Curse of Dimensionality Makes CBIR System is Necessary for Storage and Retrieval

Here we compare new candidate tags with existing tags for trademark image retrieval (Wei et al., 2009; Phan and Androutsos, 2010) or comparatively hidden secret images to obtain intellectual Identify property owner. (Zhang and Ye, 2009). CBIR systems belonging to this category include trademarks (Kato, 1992), STAR (Wu et al., 1996) and ARTISAN (Eakins et al., 1998). Architectural engineering design engineering design shares many common functions such as the use of stylized 2D and 3D models to represent design objects (Khokher and Talwar, 2012).

There is the concept of dealing with high dimensional data called dimension emergency (that is, many features). Dimensionality means the computation speed required and the exponential increase in memory as the number of data dimensions (features) increases. Dimension reduction is a technique used to reduce the number of features included in the machine learning process. This helps to reduce complexity, reduce computational costs, and improve the computational speed of machine learning algorithms. This can be thought of as a technique to transform it into a new set of new predictors used to fit the original prediction to the model.

Due to problems related to curse of dimension, we can significantly improve the number of features / dimensions to allow us to improve the performance of our model and find the best solution for our machine learning model It is necessary to reduce it. Fortunately, in most real situations, you can reduce the dimensions of the training set without losing significant data differences. For example, some of our data points may have no meaning when describing the target variables we expect. Therefore, we may want to remove them from our analysis. In addition, since two data points are usually highly correlated with each other, you do not lose too much information by merging them into one data point.

One of the most famous algorithms in machine learning is called the gradient descent method. Its main advantage is that it avoids a lot of "curse of dimension". This problem has too many variables to execute brute force calculation on its optimum value in a system like a neural network. However, the gradient descent breaks the dimensional curse by amplifying the local minimum or minimum of the multidimensional error or cost function. This allows the system to determine the adjustment values ​​or weights assigned to each unit in the network and the accuracy will return to normal.