead. We show that our method outperforms a range of state-of-the-art methods on ImageNet classification and it can be generalized to more challenging tasks like PASCAL VOC object detection.