Articles — Project / Lab Report

This project focuses on a modification of a greedy transition based dependency parser. Typically a Part-Of-Speech (POS) tagger models a probability distribution over all the possible tags for each word in the given sentence and chooses one as its best guess. This is then pass on to the parser which uses this information to build a parse tree. The current state of the art for POS tagging is about 97% word accuracy, which seems high but results in a around 56% sentence accuracy. Small errors at the POS tagging phase can lead to large errors down the NLP pipeline and transition based parsers are particularity sensitive to these types of mistakes. A maximum entropy Markov model was trained as a POS multi-tagger passing more than its 1-best guess to the parser which was thought could make a better decision when committing to a parse for the sentence. This has been shown to give improved accuracy in other parsing approaches. We shown there is a correlation between tagging ambiguity and parsers accuracy and in fact the higher the average tags per word the higher the accuracy.

Security Insider Lab

As part of the NASA IMPACT program to model mechanical interactions with bodies in space such as comets, asteroids, and other Near Earth Objects (NEOs), the Colorado School of Mines team is designing a motor stage apparatus to interact with JSC-1A regolith simulant surfaces. This initial study involved development and testing of a prototype motor stage apparatus which was used to drive three types of probes into JSC-1A surfaces while collecting force data under standard Earth atmospheric conditions. The probes used were a conical-tipped probe, a wedge-tipped probe, and an anchoring probe. Main goals of the prototype system were to acquire general force trends for each interaction, and to isolate the most important design features for a more-complex in-vacuum system. Our data revealed force interactions that were very small in magnitude—on the order of tenths of Newtons—and more complex than our simple stiff beam probe and mount model could accurately predict. Our results lead us to recommend a more complex experimental model that can accurately represent deflection in the probes while also allowing for better measurement of regolith movement near the tip of the probes. Specifically, we recommend the following design features for the in-vacuum system: a load cell capable of measuring at very low ranges (thousandths of Newtons), reliability of structural axis alignment between trials, a robust mounting system that can accommodate each different type of probe, and consistency of sample preparation between trials.

Trabalho de Cálculo 4 - CEFET/MG

Esse trabalho tem como objetivo formular problemas como busca.

Proyecto: Amazon.com

Accurate prognosis and prediction of a patient's current disease state is critical in an ICU. The use of vast amounts of digital medical information can help in predicting the best course of action for the diagnosis and treatment of patients. The proposed technique investigates the strength of using a combination of latent variable models (latent dirichlet allocation) and structured data to transform the information streams into potentially actionable knowledge. In this project, I use Apache Spark to predict mortality among ICU patients so that it can be used as an acuity surrogate to help physicians identify the patients in need of immediate care.

The goal of this project is to explore both the theory behind the Extended Kalman Filter and the way it was used to localize a four-wheeled mobile-robot. This can be achieved by estimating in real-time the pose of the robot, while using a pre-acquired map through Laser Range Finder (LRF). The LRF is used to scan the environment, which is represented through line segments. Through a prediction step, the robot simulates its kinematic model to predict his current position. In order to minimize the difference between the matched lines from the global and local maps, a update step is implemented. It should be noted that every measurement has associated uncertainty that needs to be taken into account when performing each step of the Extended Kalman Filter. These uncertainties, or noise, are described by covariance matrices that play a very important role in the algorithm. Since we are dealing with an indoor structured environment, mainly composed by walls and straight-edged objects, the line segment representation of the maps was the chosen method to approach the problem.

In this experiment, we attempt to better understand how materials properties are tested. We tested a number of simple beams of different materials under a stress. The bending of the materials allowed for us to calculate the Poisson's Ratio and elastic moduli for each material. From this, we were able to not only compare materials but also methods of measuring elasticity. Despite some error in our results, which can be explained by the scale of our measurements in relation to the stiffness of certain materials, we find both strain gauges and equations of cantilever to be appropriate measurement techniques for measuring the elastic modulus of simple beams.