Result: Improving Machine Learning Models for Large Data and Image Detection

The combination of Big Data with the memory size limitations common in the moderately-sized machines used in many complex ML applications motivates a search for fast parallel computation methods. In this work, a method called Software Alchemy, a technique to develop generally applicable parallel ML methods, while avoiding the need for a different approach to every algorithm will be demonstrated.Note, that this a multi-topic dissertation. Imaging processing is another major area where there is an increasing amount of data to deal with. Classic methods of image fraud detection often require prior knowledge of the method of forgery in order to determine which features to extract from the image to localize the region of interest. Thus, as the second main topic, certain classic and newer machine-learning based techniques of image fraud detection will be modified, combined, or parallelized to produce improved results compared to the classic, stand-alone, serial technique.Machine learning methods outperform the traditional methods by quickly and automatically extracting a com- bination of complex/hierarchical features. However, most of these new methods are basic CNN modifications with various filters added (targeting a type of forgery), which still lack the ability to be generalized to a wide range of forgery types and the ability to localize the tampered region. Consequently, object detection frameworks will then be adapted to better solve these problems by extracting various complex features within the localized region. These will out-perform both state-of-the-art classic techniques as well as those in machine learning.Breast cancer grading is crucial to assessing the stage of breast cancer development, the most common cancer among women, accounting for 50% diagnoses. Yet, it requires a tremendous amount of acquired skill and time for a pathologist to extract the relevant features from many histopathology images by hand to make an assessment. Therefore, as the third thi