I am specially working on machine learning as applied to computer vision systems, by myself, I do not get a salary for this. I am designing and building a vision system that integrates many attributes of the human visual system, in a compact and efficient framework, such as attention, gradual learning, foveal & peripheral vision and eye saccade like scanning of visual scenes when integrating multiple high-level features together. The project is called IRIS-integrated recognition and inference system started in Jan 2016.
So design is a bit of a pain in the brain for a year now and now probably I will be able to do a demo and write about the results I have been getting with this system. I want a system which can break some records on some object detection datasets. I have the scarcity of resources to scale this system up in order to train on the 1000-class ImageNet challenge so I would test it on MNIST and Pascal VOC datasets for object detection. After this my goal is to develop applications that I can deploy using the completed system.
IRIS will be finalized as an API later and will be made available to the public , at a reasonable cost, for applications in robots, embedded vision systems, self-driving cars and medical applications. I have also decided to build robot systems using the IRIS API and enter into a robotics business by supplying toy or domestic robots powered by IRIS.
There are a number of projects that I have built but panorama app is one of them using novel computer vision algorithms I developed personally. Though most of the algorithms in that app are not ML based, the matching phase is, I have developed a rapid ML based feature matching scheme for the vision system behind the panorama recognition system, I call that algorithm the FastMatcher. So this was just the rough idea about this project.
I am very good at C++ programming so I use this to code system modules. I have devoted many years building a vision & machine learning library to power all my projects. So to make the vision system for the app I had to start from zero, come up with novel algorithms and write a linear algebra library, implementing algorithms such as singular value decomposition nonlinear algebra.I have further introduced neural networks, support vector machines and backpropagation + stochastic gradient descent algorithms into the library as means to help the implementation of the newer IRIS system.
I have self learnt things like Artificial intelligence (AI), ML, image processing, CV and programming in C++, Java Android development and Python. So this is all about what I do as an ML expert.
If you want to become an expert in Machine Learning, checkout this Machine Learning Certification Course offered by Intellipaat.