VBlog - June 2021

Its been awhile since I last posted on LinkedIn and have been pretty much off-the grid for past year. Lots changed in the last year (both globally and personally). In the spirit building in public and tracking interests I'm kick-starting this monthly VBlog (Vaibhav's Blog). It'll essentially track what have I been up to, things that caught my attention and some of my immediate bets/ goals.

Life

I quit my full-time consulting job at Deloitte in October 2020, packed my bags and moved to Stuttgart, Germany to pursue my MS in Computational Linguistics.

It has been a huge shift both in-terms of responsibilities and the overall control I have over what I want to invest my time in. I also took up a Working Student job at Deloitte's Tax Management Consulting group here in Stuttgart, where I am helping automate data workflows and building data analytics frameworks.

I've taken a break from public speaking, hopefully I'll be able to get back to it as I settle in to Deutschland.

Research

After spending an entire semester filled with exams and assignments, I decided to solely focus solely on applied research for the next 6 months.

• Abductive Natural Language Inference - Predicting whether a given observation is related to context. Trying to beat the benchmark with transformer models.
• Interpreting Attention Models with Human Visual Attention - Comparing attention scores in Transformer models with eye-tracking datasets. Focusing on BERT and GPT family.
• Approximating attention matrix to reduce quadratic memory and time footprint - Using SoTA methods like Nystromformer and Linformer for Text to Speech tasks.
• Context based Question Answering - Extending an existing library to include sentence based transformer methods for semantic search and similarity. Still refining this, more on this next month
• Design Thinking experiment with Mercedes - Envisioning the next generation of UX/ UI capabilities for Mercedes S class of the future (2030)

What caught my attention?

Apart from work and research whatever little time I get, I spend on going through my social and academic feeds and below are just couple things which I found interesting

• Bitcoin - Apart from El Salvador making BTC a legal tender and the price of the asset, there's lots of interesting stuff happening on the BTC chain, BTC is likely go through Taproot activation, which not only helps makes the chain more privacy preserving but also faster (What is Taproot?)
• How normal am I? - Essentially an interactive app which employs myriads of on-device AI algorithms to help you understand how advanced these models really are.
• Huggingface now supports vision models - I absolutely love huggingface and every time they support a new model/ class it just makes my day. It has become a defacto library for pretty much all my projects (Speech and NLP)
• Mike Bostock (creator of D3.js) open sources a JS library for Exploratory Data Analysis - Observable Plot, On the same page he also makes a pretty strong case for how on-device last mile analysis can help bridge the friction between data and insight. Watching this space closely!

Volunteering

Some of my most enjoyable work. Currently helping with sponsor management with EuroPython 2021, The conference is sadly online, but man are we planning a killer conference. Tune-in to the action around July end.

I also happen to be one of the Student Representative for my MS program and am currently working really hard to get a snooker table within our lounge (the most important task of all)

Immediate Goals

I'd like to focus on making my research public and accessible to others (Put it up on my Public GitHub?), revamp my old website, finish my Pimsleur level 1 course and run at least 60Kms this month.

Thank you for reading! I promise I'll be more regular with updates, in the meantime if you have any feedback on how I can make this more interesting for you then hit me up in the comments or on vaibhavs10[at]gmail[dot]com