I attended a meetup on artificial intelligence and machine learning sponsored and organized by PayU at Mumbai on Oct 7th 2017. I am seeing lot of medical practitioners in these places. There were some data scientist, ai programmers.
I came across terms like Deep learning, neural network, Vega (Deep learning platform), Value of pretraining, data distribution, cost of misclassification, Activation maps in Image analysis, Counter factual process, openai, epochs
Vega platform has layers, GUI based approach. Python has pipeline approach. Cypher datasets used in Image classification. Deep learning is a subset of Machine learning.
Qure.ai specializes in reading and interpreting x-ray, mri scan images. AI is all about one number multiplied with another number which in turn is multiplied with another number and it goes on. AI is matrix multiplication.
AI has three broad categories namely Language, Vision and Speech. Qure.ai focuses on vision analysis.
Vision analysis steps are classification (which objects are there in image such as man, sheep, dog), object detection (ability to create a box around each image for further analysis), segmentation (ability to extract the object totally out the image and process it deeper such as what is the eye colour, body language etc )
Perturbation method , one of the technique used in image analysis where image is divided into tabular rows and column cells. Each cell is analyzed as a separate image. And in final approach, try to corelate all cells.
Another technique is Super pixels classification. In this pixels with similar intensity is grouped together (example sections are highlighted as per brightness of every pixel). It's probably a LIME heatmap representation of an image.
Torch lights are eyes of Image analysis. Different type of torch lights are used to decipher an image from various perspectives.
These analysis are run through AI model to create different types of probability plots.
Recommended coursera course by Andrew Ng. CS219 lectures by Andrej. Some tools in market are Pytorch, Keras, Tensorflow, Theano, OpenCV, Skimage
No comments:
Post a Comment