Abstract-This paper aims to accelerate the test-time computation of convolutional neural networks (CNNs), especially very deep CNNs [1] that have substantially impacted the computer vision community. Unlike previous methods that are designed for approximating linear filters or linear responses, our method takes the nonlinear units into account. We develop an effective solution to the resulting nonlinear optimization problem without the need of stochastic gradient descent (SGD). More importantly, while previous methods mainly focus on optimizing one or two layers, our nonlinear method enables an asymmetric reconstruction that reduces the rapidly accumulated error when multiple (e.g., ≥10) layers are approximated. For the widely used very deep VGG-16 model [1], our method achieves a whole-model speedup of 4× with merely a 0.3% increase of top-5 error in ImageNet classification. Our 4× accelerated VGG-16 model also shows a graceful accuracy degradation for object detection when plugged into the Fast R-CNN detector [2].

Inkjet printing has been considered an available way to achieve large size full-color RGB quantum dots LED display, and the key point is to obtain printed film with uniform and flat surface profile. In this work, mixed solvent of 20 vol % 1,2-dichlorobenzene (oDCB) with cyclohexylbenzene (CHB) was used to dissolve green quantum dots (QDs) with CdSe@ZnS/ZnS core/shell structure. Then, by inkjet printing, a flat dotlike QDs film without the coffee ring was successfully obtained on polyetherimide (PEI)-modified ZnO layer, and the printed dots array exhibited great stability and repeatability. Here, adding oDCB into CHB solutions was used to reduce surface tension, and employing ZnO nanoparticle layer with PEI-modified was used to increase the surface free energy. As a result, a small contact angle is formed, which leads to the enhancement of evaporation rate, and then the coffee ring effect was suppressed. The printed dots with flat surface profile were eventually realized. Moreover, inverted green QD-LEDs with PEI-modified ZnO film as electron transport layer (ETL) and printed green QDs film as emission layer were successfully fabricated. The QD-LEDs exhibited the maximum luminance of 12 000 cd/m and the peak current efficiency of 4.5 cd/A at luminance of 1500 cd/m.

No abstract

This paper aims to accelerate the test-time computation of deep convolutional neural networks (CNNs). Unlike existing methods that are designed for approximating linear filters or linear responses, our method takes the nonlinear units into account. We minimize the reconstruction error of the nonlinear responses, subject to a low-rank constraint which helps to reduce the complexity of filters. We develop an effective solution to this constrained nonlinear optimization problem. An algorithm is also presented for reducing the accumulated error when multiple layers are approximated. A whole-model speedup ratio of 4× is demonstrated on a large network trained for ImageNet, while the top-5 error rate is only increased by 0.9%. Our accelerated model has a comparably fast speed as the "AlexNet" [11], but is 4.7% more accurate.

The aggregation-induced emission (AIE) phenomenon is important in organic light-emitting diodes (OLEDs), for it can potentially solve the aggregationcaused quenching problem. However, the performance of AIE fl uorophorbased OLEDs (AIE OLEDs) is unsatisfactory, particularly for deep-blue devices (CIEy < 0.15). Here, by enhancing the device engineering, a deep-blue AIE OLED exhibits low voltage (i.e., 2.75 V at 1 cd m −2 ), high luminance (17 721 cd m −2 ), high effi ciency (4.3 lm W −1 ), and low effi ciency roll-off (3.6 lm W −1 at 1000 cd m −2 ), which is the best deep-blue AIE OLED. Then, blue AIE fl uorophors, for the fi rst time, have been demonstrated to achieve high-performance hybrid white OLEDs (WOLEDs). The two-color WOLEDs exhibit i) stable colors and the highest effi ciency among pure-white hybrid WOLEDs (32.0 lm W −1 ); ii) stable colors, high effi ciency, and very low effi ciency roll-off; or iii) unprecedented effi ciencies at high luminances (i.e., 70.2 cd A −1 , 43.4 lm W −1 at 10 000 cd m −2 ). Moreover, a three-color WOLED exhibits wide correlated color temperatures (10 690-2328 K), which is the fi rst hybrid WOLED showing sunlight-style emission. These fi ndings will open a novel concept that blue AIE fl uorophors are promising candidates to develop high-performance hybrid WOLEDs, which have a bright prospect for the future displays and lightings.

White organic light-emitting devices (WOLEDs) have drawn intense attention in both scientific and industrial communities due to their potential applications in full-color flatpanel displays, back-lighting sources for liquid-crystal displays and solid-state lighting sources. [1,2] To achieve white emission, mixtures of the three red, green and blue (RGB) primary colors or two complementary colors, are typically required. Various approaches towards realizing WOLEDs have been reported, including multilayer structures capable of sequential energy transfer, [3][4][5] multiple component emissive layers containing an appropriate ratio of RGB phosphorescent or fluorescent dopants, [6][7][8][9][10][11][12][13][14] polymer blends containing RGB emitting species, [8][9][10][11][12][13][14][15][16] charge transfer exciplexes or excimers broad emission [4,17] and single component layers that utilize a polymer with broad emission. [18][19][20][21] Polymer light-emitting devices (PLEDs) [8][9][10][11][12][13][14][15] offer the advantages of solution processing, including screen printing and ink-jet deposition, large area coverage, and low power consumption. However, white emission PLEDs are less efficient with respect to power efficiency (PE) and luminous efficiency (LE), when compared with devices fabricated using vacuumdeposition technologies. For example, vacuum deposited WOLEDs have efficiencies that exceed that of the incandescent light bulb (12-17 lm W -1). [1] In contrast, to the best of our knowledge, the most efficient white emission PLEDs are below 10 lm W -1 , [12][13][14] still far away from practical applications for solid-state lighting.In this communication, we report single layer white PLEDs with an emission layer containing a blend of two or three phosphorescent iridium complexes [22] within a poly(N-vinylcarbazole) (PVK)/1,3-bis[(4-tert-butylphenyl)-1,3,4-oxadiazolyl] phenylene (OXD-7) host matrix. Despite their simple architectures and straightforward fabrication procedures, these devices show excellent color purity and considerably improved power efficiencies and luminous efficiencies, when compared to other polymer-based white emission devices. Ideal white light emission occurs near the equi-energy white point with Commission Internationale de L'Eclairage (CIE) coordinate of (0.333, 0.333). In a blend, the contributions from independent emitters need to be carefully balanced so that their collective integrated emission approaches this coordinate. We relied on the previously reported iridium bis(2-(4,6-difluorophenyl)-pyridinato-N,C 2′ ) picolinate (FIrpic, blue emission), iridium tris(2-(4-tolyl)pyridinato-N,C 2′ ) (Ir(mppy) 3 , green emission), and iridium bis(1-phenylisoquinoline) (acetylacetonate) (Ir(piq), red emission); see Figure 1 for molecular structures. PVK was chosen as a host material due to its higher-lying triplet states (3.0 eV), excellent film-forming properties, high glass transition temperature (T g ) (∼ 160°C) and hole transport characteristics.[23] OXD-7 was included into the host matrix...

We integrated plastic into mesoporous paper to achieve a new type of substrate plastic–paper for high performance optoelectronics.

White polymer light‐emitting devices with a peak forward‐viewing power efficiency close to 40 lm W−1, corresponding to an external quantum efficiency of 28.8% and a luminous efficiency of 60 cd A−1, are demonstrated. The devices are based on two newly synthesized yellow‐emitting iridium complexes functionalized with the sterically hindered diarylfluorene chromophores and are fabricated by a simple solution‐processing method.