从Berkeley CS的summary中看计算机最新发展

Berkeley 的计算机系是世界上最好的计算机系之一，看到它2006年的总结，很长，但是我们摘取一些，看看最新的成果，看看发展趋势：

1、通信

2、VLSI的CAD

3、控制、机器人和生物系统

An Automated High-Resolution Room Surveillance System Using a Robotic Camera, a Sensor Network, and Particle Filtering



4、嵌入式软件和系统

5、集成电路

6、MEMS & 纳米科技

人工智能

A Discriminative Matching Approach to Word Alignment

Simon Lacoste-Julien and Ben Taskar
(Professors Michael Jordan and Daniel Klein)

We present a discriminative, large margin approach to feature-based matching for word alignment. We cast the word alignment as a maximum weighted matching problem in which each pair of words receives a matching score based on features of that pair such as measures of association, similarity of the orthographic form, and so on. Large-margin estimation of the model is accomplished by solving a (relatively) small quadratic program. Even with only 100 labeled training examples and simple features which incorporate counts from a large unlabeled corpus, we achieve alignment error rate (AER) performance close to IBM Model 4, in much less time. Including Model 4 predictions as features, we achieve a relative AER reduction of 22% in over intersected Model 4 alignments.

In an extension to the model [2], we introduce a flexible, learnable scoring scheme that captures desirable fertility trends. We also model first-order interactions between alignments of consecutive words by formulating the problem as a quadratic assignment problem (QAP) in which we define scores for pairs of edges that connect consecutive words. Even though QAP is NP-Hard, the problems we encounter in practice in word alignment are small enough to be solvable using standard integer programming solvers. We use a relaxation of QAP to cast the large-margin estimation problem as a QP. Preliminary results show that these two extensions are capable of improving error rates over our original model.



18、编程系统

Autolocker: Synchronization Inference for Atomic Sections

19、科学计算

Berkeley UPC: Unified Parallel C