AI Vs Humans! Can It Really Beat Humans?
3 main points
✔️ Evaluate natural language task performance between AutoML tools
✔️ Evaluate performance between AutoML tools and people
✔️ Results that could serve as a benchmark for future work
Leveraging Automated Machine Learning for Text Classification: Evaluation of AutoML Tools and Comparison with Human Performance
written by Matthias Blohm, Marc Hanussek, Maximilien Kintz
(Submitted on 7 Dec 2020)
Comments: Accepted for ICAART2021 conference as a Short Paper
Subjects: Machine Learning (cs.LG)
In recent years, AutoML has made a lot of progress, such as in AI-SCHOLAR: " Developed by Google! AutoML-Zero, which automatically generates machine learning algorithms by evolutionary search " and "A new paper on AutoML that automatically learns with a single line of code! The latest paper on AutoML to automate machine learning! and others have introduced the latest successes. That's how much attention AutoML is getting.
However, most of them are for organized table data. In this paper, we investigate whether AutoML is equally successful for natural language processing tasks. In addition to performance evaluation, we also evaluate the accuracy that humans have achieved so far. In other words, it is an AI vs. human evaluation. Let's take a look at the basic evaluation of AutoML.
To read more,
Please register with AI-SCHOLAR.OR
Categories related to this article