We show that in language learning, contrary to received wisdom, keeping exceptional training instances in memory can be beneecial for generalization accuracy. We investigate this phenomenon empirically on a selection of benchmark natural language processing tasks: grapheme-to-phoneme conversion, part-of-speech tagging, prepositional-phrase attachment, and base noun phrase chunking. In a rst ser...

Linux is increasingly used to power everything from embedded devices to supercomputers. Developers of such systems often start with a mainline kernel from kernel.org and then apply patches for their application domain. Many of these patches represent crosscutting concerns in that they do not fit within a single program module and are scattered throughout the kernel sources—easily affecting over...

Attacks on host computers by malicious peripherals are a growing problem; the inexorable advance of powerful, small, cheap peripherals, which attach to plug-and-play buses, has made such attacks easy to mount. At the same time, commodity operating systems lack systematic defenses. We present Cinch, a pragmatic response to this threat. Cinch transforms USB peripherals to untrusted network endpoi...

Methods based on projective geometry have become popular in machine vision because they lead to elegant mathematics, and easy-to-solve linear equations [Longuett-Higgins 81, Hartley 97a, Quan & Lan 99]. It is often not realized that one pays a heavy price for this. Such method do not correctly model the physics of image formation and as a result require more correspondences and are considerably...