Internship at MPI-SWS and a summer to cherish

I spent summer of 2018 (May – August) as an intern in Prof. Krishna P. Gummadi’s research group at MPI-SWS in Saarbrücken, Germany. It was an enthralling and enriching experience to work on cutting-edge and high impact research problems, interacting with some of the best researchers and PhD students in the world, and also making some amazing friends from across the globe. In this blog, I’ll try to recapitulate what was the most happening and enjoyable summer for me thus far.

First off, the question which I have been asked very often since I updated my internship status on LinkedIn and FB: how did you get in? Short answer: PK’s recommendation can take you places. Long Answer: I have been working with Precog for the past 2+ years now. My first project at Precog (and with PK) was a collaboration with Rijurekha Sen (who back then was a postdoctoral researcher at MPI-SWS). Our work got accepted to SIGMETRICS’17 as a poster, and then later to ICTD’17 as a full paper, which I believe buttressed my application greatly. I applied to MPI-SWS around October 2017 through their portal along with PK’s recommendation to my eventual advisor, Krishna P. Gummadi. I got an email from Krishna sometime in late December informing me that I had been selected! And that is my long story short. To sum it up, MPI-SWS has an amazing and very competitive internship program for which you’d be competing against Masters and PhD students from some of the top universities in the world like ETH Zürich, EPFL, GaTech to name a few. In such a case, being an undergrad from India, your best bet is to stand out through your work experience and most importantly, your recommendations.

After a tumultuous semester, summer finally arrived. At Precog we had been reading research papers by Krishna’s group and often considered those papers as seminal works in the field. The thought of working with pioneers of their field was exhilarating. I explored multiple projects during my time at MPI. I started by working on adversarial machine learning, then shifted gears to work on data fairness in automated decision making and then finally settled on dabbling with detecting and mitigating discrimination in targeted advertisements. I won’t make this blog insipid by delving into the nuances of these projects, but, in the process of ruminating on these nuances I gained a lot of insights into both technical and philosophical aspects of research some of which have had an indelible effect on me.

On the technical side, I learnt about various notions of fairness in decision making, learnt a great deal about attacks and defences on machine learning systems (particularly neural networks) and even proposed my own attacks and defences on targeted advertisement systems. In addition, we used to have weekly reading groups where we would take up some interesting concept and deconstruct it to get an intuitive feel of what’s going on. However, the most important skill I gained out of my internship was to understand the importance of asking the right questions. Often times in the past, I had been guilty of over-complicating things and selling my work behind a veil of buffed up jargons. However, it was during my internship I realised that science is about elucidating rather than obscuring. After long thought provoking discussions, meetings which would warp time, my outlook towards research has been broadened.

Here’s a picture with the group at MPI. L-to-R: Till; Junaid; Reza; Prof. Niloy Ganguly, IIT Kgp; Prof. Krishna P. Gummadi; Bilal; Myself; Koustuv (intern, PhD at GaTech); Ashmi (intern, MS CS at TUM)

While it may seem at this point that my internship was only work, it was far from it. Being in Europe bestows you to some of the most picturesque places in the world. Being a football fan, in the midst of a world cup, was an icing on the cake. I got to watch the final in Metz, a small football crazy city in France and watched the 2 french open finals: women’s singles and men’s doubles on court Philippe Chatrier. Of course my journey in Europe would be incomplete without mentioning the amazing (and not so expensive) ice creams! I’ll let the pictures do rest of the talking.


Some culinary delights.


French Open (above) and Le Petit France, Strasbourg (below)

To sum it up, I would strongly encourage anyone looking for research internships to apply to MPI-SWS. I sincerely thank Krishna for having me at MPI and PK for recommending me.

Preventing KillFie: A crowdsourced approach

Selfies have become a prominent medium for self-portrayal on social media. Certain social media users go to extreme lengths to click dangerous selfies, which puts their lives at risk. A hundred and twenty seven individuals have died since March 2014 until December 2016 while trying to click selfies out of which 76 deaths were from India alone.

This disturbing trend can be traced to users taking selfies in “dangerous” locations which in turn can be linked to the concept of self representation on online social media. A user will be perceived more bold and adventurous if he posts from a cliff top or in front of a moving train. Such acts of portraying oneself as being a daredevil leads people to go to life threatening lengths to get the perfect selfie. The engrossing hunt for the ultimate display picture momentarily distracts the selfie taker from their surroundings, which might result in tragedies.

Our goal is to get a dataset of all such locations around the globe which are popular selfie spots, but the lives of people clicking selfies at these locations are inadvertently put at risk. We go about collecting this data through an android app and a chatbot(on facebook) that we made. Users can report locations along with the type of risk(eg : height related, water related, vehicle related etc) associated with taking a selfie on that location. Using this data, our app nudges the user (through a notification) whenever the user goes near such a location.

As we become more engrossed in mobile phones, and go to extreme lengths while being absorbed in our phones and digital world, it is possible that we lose sense of our physical surroundings. For instance, there were several incidents when people got injured while using Pokemon Go (an immersive augmented reality application). Tagging such dangerous locations might be helpful for such apps to get relevant warnings from the database.

If you know of a selfie lover or are one yourself, please download this app. Even a single location which you report may prevent many unfortunate deaths. And since what goes around comes around, one day the app might save you from unknowingly putting yourself in danger.

To interact with the chatbot, go to the facebook page and click on send message. Send a location by clicking on the location icon(can be done only through the messenger app) and just follow the instructions given by the bot.

Click here to download the app.

Click here to interact with the chatbot.