Sean Benhur | AIcrew Stories
🤝🏼 Welcome back to AIcrew Stories!
AIcrew blog series share real-life machine learning journeys of our fellow challenge winners and also to some extent explore the winning solutions that got them the title. The aim of the series is to in some way inspire future participants and encourage them to partake in these challenges.
🧭Sean’s Machine Learning Journey
Hailing from Coimbatore, Sean is currently pursuing a Master's Degree in Software Systems at the PSG College of Arts and Science. He got introduced to Machine Learning in the very first year of his college through a course offered by IIT Madras.
The course focused on detailed theoretical knowledge before diving into the practical implementations of the various algorithms. At first, Sean was intimidated by the sheer amount of Mathematical prerequisites the field requires and so he dropped out of the field.
However, in his fourth semester, Sean’s interest in AI rekindled and he took the famous Andrew NG's Machine Learning course on Coursera. From there on out, motivated by Andrew's words, Sean took up a plethora of ML and DL projects. You can find his projects on his Github. Sean currently works as a Machine Learning Intern at Impiger Technologies.
🙋🏼♂️Getting started with Challenge
Sean was introduced first to AIcrowd last year through the buzz created by the Procgen Benchmark Challenge. This NeurIPS challenge was hosted on our platform by OpenAI as a way to measure sample efficiency and generalization in Reinforcement Learning. Curious to know more? Click here!
NeurIPS Procgen Benchmark 2020
Exploring the platform back then, Sean decided to sign up for the AIcrowd Newsletter. As an AIcrowd participant, Sean received the challenge update email newsletters. Through them, he eventually got introduced to the Blitz challenges. He signed up for the three-week-long, 5 puzzle format as it was manageable and worked well with his schedule.
🎙Sean's Advice to you
Sean rebukes his own experience and learning pace that was allowed by his ongoing studies. Sean talks about the constant juggling between curriculum and self-learning during his college life. However, he advises with a steady progression in the ML course and regular implementation of concepts through projects enabled him to get familiar with ML concepts in 3 months.
“Submit whatever baseline you can come up with”, says Sean. His advice for beginners jumping into the competitive ML challenges is to be proactive and come up with a solution. “Once the standings are clear, assess your performance, research where you lacked, and come back with a better submission”, he adds. Sean employed this technique to achieve the highest score in one of the puzzles in Blitz 9, Emotion Detection.
♟About his submission
Sean was greatly inspired by the community posts and previous community contributions by people in various AIcrowd Challenges. Sean has noticed people often used techniques without knowing the logic behind them especially in the field of Natural Language Processing. He tried to address this by creating an in-depth walkthrough on transformer and transduction models. His notebook explained how transduction models that rely entirely on self-attention to compute representations of its input and output, like BERT help with text encoding and feature extraction.
In his submission, Sean gives us a walkthrough of different methods for data cleaning and feature engineering for conventional approaches to state-of-the-art methods in NLP. Look out for a detailed breakdown of his submission in the upcoming Challenge Unboxed📦!
What stories do you wanna read next? Comment down below or Tweet us @AIcrowdHQ!