My Journey from SRM to National University of Singapore

Landing a research internship in an esteemed university is a task in itself. Given the fact that it can be a herculean task to get one, thanks to CG I was able to pursue my interests in the field of Machine Learning and Deep Learning at National University of Singapore.

Inherently speaking, the thought of going overseas is a scary one, especially to those who have never really stepped out of the country. On top of that, not going on a vacation and rather embarking on a journey to advance one’s academic profile is an even greater task.

To give a brief overview on the importance of Machine Learning and Deep Learning in today’s world, we need to retrospect and think about how exponentially machines have transformed our lives. Starting from basic decision making to automation of thousands of tasks by softwares like Ansible, we now rely on machines for effective yet near perfect execution of tasks that would’ve taken us a lot more time.

Being a first-time visitor, I was rather relieved to see how welcoming the people of Singapore were. Starting from basic assistance to helping in terms of navigating through the spiderweb of a train network, they did it all with a smile. Coming to the NUS campus, it is one of the most brilliant pieces of work in the entire country; an ideal university, one might say. Call it luck or a calculated move, I got to experience Singapore in all of its might right before the pandemic struck.

The course I pursued in a 3 week long internship was majorly focused on the theoretical and practical applications of Machine Learning and Deep Learning. It was focused at individuals, primarily undergraduate students who had a rather intermediate knowledge on the topic and wanted to take their intellectual prowess to the next level.

The primary goal was to understand why Machine Learning, which is explicitly programmed can be used as a precursor to Deep Learning. In essence, Deep Learning heavily relies on unsupervised learning. To aid with the process, the University has few of the brightest minds in the entire world, selflessly teaching beginners to jump start their data science career.

A trivial yet vital part of the learning was learning the practical viability of the technology we harnessed and how feasible and ethical it can be. In retrospect, it’s astounding to fathom how universities usually neglect the business aspect of any project that they expect their students to complete. NUS being the stark opposite, gives major importance to the practicality, usability and the ethicalities of any prospective projects that the students come up with.

Being an internship focused on both the theoretical and practical aspects of ML/DL which lie in the ambit of artificial intelligence. We were also given the task to make research projects which were to be coupled with appropriate business models so as to explain the usability and practical need of the model that we developed.

A team, usually of 5-6 people, consisted of individuals with differing skill-sets and knack for the different aspects that the project was supposed to cover. To give a brief overview of the project, we worked on creating a Convolutional Neural Network (CNN) to analyze and recognize human emotions through still images as well as through the live camera feed. We used concepts that involved image classification, labeling the unlabeled data, image recognition and computer vision.

To explain in absolute rookie terms, we used the Visual Geometry Group (VGG) to create a Deep Learning model that analyses human emotions through visual expressions, (such as happy, sad, angry) and then uses these recorded details to provide an insight into various different avenues as and when needed.

It all started with us learning about the initial machine learning models what data crawling is. Once we learnt about how data is crawled and its ethical implications, we went on to make an image crawler for ourselves. Essentially, its purpose was to find images off of the internet which are desired to our individual needs and eventually use these images as training and testing data in the project that we were going to create.

When I talk about crawled data, it is imperative to remember that the data is not completely clean. It always ends up being riddled with noise (like watermarks and unnecessary text) so collection of data is just one of many steps needed in the project. After we collected enough samples off of the internet, we had the Herculean task of data cleaning upon ourselves. What this essentially meant was that we had to individually go through images and make sure that we remove (to the best of our abilities) all of the noise that was present in the image which could lead to possible data noise and accuracy problems.

As I am talking about clean data, it is only logical to point out that clean data is very vital for any model to show the correct and precise accuracy. This is the reason why there are clean datasets on the internet which are considered ideal for the testing of models as they have been cleaned over time and are near perfect. (ex: the kaggle datasets).

Moving along, the rest of the project was rather streamlined and easy to incorporate as we already had a working model of VGG-16, all we had to do was incorporate our custom datasets to test it out. Unfortunately, we did not achieve the best of accuracy with our own custom data set; so we ended up mixing our custom images with precreated kaggle data set images for enhanced accuracy.

A striking feature of the project created, was the business model proposition. As I mentioned, technical as well as business details carry equal importance. The business model consisted of various different categories.

Along the lines of the mentioned categories, we had to devise viable yet crucial steps so as to create a unique project. Namely, they were:

• Key partners
• Key activities
• Key resources
• Value propositions
• Customer relations
• Channels
• Customer segments
• Cost structure
• Revenue streams

Along with these, we also had a product development life cycle which highlighted the growth and maturity of our project, as a product. For a brief overview of the PDLC, we talked about how our product goes from the introductory stages, to the growth, maturity and decline phase.

We also had to devise methods to make sure that we keep away from the decline phase and re-enter the growth phase. Another notable mention goes to the consistent internet and network provided throughout the campus at all times which ensured that our tasks of training the model over google colab ran without any speed breaks. To sum it up, both the technical and non-technical aspects of the project were covered in great detail.

By working on the model as well as thinking about the business implications, we need to also think about our target audience. As the model we created has a potential to expand on a large scale, we also discussed the industrial applications of the model and how we think it would truly benefit the society.

To outline a few, we deliberated upon how an emotion detection system can be vital in schools for young children where we know that individual attention is more or less crucial but due to the strength in each class, it becomes a task to monitor the mental health of each child. Using the model, we could work on the behavioural patterns displayed by a child over the course of a few weeks or months and act upon any sudden or unpredictable changes in their emotions.

By now you might’ve realized that the internship as a whole was extremely intensive but thanks to the faculty and the teaching, it felt rewarding and not as an enforced task. We also had the weekends to ourselves to explore the country and visit pretty much all of the landmarks.

Being a festive month in general, visiting Singapore in the December-January can be a lot of fun. There’s a lot of places you can visit, such as Marina Bay Sands, Universal Studios, Sentosa island, Merlion, botanical garden, the flyer and a lot more. A rather handy tip to make sure you cover most of the attractions is to get an EZ link card for the MRTs. The country has an equally bustling nightlife with a lot of nightclubs, especially in Clarke Quay and the Marina Bay Sands area.

The New Year’s Eve is usually celebrated near the Marina Bay Sands (MBS) area and people gather from all over the country. In all of its might, it looks something like this.

In all, the internship was a very exciting, rewarding as well as an enriching experience. The exposure an individual gets is unparalleled and the teamwork that is built between complete strangers over the course of a few weeks, acts as a catalyst when it comes down to being a team member and also aids in understanding how to adapt to the different types of people that a team could comprise of. In conclusion, the entire ordeal rests in the hands of the individual.

There was an unending inflow of information and whether one chooses to absorb it all or let if flow away is their decision. To give a very personal opinion, I am more than glad to have attended the internship as it not only helped me understand the intricacies of the Deep Learning realm, but also made me understand and imbibe the vital qualities needed in surviving this competitive day and age.

ABOUT THE AUTHOR