Introducing some common terms in Machine Learning

If you are new to Machine Learning, this article might help you get introduced to some commonly used terms. The goal behind this article is to be a soft introduction to these terms. If you intend to become a machine learning researcher, the usage of at least SOME of these technical terms should be like second nature to you. If the last sentence overwhelmed you, then, instead of this being a test of your knowledge, you can also bookmark this article. One use is when preparing for interviews. How many times has it happened to each one of us that we can’t find the right words to explain a concept that we use daily in our work?

1. End-to-end ML model

Reference : 2203.03605.pdf (arxiv.org)

2. Inductive Priors

Video at 21:00

3. Online vs Offline

Reference : Colar: Effective and Efficient Online Action Detection by Consulting Exemplars (thecvf.com)

4. Fully-supervised vs Weakly-supervised vs Zero-Shot

Fully-supervised has bounding box for the human and object and verb-object combination (text) along with the RGB image as ground truth data, weakly-supervised has only verb-object combination (text)along with the RGB image and zero-shot has only the RGB image.

Reference : 2309.05069.pdf (arxiv.org)

5. Long-tail distribution

Reference: Learning to Detect Human-Object Interactions With Knowledge (thecvf.com)

6. Sampling

Reference: 2307.12532.pdf (arxiv.org)

7. Frozen weights

Reference : Efficient Two-Stage Detection of Human–Object Interactions with a Novel Unary–Pairwise Transformer

8. Set-based prediction loss

2203.03605.pdf (arxiv.org)