Random data augmentation is a critical technique to avoid overfitting in training deep neural network models. However, data augmentation and network training are usually treated as two isolated processes, limiting the effectiveness of network training. Why not jointly optimize the two? We propose adversarial data augmentation to address this limitation. The main idea is to design an augmentation network (generator) that competes against a target network (discriminator) by generating "hard" augmentation operations online. The augmentation network explores the weaknesses of the target network, while the latter learns from "hard" augmentations to achieve better performance. We also design a reward/penalty strategy for effective joint training. We demonstrate our approach on the problem of human pose estimation and carry out a comprehensive experimental analysis, showing that our method can significantly improve state-of-the-art models without additional data efforts.

Negative photoconductivity (NPC) and positive photoconductivity (PPC) are observed in the same individual InAs nanowires grown by metal-organic chemical vapor deposition. NPC displays under weak light illumination due to photoexcitation scattering centers charged with hot carrier in the native oxide layer. PPC is observed under high light intensity. Through removing the native oxide layer and passivating the nanowire with HfO, we eliminate the NPC effect and realize intrinsic photoelectric response in InAs nanowire.

In this paper, we propose quantized densely connected U-Nets for efficient visual landmark localization. The idea is that features of the same semantic meanings are globally reused across the stacked U-Nets. This dense connectivity largely improves the information flow, yielding improved localization accuracy. However, a vanilla dense design would suffer from critical efficiency issue in both training and testing. To solve this problem, we first propose order-K dense connectivity to trim off long-distance shortcuts; then, we use a memory-efficient implementation to significantly boost the training efficiency and investigate an iterative refinement that may slice the model size in half. Finally, to reduce the memory consumption and high precision operations both in training and testing, we further quantize weights, inputs, and gradients of our localization network to low bit-width numbers. We validate our approach in two tasks: human pose estimation and face alignment. The results show that our approach achieves state-of-the-art localization accuracy, but using ∼70% fewer parameters, ∼98% less model size and saving ∼75% training memory compared with other benchmark localizers. The code is available at https

We report a systematic study on the correlation of the electrical transport properties with the crystal phase and orientation of single-crystal InAs nanowires (NWs) grown by molecular-beam epitaxy. A new method is developed to allow the same InAs NW to be used for both the electrical measurements and transmission electron microscopy characterization. We find both the crystal phase, wurtzite (WZ) or zinc-blende (ZB), and the orientation of the InAs NWs remarkably affect the electronic properties of the field-effect transistors based on these NWs, such as the threshold voltage (VT), ON-OFF ratio, subthreshold swing (SS) and effective barrier height at the off-state (ΦOFF). The SS increases while VT, ON-OFF ratio, and ΦOFF decrease one by one in the sequence of WZ ⟨0001⟩, ZB ⟨131⟩, ZB ⟨332⟩, ZB ⟨121⟩, and ZB ⟨011⟩. The WZ InAs NWs have obvious smaller field-effect mobility, conductivities, and electron concentration at VBG = 0 V than the ZB InAs NWs, while these parameters are not sensitive to the orientation of the ZB InAs NWs. We also find the diameter ranging from 12 to 33 nm shows much less effect than the crystal phase and orientation on the electrical transport properties of the InAs NWs. The good ohmic contact between InAs NWs and metal remains regardless of the variation of the crystal phase and orientation through temperature-dependent measurements. Our work deepens the understanding of the structure-dependent electrical transport properties of InAs NWs and provides a potential way to tailor the device properties by controlling the crystal phase and orientation of the NWs.

No abstract