Samarth - Research

Research is key because "It is a capital mistake to theorize before one has data!" - Sherlock Holmes

 

 

The hardest part about gathering information

Researching for projects in the field of AI is complex and, in my mind, slightly different than researching for non-AI projects. Some of the most challenging issues are trying to get reliable data sets; further, there are restrictions around getting data that includes personal information because of different laws. For example - getting anonymized data to train models for AI projects in healthcare is very tough. So, while I do research and take courses to develop a skillset for my project, a bulk of my research involves getting data in the public domain that are reliable and legal sources. 

While there is a lot of available information, sifting through the data, sorting the data, and finalizing the data that one can use takes time and careful evaluation.

Can you solve a problem without researching it?

Not at all! Fundamentally, research tells you if a problem needs to be solved or if a solution already exists. Also, there is a high chance that a project will fail if there is a lack of due diligence or research. (1) Research helps streamline the problem statement and helps bring clarity and focus. I use research to not just add on but also to eliminate. For example, research helps me eliminate certain sub-statements from my project. For example – if a solution for a part of my initial problem statement already exists, I may focus on either a novel method or another sub-statement. For example – A few years ago, I researched what fitness apps exist in the market. After my research, I decided to create an app for strength training only because many of the fitness apps focused on cardio and diet.

How much information do you need to know before you get started? Do you think it's important for scientists to be organized?

It isn't easy to quantify the amount of research before getting started on a project. Further, the constant influx of newly published research makes it challenging to stay current with the latest and greatest (information overload) (2) and decide what research is relevant to the project. And that is why the organization is fundamental for scientists. I work backward, usually based on deadlines. My major research has to stop before the design and implementation timeline through peripheral research can continue through the project – to help operate, enhance and fix bugs. While some deadlines [for personal projects] may move, the organization helps me keep focus and the required pace. 

Please see an example below.

What role does research and gathering information play in coming up with a solution to your problem?

Research is super important because it helps me understand what AI work may have been done in the same field. It allows me to focus on establishing my goal – If work has been done on a lot, can I train my model to get better accuracy, should I focus on a novel solution to the same problem, or should I change my problem statement and work? On something new. 

 

Citations for Blog & Video

1) According to TechRepublic, 85% of AI projects eventually fail to bring their intended results to the business. - https://www.techrepublic.com/article/why-85-of-ai-projects-fail/

Most organizations reported failures among their AI projects, with a quarter of them reporting up to a 50% failure rate. https://www.fastcompany.com/90449015/this-much-hyped-technology-is-failing-businesses-heres-why

2)    https://rebelem.com/information-overload-drinking-from-the-firehose/

3)    https://web.mit.edu/tslvr/www/lessons_two_years.html

4)    https://towardsdatascience.com/research-oriented-code-in-ai-ml-projects-f0dde4f9e1ac

5)    https://hbr.org/2018/11/how-to-set-up-an-ai-rd-lab

6) Gifs from Giphy

7) Graphics from Canva Library