4 Months into deep learning - NLP

It has been almost 3 months since I joined Sentient.io on the 21 July. One of the most valuable highlights is the belief in my supervisors and mentors that this internship should be constructive. This means I would be given the opportunity to get hands on with machine learning projects. The key goals my mentors set for me in this internship are to discover if AI is the field I truly want to pursue and to instil the motivation to add value across various industries with the skillsets I attained.

# A little introduction of my company

Sentient.io is a 3 year old start up, aiming to democratise AI technology by providing AI-as-a-service platform. Our mission is to empower enterprises and software developers to harness the power of AI technology. The purpose in making AI (or technology in general) more accessible and beneficial to the society resonates with what I always believe in. It has been the work I long to pursue - the process of turning theoretical knowledge to build products which tackle real-life problems. The open culture in Sentient.io encourages interactions between all team members. I am fortunate to be blessed with such a lovely team and caring mentors guiding my machine learning journey every steps of the way. The freedom to schedule meetings with my mentors or even the founder is an unique experience to learn from their sharing and clarify my doubts.

Photo by Ian Schneider

# The work I do

Working as a science intern under the science and research team, I'm assigned to the project of text summarization. Automatic text summarization is a challenging problem under the field of Natural Language Processing (NLP). It is the process of filtering the most important information from the input source and generating a concise version of it without losing its original meaning.

Here I will be building a encoder-decoder model that is able to take in a source document and produce abstractive summary.

# Applications of text summarization

Before starting to work on the project, it'd be meaningful to know the applications of text summarization, why are we doing text summarization and how can it benefit us.

Some of us who have tried summarizing a text would understand the pain in distilling key information from a lengthy document or text. The process of manual text summarization is tedious and time-consuming. This effort is compounded when the number and size of the document increases, multiplying the labour intensity needed to complete this task. Besides, the quality of summarization may deteriorates as human performance is influenced by other external factors. With automatic text summarization, these drawbacks can be eliminated.

There are various ways text summarization can be used across the industries. One common text summarization case we encountered daily is the news highlights and another could be the generation of newsletters for organisations.

# Opportunities and challenges

Having barely to dabble with machine learning, I am keen to stretch myself beyond my comfort zone and to conquer the steep learning curve. The new skills I have acquired can be broadly classified into two sections - the mathematics behind deep learning and python programming. Essentially, machine learning is heavily driven by mathematics and statistics. Although many of the libraries such as Tensorflow, Pytorch and Keras have simplified complex mathematics operations into a single, easy-to-use function, it is best for a data scientist to have a clear understanding of the model which is mathematical in nature. This area poses a huge hurdle to me due to the knowledge and experience gap in AI. There are times when I am able to run a piece of code smoothly without having a clear sense of what is happening behind the machine. This includes identifying the input and output dimension of the neural network layer.

Python is a popular programming language widely used in the industry because of its simplicity and powerful application. Working on my project, I have the opportunity to gain exposure to the Tensorflow and Keras library. This is definitely something new and exciting for me. Coming from a background with basic coding skills is not sufficient to fulfil the task given. It is necessary to hone my skills by looking into online resources and documentation. For instance, I have to find out about Python classes and objects, how to use them and incorporate into my code. The ability to absorb new knowledge plays such a prominent role as I was figuring out things on the fly. Nevertheless, tying both the theoretical and technical skills into real-life application has always been rewarding. Since the methods of achieving the state-of-the art performance has been evolving rapidly, learning how to learn is the only way to keep up with the speed of AI revolution.

While grasping these concepts can be daunting, I have some fun scrolling through AI memes

# Keep diving!

AI is poised to disrupt our lives, businesses and economies. While the intrinsic intricacy of AI is often intimidating, I am so glad to be get involved in the AI scene. Every research paper and online article I read leads me to a realm of new discoveries. I can now better relate to what is going on behind our chatbots, recommendation system and the annoying advertisement that keeps popping up due to my search history. Is it fun to swim into a sea of unknowns? Definitely not. But the adrenaline to explore the beauty under the sea keeps me going. And the only way is to dive into it.

"In the growth mindset, failure can be a painful experience. But it doesn’t define you. It’s a problem to be faced, dealt with, and learned from.” -Dweck

Discomfort forces growth. Overcoming new challenges and gaining new insights allow me to unlock new opportunities. And what is more wonderful than working with a team of motivated problem solver who share the common purpose on adding values across various industries with the power of technology. I'm inspired to learn from each individual's unique gifts, talents and skills.

Photo by Greg Rakozy

# Little reflection

It has been a roller coaster ride as I have been feeling a mixture of emotions throughout this journey. The stress of not being able to understand a concept, the struggle to produce code which performs the task accurately and the joy in fixing a probelm can happen all at once within a day of work. This internship burst the glamorous bubble of a life as a data scientist, the advantages of remote working come with the promise of meeting deadlines and the ease of launching a successful project is only possible with strong proficiency in the area. It requires deliberate practice and having a growth mindset. While that is the harsh reality, it reaffirms my passion and keen in learning about data science/machine learning. The work at Sentient.io from the research phase to the engineering phase, the drive in solving problems with technology, providing a customer centric solution and the communication skills to liase with various stake holders have further convinced me in this specialization.

It doesn't matter which sector or industry you're in. Go and explore different sectors and understand how they operate. The key is to add value in any of these industries with the skills you have.

I still believe in that and I will continue to do so.