In this article, we propose a noisy channel/information restoration model for error recovery problems in Chinese natural language processing. A language processing system is considered as an information restoration process executed through a noisy channel. By feeding a large-scale standard corpus C into a simulated noisy channel, we can obtain a noisy version of the corpus N. Using N as the inp...

with the arrival of increasingly higher resolution pet systems, small amounts of motion can cause significant blurring in the images, compared to the intrinsic resolutions of the scanners. in this work, we have reviewed advanced correction methods for the three cases of (i) unwanted patient motion, as well as motions due to (ii) cardiac and (iii) respiratory cycles. for the first type of motion...

Gunshot detection finds application in the fields of law enforcement, forensic science, and defense (military applications). The first task of a sniper detector, aiming to estimate the direction of arrival of a given gunshot, is to detect automatically the presence of this audio event. Since it is a typical on-line application where a fast response is of paramount importance, a non (computation...

Spontaneous human speech contains a lot of sounds that are not proper speech, yet carry meaning, laughter being a good example. Recognizing such sounds from speech-sounds could improve speech recognition systems as well as widen the communicative range of automatic dialogue systems. Our goal is to develop methods for automatic classification non-speech vocal sounds. As laughter varies widely be...

In damage detection the number of elements is generally more than the number of measured frequencies. Consequently, the corresponding damage detection equation is undetermined and thus has infinite solutions. Since in the damaged structures most of their elements remain healthy, the sparsest solution for the damage detection equation is mostly the actual damage. In the proposed method, the dama...

Existing works on anomaly detection (AD) rely clean labels from human annotators that are expensive to acquire in practice. In this work, we propose a method leverage weak/noisy (e.g., risk scores generated by machine rules for detecting malware) cheaper obtain detection. Specifically, ADMoE, the first framework algorithms learn noisy labels. nutshell, ADMoE leverages mixture-of-experts (MoE) a...

This study introduces a biologically-inspired model designed to examine the role of coincidence detection cells in speech segregation tasks. The consists three stages: time-domain cochlear that generates instantaneous rates auditory nerve fibers, amplify neural activity synchronously with presence, and an optimal spectro-temporal presence estimator. A comparative analysis between estimation bas...