Version 3.0

Project descriptions of backgrounds and goals.

A company installed sensors in a city to collect temperature data but the sensors show gaps in the same region depending on the environment such as latitude, distance from rivers, etc.. The goal is to build a model to predict temperature of the sensors by using temperature data by sensors and National Weather Service.

One of big issues in natual language problems is to handle large scale datasets. Several approaches have been introduced to solve text classification problem but have not tried large scale datasets. The goal is to build graph-based text classification model which can cope with large scale datasets.

Computer vision consists of different problem such as image classification, segmentation and object detection. Among those, image classification can be considered as the fundamental problem. The goal is to apply different ConvNet with different image sets.

Although many word embedding methods have been introduced, there is no grounded standard method to evaluate abilities of word embeddings. By evaluating popular word embedding methods, I'd like to focus on the abilities of word embedings.

Depending on individual perspectives, it predicts whether the individual votes or not. The dataset has following information: age group, education, gender, race, religion, etc.

For time series problems, many RNN based methods have been developed. But, transfering time series models from RNNs to CNNs outperforms RNNs based models. The dataset was collected by 10 people and has 3 labels:crouching, running and walking. Here compares time series results in Vanilar RNN, FCN(Fully Convolution Networks) and ResNets.