Friday, January 6, 2017 7:06:26 AM
- Data scientists have been using machine learning in the fields of image classification, video analytics, speech recognition and natural language processing.
- What is deep learning? It consists of the use of multilevel deep neural networks to build systems that can perform feature detection on large datasets that is unlabeled training data.
- ML classification and training models are being run on the cloud. GPU’s combined with machines of 1000’s of cores are able to give 10-100x speedup in application speeds. GPU’s the super fast dedicated computer chips were initially developed for home computer gaming.
- What are some popular machine learning tools? Caffe, cuda-convnet, Theano, Torch7, cuBLAS, MATLAB and mxnet are a few currently popular frameworks and technologies in popular use across the field of ML.
- Deep learning systems will get better and better as more and more data is fed into it. This currently seems the logical path to true artificial intelligence in the future.
- Google Brain was a network of 1000 computers at Google X lab whose goal was to soak up the worlds information. With 1 million simulated neurons and 1 billion simulated connections, it was ten times larger than any deep neural network before it. It was one of the first to identify that the internet is full of cat videos. This sparked the resurgence of machine learning.
- Deep learning basically means the use of neural networks. What are neural networks? inspired by the densely interconnected neurons of the brain, mimic human learning by changing the strength of simulated neural connections on the basis of experience.
- For everyday users, ML led to Google Photos that can sort and recognize photos while automatically creating albums and videos, Siri and Alexa that can understand spoken commands and a better google translate. On a scientific level ML can search for potential drug candidates, map real neural networks in the brain or predict the functions of proteins.
- Unsupervised learning is hard to achieve because it comes without any meta data, tags, names, titles, categories etc and the computer needs to figure it out all on its own.
- Deepnets are great at finding patterns in data sets. Thats one of the fundamental reason behind this field.
- How does the layering in neural networks work? Given a set of images, layer one would identify light and dark pixels, layer two identifies edges and shapes, layer three will identify more complex shapes and objects while layer four will build on the previous layers and finally map the shapes and objects to define a human face. The basic idea is that each layer will build on the work of the previous layer.
Sunday, December 25, 2016 8:48:16 PM
- Whats the difference between machine learning, artificial intelligence and deep learning? Think of them as concentric circles with AI — the idea that came first — the largest, then machine learning — which blossomed later, and finally deep learning — which is driving today’s AI explosion — fitting inside both.
- Several technological advancements have propelled machine learning forward such as: GPU’s (graphical processing units), infinite storage and big data with growth in the area starting around 2012 and being on a peak ever since 2015.
- What is the ultimate goal? It is to build a computer that can perform as well as or better than a human being. Is that possible? As of now, narrow AI is whats possible - only a small part of human actions are being performed better by the computer.
- Tensor flow is an example of a machine learning system built and open sourced by Google. It started out as an internal project called disbelief. At the core of it, tensor flow is built completely using graphs. There’s no black magic involved here!
- If there are no dependencies between the nodes of the graph then those nodes can be parallelized and thats powerful. Overfitting and under fitting are both problems that must be overcome in machine learning.
- Machine learning basically involves training the system first and then running the system in real time on new data. Example-> Lets teach it what a cat means. Then in the future, it will identify every input that has a cat from that which is not a cat.