The AI journey of an undergraduate research student

source: google images

I wanted to share the things that I said to myself “If I had heard of these experiences before, I would not have let some of my thoughts affect me badly.” in my early times. This story for those who want to start this journey. A perspective from a student that passionate about AI.

Abstract of my nerdy story.

I will be starting with my ignorance at the beginning. After that, I will be talking about how I passed the stages which include my robotic experience with computer vision and changing my direction to be researcher on my advisor’s research project at university. At the end, I will be explaining the being an active contributor of this field and feeling confident to start as a researcher at a company as a student.

The Ignorance Era

It has been nearly 2 years since I start doing things related with AI as an undergraduate student. I was extremely excited about developments I see in this field. I was thinking “I need to do something about artificial intelligence”, even though I have no idea where I am going to start and what I need to do next.

Everyone was saying around me “That’s pointless to start AI in the first grade. You will not be able to go further anyways.” Well. I knew that will be hard. But I was not agree at all. I was aware of the requirements which are deep math knowledge and real life problems to actually change something, and I was sort of blind when it comes to start. My knowledge was really insufficient to understand specific AI research areas and decide what I need to do next in that time.

Photo by Tim Mossholder on Unsplash

I was spending my very first months mostly looking for information about AI. “What actually AI is, how I can start, what is computer vision, what is nlp…” I continue with the keywords I saw from lots of different sources about my questions. “What is machine learning, what is the difference between deep learning and machine learning, how I can start ml…” Well. My brain was about to explode from so much information. It took more than messy 2 months to start having perspective about keywords and their relations.

I decided to start with machine learning courses. However, nothing was going nice for me in reality, because my methods had no order at all. I was trying to learn things randomly. It was boring and extremely hard for me. I think myself kinda hyperactive person. I was spending my time trying to watch ml courses I found. I felt really bad in this process. Most of the courses were similar, but I was feeling I understand nothing. I mostly thought like, well here we go again. I learned lots of things and forget most of the basic knowledge again, again and again. I do not know how many times I’ve forgotten what is precision, recall and f1 etc. and studied one more time.

There is a key issue for me and I think we can generalize as psychological truth for most people. We should know that if we do not have a purpose, we will forget most of what we watch or do without understanding why we need that information exactly. From my above forgetting experience, I can easily say that I stopped forgetting most of the basic and complex things after I learned them to use on my research project or realize on our rover(will be explained). Basically, if you learn something for a specific purpose, if you learn that thing to realize on some project or if you aware of why you need that information in the near future, that information will become your actual knowledge more faster than you thought. Our brain categorizes the information every time while we are learning.
Is it general information and will be no problem with forgetting?
-Okay forget it in hours.
Is it crucial information in current situation?.
-Damn, we need this, keep it in the long memory.

Therefore, of course, I continued to work a lot of math as well in this process. I never pushed back the math because I had strong belief those will be useful. Still think it was the right choice.

Welcome to be a part of lifelong learners.

Get your hands dirty

After watching countless machine learning related videos, I found a chance to change my direction into real world applications with AI. That chance was being a member of the team that aims to preparing for URC(University Rover Challenge), ERC(European Rover Challenge) and Teknofest. Well. May seems perfect. However, everything was going to really, really tough.

Our team was newly established, a group of people who are visionary and passionate enough to create a robot from scratch in less than a year. This time interval was really challenging amount of time to build including learning time and realize it. We had no starting documentation and we should learn everything that we cannot find easily in the internet. Spoiler: We did it.

We reached experts in many fields to hear their experiences in some of the problems. We made lots of meeting with university teachers and other rover teams to learn from their knowledge. My field was determined. I was going through computer vision.

Well, I had intuition about machine learning and general stuff that we can find in every courses. The problem is that I had no experience and no idea about what are the most useful computer vision applications in real life, specifically, object detection, tracking. My main task was creating a customized dataset and model for the competition tasks which are mostly detection of a specific object in real time to trigger the specified act of rover. At the beginning, I was like, here we go again. My ignorance was giving me tension time to time. I could not be able to see what is next. At least I had a purpose that includes space discovery and competition tasks to complete to continue working with passion.

