WHO IS A RESEARCHER?

Hiya!

I am a Machine Learning Engineer and I have done exciting research works with a great team and with other independent researchers.

​

My research focus has mostly been around low-resourced NLP and Speech Processing.

​

Sharon Ibejih

Co-founder, Kadatemy

@sharonibejih

Context

​

Research is a process of investigating into and studying past works and experimenting with them to come up with new conclusions. The goal is to document (and share) findings that could improve the way problems are approached.

​

Research:

• The invention of electricity that now powers so many devices.
• The development of Google Map that helps people navigate locations.
• The development of programming languages, which enable computers do the tasks that we want them to.

​

Sample research works:

What is

Machine Learning Research?

Machine learning is a process that makes a system understand patterns in data so that it can imitate the way humans learn.

This is revolves around understanding and improving how machines learn. It’s covers a very broad scope like training method, bias, data availability, deployment techniques, performance metrics and monitoring.

Machine learning research:

Who are they?

• Members of research groups (or communities)
• Individuals who are independent researchers
• Researchers in organization labs
• Academics: Professors, PostDocs, PhD students.

Steven Kolawole

(Author: Sign-to-speech model for sign language understanding)

Their Qualities

• Curiosity
• Ability to learn from existing works
• Ability to use words like “why” and “what if”.
• Ability to share findings - including failures.
• Highly collaborative.

Some areas of ML research focus

1. Fairness and inclusion
2. Theoretical foundations of Deep Learning
3. Computation Issues
4. Data Privacy
5. Domain Specific - Health, Climate, Agriculture, Neuroscience.
6. Societal Impact - Speech, Vision, Robotics, Pandemic response.
7. Data Collection and Annotation.
8. Model Evaluations

Keeping up with research works in ML could be so much work. Sigh!

​

But here are ways to navigate it:

• Being part of a research community.
• Following ML researchers on Twitter.
• Using PapersWithCode and/or arxiv-sanity
• Having researchers as friends.

How to Carry an ML Experiment

​

ML researchers excel in research environments.

​

This environment involves:

• Understanding a problem
• Studying existing approaches that solves that problem (if any)
• Exploring/Improving those approaches using a Jupyter Notebook
• Documenting observations (even failures)/results
• Concluding on the best approach that could solve that problem

​

How to Publicly Present Findings

For machine learning advancement, it’s always good to share findings. This will save someone else the stress of repeating things you have done.

​

• Write an article/paper
• Submit the paper at workshops/conferences for review
• *Accepted works can be presented at these places as talks or posters.

Defining your research niche

You are a better researcher when you are deeply connected to the problem statement.

​

• What topics draw your interest the most?
• What aspect(s) of ML do you think need(s) to be improved?

​

Sometimes, it takes working on a couple of projects or papers to identify this. OR reading lots of papers.

​

Research Challenges in Africa

• Data availability
• Inability to access computation resource
• Very few past research works to study.
• High dependence on large pretrained models - which may not fit our use case.
• Lack of funds
• Very few research labs.

• Reach out to researchers for collaborations on papers that focuses on a problem statement you understand.
• Ensure the writing is clear and that it addresses the problem statement.
• Your results should show a comparison with other methods.
• State what contribution your paper is making.
• Take out time to proofread and share with a reviewer.
• Submit your work at an appropriate workshop/conference.
• Be open to corrections and rejections.

Writing your first research paper

​

A Wrap

• You do not need a PhD to do research
• Being a researcher in Africa will require doing lots of work - which is an interesting hands on.
• You can co-author a paper simply by carrying a task.
• Being a part of a research community is really helpful. Examples are SautiDB, DSN Groups, Masakhane, ML Collective, Black in AI, Women in Machine Learning.

THANK YOU!

Sharon Ibejih

Co-founder, Kadatemy

@sharonibejih

Thank you Sponsors