Visual Question Generation: Which Question Types are Most Similar to What We Attack? – In this paper, we propose a new framework for using machine learning algorithms to identify the most closely related questions and answer set from a large corpus of questions over a series of videos. This framework is very simple, yet extremely effective. The key idea is to use a deep neural network (DNN) to predict whether a question is related to a particular answer set. The DNN can learn the answer set using the response set, which is given by a model. The problem is to predict the most likely answer set of a question set, not the most likely answer set that is given by a model.

In this paper we show that a simple linear regression, with no explicit estimation of parameters, can achieve comparable or even better performance to a linear one. This results means that the time-series data of interest are more suitable for estimation and also easier to obtain than the time-series data in the case of high-dimensional time-series data. To make this work useful, we present a novel statistical model called the Gaussian distribution over time-series, which is able to compute the underlying time series. Based on our method, we obtain approximate statistics for the Gaussian distribution. To compare our approach to the traditional linear regression approach, we first propose a new model with a simple formulation: each time-series data is represented by a fixed point. We use the Gaussian process method and demonstrate that the Gaussian process approach achieves comparable or higher performance to the linear regression approach, and even outperforms the one based on the conventional linear time-series model.

LIDIOMA – A Deep Neural Network for Interactive Object Detection

# Visual Question Generation: Which Question Types are Most Similar to What We Attack?

Scalable and Robust Estimation of Feature-specific Temporal Discretization in Multivariate Time-SeriesIn this paper we show that a simple linear regression, with no explicit estimation of parameters, can achieve comparable or even better performance to a linear one. This results means that the time-series data of interest are more suitable for estimation and also easier to obtain than the time-series data in the case of high-dimensional time-series data. To make this work useful, we present a novel statistical model called the Gaussian distribution over time-series, which is able to compute the underlying time series. Based on our method, we obtain approximate statistics for the Gaussian distribution. To compare our approach to the traditional linear regression approach, we first propose a new model with a simple formulation: each time-series data is represented by a fixed point. We use the Gaussian process method and demonstrate that the Gaussian process approach achieves comparable or higher performance to the linear regression approach, and even outperforms the one based on the conventional linear time-series model.

## Leave a Reply