I constantly did my research about computer vision and robotics. I watched countless videos on the youtube and on the other sites. I faced with tons of errors with integrating the system on Jetson Cards which we can consider them as brain of the rover while using Linux and ROS.

I can freely say that solving errors was taking more than half of my time, but errors are your friends. You mostly learn from them. It makes you feel really bad in that time, but they are good friends.

Of course, accidents happens as well: One of my longest model training on the local card was unplugged, so my 10 hours were wasted. Do not even say labelling data wrong and realize that after training…

With the approaching of summer, our work got accelerated. We were spending most of our time in our atelier. Everything I learned was changing my perspective because the knowledge I have was limited and I were constructing it right in that time.

Sadly we could not find a chance to compete at URC and ERC because of the pandemic. However, Teknofest time was coming closer. In Teknofest, we need to classify weeds and disinfect them autonomously (without human interaction).

The test day of competition we saw the competition field and you cannot imagine our faces :) There was nothing close the instructions. Every plant was nearly same (you cannot classify them even with human eyes in distance) and I needed to classify and detect just 2 out of 64 plants as weed to disinfect. Well. Crisis must be handled. We took new pictures from the test field. A lot of pictures. Without sleeping, at university I created new dataset with my friend til morning and trained them in our model to compete on the next days. We did it. Yes, we did it. We were the only team that can classify properly and autonomously interfere the plants. However, we got the 3rd place at the competetion.

After we got our results, with lots of experience I earned, I realized that I do not want to do anything with robotics. I really wanted to learn natural language processing and its research.

The Academic Research Era

I started to study natural language processing rest of the summer. I followed https://web.stanford.edu/~jurafsky/slp3/

This book is literally get you inside and teach lots of things.

At this point I felt more comfortable with learning AI stuff even though I still time to time forget most of I learned. Topics increase their connections with other topics for me. Things start making sense that I never had before. I met with inzva which is sanctuary of the Turkish hacker community. I participated Deep Learning and Applied AI study group at inzva. Studying Andrew NG’s deep learning course together was extremely useful to be more involved about math of deep learning. It allows you to learn things in shorter period of time because of the structure of the system.

At the same time, I applied to my advisor’s research project to learn more about process of academic studies. He accepted me as scholar and officially the idea of being part of actual research started to realize itself. I learned how to make reliable tests on machine learning and report them. I still work with the same advisor. I continue to learn variety of advanced topics in both natural language processing and computer vision.

Currently, after constantly working 2 years, I have the knowledge about the math and I know the relations about wide-range of AI gave me the confidence to get a job as NLP Research Engineer as a student. I still have long way to go. I will never stop learning.

I started making contributions to Kaggle with notebooks that includes more advanced techniques and my experience notes. Kaggle is one the most informative site for data-related studies. You can find the codes about almost everything with its free GPU and TPU offers. My Kaggle account: [here]

I am fully motivated by the idea of helping people in the future using AI so I can constantly work even though I fell again and again. I hope you will find yours that will boost you.

Thank you for reading until here. I hope the experiences I wrote will give you an intuition about the learning journey.

--

--

--

Cognitive Science & AI Enthusiast

Love podcasts or audiobooks? Learn on the go with our new app.

Recommended from Medium

Keeping staff onboard and motivated… 101

Qualities of a Top property Broker by Joseph Armato

The Credit Stealer

The V-Interview Mania!

Introducing Coaching for a Living - The Podcast

“A candidate should not have to fit into the company culture; the reason for prioritizing diversity…

Shri Ram College of Commerce (SRCC) Placement Cell rides the crest in the recruitment scene

Shri Ram College of Commerce (SRCC) Placement Cell rides the crest in the recruitment scene

So, how have we been feeling?

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store
Toygar

Toygar

Cognitive Science & AI Enthusiast

More from Medium

Ethics of Artificial Intelligence

The Making of a Rapbot

#1 How AI would respond to the UPSC essay questions?

Artificial Emotional Intelligence