基于切空间判别学习的流形降维算法

在基于图像集的流形降维问题中,许多算法的核心思想都是把一个高维的流形直接降到一个维数相对较低、同时具有的判别信息更加充分的流形上.投影度量学习（projection metric learning,简称PML）是一种Grassmann流形降维算法.该算法是基于投影度量,并且使用RCG（Riemannian conjugate gradient）算法优化目标函数,其在多个数据集上都取得了较好的实验结果,但是对于复杂的人脸数据集,如YTC其实验结果相对较差,只取得了66.69%的正确率.同时,RCG算法的时间效率较差.基于上述原因,提出了基于切空间判别学习的流形降维算法.该算法首先对于PML中的投影矩阵添加扰动,使其成为对称正定（symmetric positive definite,简称SPD）矩阵;然后,使用LEM（log-euclidean metric）将其映射到切空间中;最后,利用基于特征值分解的迭代优化算法构造判别函数,得到变换矩阵.对提算法在多个标准数据集上进行了实验验证,并取得了较好的实验结果,从而验证了该算法的有效性.     

Novel CMOS multi-bit counter for speed-power optimization in multiplier design

The paper introduces a novel multi-bit counter for efficient binary multiplication. A 7:3 counter is proposed, customized and optimized by 3-pronged approach, firstly by group-wise parallel addition using half-adders, secondly by eliminating redundant carry-generators from the design and finally, by optimizing the resultant hardware. The circuit is designed and optimized on standard static-CMOS. Corner analyses with TT, FF and SS for PVT-variation have been performed on proposed design to study reliability and robustness. A benchmarking exercise on Power-Delay-Product (PDP) with reported candidate designs demonstrates superiority of the proposed design. The study reveals that the 7:3-counter, designed using proposed strategy, has up to 36% less PDP, compared to the best candidate-design. Next, the proposed counters are employed, first to design an 8-b × 8-b Column-Compression (CC) multiplier and thereafter, using decomposition-logic on thus-designed-8-b × 8-b-multipliers, to design a 16-b × 16-b multiplier. The multipliers are optimized on 90-nm standard-CMOS technology and compared for speed-power performance with reported candidate designs at 500 MHz. Simulations show that the presented design of 16-b × 16-b multiplier using proposed 7:3 counter offers 55% less PDP, compared to the best candidate design under identical conditions. Once again, all simulations are performed on TSMC 90-nm CMOS technology at 25 °C temperature and 1.0 V supply-rail.     

Failure analysis of a thermal tile on the space shuttle Columbia

Several high-temperature reusable surface insulation (HRSI) tiles from the space shuttle Columbia were analyzed for their response to the catastrophic events during re-entry. The analysis of the HRSI tile focused on the outer surface, which was composed of silica tile with a reaction-cured glass (RCG) coating, and the inner surface strain isolation pad, which is bonded to the tile using a room-temperature vulcanizing silicone adhesive. Light optical microscopy, scanning electron microscopy, and energy-dispersive X-ray spectroscopy were used to evaluate the tile surfaces. The tiles were found to have a heavily damaged outer surface, with much of the top RCG coating eroded by impact from shuttle debris and/or plasma flow during the shuttle’s breakup and re-entry. The RCG experienced slumping and the silica exhibited melting and glassification due to temperatures exceeding the glass transition temperature and the anticipated tile operating temperatures. Deep cavities present in the material were found to contain metal particles that are not normally present and suggest impact with shuttle debris. Silicone adhesive degradation was also found. The slumping of the silica, the degraded adhesive, and the erosion of the tile contributed to a complex process during which the individual tile delaminated from the shuttle airframe.     

主动数据库中规则可观察确定性判定问题的一个解决方法

判定规则集的可观察确定性,是主动数据库中的核心问题之一.本文对于数据检索采用了执行状态图,逐步求得每条执行规则对检索结果的影响,最后累积出最终的影响结果,从而确定其可观察确定性.对事务回退操作采用抽象操作符记录动作的执行顺序,通过动作的逆操作实现事务的回退操作.从而实现了规则集可观察确定性的判定.