arXiv subjects

Xuan Shen

Publications and source records attributed to Xuan Shen.

39 records · Page 3Linked to original sources

arXiv2021

NPAS: A Compiler-aware Framework of Unified Network Pruning and Architecture Search for Beyond Real-Time Mobile Acceleration

With the increasing demand to efficiently deploy DNNs on mobile edge devices, it becomes much more important to reduce unnecessary computation and increase the execution speed. Prior methods towards this goal, including model compression and network architecture search (NAS), are largely performed independently and do not fully consider compiler-level optimizations which is a must-do for mobile acceleration. In this work, we first propose (i) a general category of fine-grained structured pruning applicable to various DNN layers, and (ii) a comprehensive, compiler automatic code generation framework supporting different DNNs and different pruning schemes, which bridge the gap of model compression and NAS. We further propose NPAS, a compiler-aware unified network pruning, and architecture search. To deal with large search space, we propose a meta-modeling procedure based on reinforcement learning with fast evaluation and Bayesian optimization, ensuring the total number of training epochs comparable with representative NAS frameworks. Our framework achieves 6.7ms, 5.9ms, 3.9ms ImageNet inference times with 78.2%, 75% (MobileNet-V3 level), and 71% (MobileNet-V2 level) Top-1 accuracy respectively on an off-the-shelf mobile phone, consistently outperforming prior work.

cs.LG↗

arXiv2021

Towards Fast and Accurate Multi-Person Pose Estimation on Mobile Devices

The rapid development of autonomous driving, abnormal behavior detection, and behavior recognition makes an increasing demand for multi-person pose estimation-based applications, especially on mobile platforms. However, to achieve high accuracy, state-of-the-art methods tend to have a large model size and complex post-processing algorithm, which costs intense computation and long end-to-end latency. To solve this problem, we propose an architecture optimization and weight pruning framework to accelerate inference of multi-person pose estimation on mobile devices. With our optimization framework, we achieve up to 2.51x faster model inference speed with higher accuracy compared to representative lightweight multi-person pose estimator.

cs.CV↗

arXiv2012

Magnetic ordering and structural phase transitions in strained ultrathin SrRuO$_{3}$/SrTiO$_{3}$ superlattice

Ruthenium-based perovskite systems are attractive because their Structural, electronic and magnetic properties can be systematically engineered. SrRuO$_3$/SrTiO$_3$ superlattice, with its period consisting of one unit cell each, is very sensitive to strain change. Our first-principles simulations reveal that in the high tensile strain region, it transits from a ferromagnetic (FM) metal to an antiferromagnetic (AFM) insulator with clear tilted octahedra, while in the low strain region, it is a ferromagnetic metal without octahedra tilting. Detailed analyses of three spin-down Ru-t$_{2g}$ orbitals just below the Fermi level reveal that the splitting of these orbitals underlies these dramatic phase transitions, with the rotational force constant of RuO$_6$ octahedron high up to 16 meV/Deg$^2$, 4 times larger than that of TiO$_6$. Differently from nearly all the previous studies, these transitions can be probed optically through the diagonal and off-diagonal dielectric tensor elements. For one percent change in strain, our experimental spin moment change is -0.14$\pm$0.06 $μ_B$, quantitatively consistent with our theoretical value of -0.1 $μ_B$.

cond-mat.str-el↗