How ML helps with General Relativity "Machine learning is a branch of artificial intelligence that works by making inferences from raw data using sophisticated algorithms and powerful computers. For online shoppers, that means improved "you might also like..." suggestions, as well as, for scientists, machine learning tools reveal profound insights hiding in ballooning datasets." - Quote from Berkeley Lab. Berkeley Lab scientists research nanotechnology, high-performance computing, synchrotron x-ray research, networking, genomics, and more.
It’s great to see how much Math, Coding, Science, AI, and Deep Learning - are progressing! It’s so inspiring! We hope that it will solve some problems that humans face as a species and will improve upon over the course of our lives.
100 years ago Einstein presented his theory of general relativity. Recently, computer scientist Katherine Bouman was able to create an image of a black hole, confirming Einstein’s theory of general relativity.
“Bouman prepared a large database of synthetic astronomical images and the measurements they would yield at different telescopes, given random fluctuations in atmospheric noise, thermal noise from the telescopes themselves, and other types of noise. Her algorithm was frequently better than its predecessors at reconstructing the original image from the measurements and tended to handle noise better. She’s also made her test data publicly available online for other researchers to use.” Source: MIT News
What if ML can help us to control Thermonuclear Fusion? What is Thermonuclear Fusion you may ask me? This is how our Sun works. People are even trying to recreate controllable thermonuclear fusion processes in labs. Cool right?
But Bouman Couldn’t Track her Machine Learning Code. Imagine if She Could.
You can track it though.
Machine Learning projects are hard to build. We have datasets and we have our code. Ideally, we would run the code, view results, improve our code and repeat this cycle. And then we get a "perfectly" trained model.
Then time passes by, training data become outdated or might even get deleted. Sometimes you simply forget. Did you ever feel that abyss between the model and the process to create a model? Complicating it all is the underlying fact - model training requires a lot of resources to train.
It's not your Nodejs side-project that you can run from any stone :)
Ok, but what about this situation: dataset was overwritten or changed? It can be a big problem. A good tool that supporting transparency and reproducibility for an ML project can help you a lot.
"Work smarter, not harder" - remember? Machine Learning is not different here. In order to get better results, in order to achieve your goals effectively - you definitely should track the code of your machine learning experiments.
If you have a few (or many) experiences to share, you can become a contributor on Hacker Noon too. Simply create an account, and join 10,000+ contributors sharing their knowledge and expertise with the rest of us. Maybe you'll get featured on our next newsletter too.
