DC Field | Value | Language |
---|---|---|
dc.contributor.advisor | Yang, Eunho | - |
dc.contributor.advisor | 양은호 | - |
dc.contributor.author | Lee, Jung Hyun | - |
dc.date.accessioned | 2022-04-13T05:40:05Z | - |
dc.date.available | 2022-04-13T05:40:05Z | - |
dc.date.issued | 2021 | - |
dc.identifier.uri | http://library.kaist.ac.kr/search/detail/view.do?bibCtrlNo=963745&flag=dissertation | en_US |
dc.identifier.uri | http://hdl.handle.net/10203/292500 | - |
dc.description | 학위논문(석사) - 한국과학기술원 : AI대학원, 2021.8,[iii, 20 p. :] | - |
dc.description.abstract | Network quantization, which aims to reduce the bit-lengths of the network weights and activations, has emerged for their deployments to resource-limited devices. Although recent studies have successfully discretized a full-precision network, they still incur large quantization errors after training, thus giving rise to a significant performance gap between a full-precision network and its quantized counterpart. In this work, we propose a novel quantization method for neural networks, Cluster-Promoting Quantization (CPQ) that finds the optimal quantization grids while naturally encouraging the underlying full-precision weights to gather around those quantization grids cohesively during training. This property of CPQ is thanks to our two main ingredients that enable differentiable quantization: i) the use of the categorical distribution designed by a specific probabilistic parametrization in the forward pass and ii) our proposed multi-class straight-through estimator (STE) in the backward pass. Since our second component, multi-class STE, is intrinsically biased, we additionally propose a new bit-drop technique, DropBits, that revises the standard dropout regularization to randomly drop bits instead of neurons. As a natural extension of DropBits, we further introduce the way of learning heterogeneous quantization levels to find proper bit-length for each layer by imposing an additional regularization on DropBits. We experimentally validate our method on various benchmark datasets and network architectures, and also support a new hypothesis for quantization: learning heterogeneous quantization levels outperforms the case using the same but fixed quantization levels from scratch. | - |
dc.language | eng | - |
dc.publisher | 한국과학기술원 | - |
dc.subject | Network Quantization▼aCluster-Promoting Quantization▼aDropBits▼aHeterogeneous Quantization▼aNew Hypothesis for Quantization | - |
dc.subject | 네트워크 양자화▼a군집 촉진하는 양자화 | - |
dc.subject | 드랍비트▼a비동형 양자화▼a양자화에 대한 새로운 가설 | - |
dc.title | Cluster-promoting quantization with bit-drop for minimizing network quantization loss | - |
dc.title.alternative | 네트워크 양자화 손실을 줄이기 위한 군집 촉진하는 양자화 및 비트드랍 | - |
dc.type | Thesis(Master) | - |
dc.identifier.CNRN | 325007 | - |
dc.description.department | 한국과학기술원 :AI대학원, | - |
dc.contributor.alternativeauthor | 이정현 | - |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.