I started my PhD two months ago, on automated ways to classify documents, and I begin to be depressed…
Since I started University in Computer Science, I wanted to do a PhD in biocomputing, combinatorial optimization, or machine learning.
In CS class, Machine learning was the area I enjoyed the most - so it was natural to look for a PhD in ML.
For me, research activities conduct to new innovative techniques, or tools, which have to be tested on real world examples.
As a good PhD student, I started my start-of-the-art and I began to experiment two or three ideas. So, I read about 50 or 60 research articles about new ways to classify documents, and text understanding.
The research articles I read in the area of machine learning came, in large majority, from web giants: Google, Facebook, Amazon, Baidu, Criteo…, but also from companies or startups which are related to those giants, like DeepMind.
These research articles propose new approaches to classify text documents, and new ways to understand, and translate, documents.
Those approaches have been tested on a massive dataset, and clearly outperform existing methods.
Also, at NIPS (one of the greatest conference about machine learning and computational neuroscience), companies owned about 50% of accepted papers.
Now, I am wondering if innovative things in this research area still belong to Universities.
Maybe you are wondering why those companies are first, in this area.
This is my answer:
- They have massive datasets of real world examples,
- They have tech’ (for example, the last chip from Google for DL, which is just amazing),
- They own, or the work with, the best research scientists in AI (Y. LeCun, A. Ng, H. Larochelle…),
- They have the best engineers who are working hard on this subject,
- They can test their idea directly on well designed products, faster than light.
I realized, during my previous internships, that there is a hard competition between research institutions and companies.
Some research institutions, like inria, share the idea that a research idea can be part of a great product - so, they propose to scientists to share ideas with startups, or to create their own startup.
It’s the same thing in University of Luxembourg - for example, the DataThings startup has been funded thanks to the research ideas of a great team of scientists who are working on real time data series.
So, research institutions are moving from pure research to applied research, and it’s great ! But, is this change not too late ? And is AI research area already belongs to giants ?
I know that companies, especially today with DL, would not have done these wonderful tools/products without research scientists. But I am sad those research scientists are now owned by private companies, to conduct internal research.
I am really scared that my work and research ideas are became soon outdated, because of a technology that exists since a few years, but has not been released yet due to private IPs or something like that. Also, how can you be peaceful if you know that companies can explore, implement a solution and draft a paper about it faster than you ?
Actually, I am depressed… but I am not giving up.