The Google Retreat 2013 was held from 30th June, 2013 to morning of 3rd July, 2013. The main activities of the retreat were spread over two days (i.e., 1st and 2nd July, 2013) with 30th June reserved for registration and the welcome reception at the hotel where the scholars and finalists got to know each other through a very interesting networking Bingo. The final day consisted of a very brief breakfast tram tour of Zurich.
Below is a picture of the networking Bingo given to us by Google; for those unfamiliar with the term Bingo it is a card game played in United States and Canada where a 5x5 matrix has to be completed with numbers printed on a board either vertically, horizontally or diagonally. The difference in Google's version of Bingo was however that rather than making it a game of chance it was a game of socializing and networking with other fellow scholars and finalists; and it was great to know that most of them were fan of nerdy TV shows like "The Big Bang Theory" and took nerd as a compliment :-)
The retreat officially kicked off with Oliver Heckman, Engineering Director at Google Switzerland, giving an overview of the engineering initiatives at Google Zurich. Many amazing Google products are a result of hard work by engineers in this Europe-based Google office with some example products being Google Maps, Google Knowledge Graph, YouTube etc. Oliver also demoed the upcoming Google's Conversational Search which seems to be a great leap in the world of Web search engines.
Next up was a technical talk by Doug Aberdeen who holds a PhD with his topic of expertise being Reinforcement Learning prior to joining Google, and within Google he works with the Gmail product team on things like spam detection but more recently on my personal favorite namely "Priority Inbox". His talk was full of valuable insights for those working in Machine Learning which is why I enjoyed it a lot. Doug's talk was different than traditional machine learning talks in the sense that it considered machine learning from a practical and realistic point of view i.e., from point of view of how to approach machine learning when building large-scale products that have to be deployed in the real-world. He said that machine learning people may seem fascinated by the huge amount of data available to Google engineers but the fact of the matter is that even Google does not have ground truth labels all the time and this is where the real Machine Learning challenge comes in. A somewhat astonishing fact for me was that 90% of machine learning algorithms at Google are simple parallel logistic regression; however, parallelizing logistic regression algorithm at Google-scale is definitely something not trivial. Doug's talk was followed by a tech talk on Engineering behind YouTube and how YouTube detects copyrights' violations; it reminded me of the following TED Talk by Margaret Gould Stewart:
Moving on we entered the Product Design workshop which was a fun experience and this activity turned out to be wonderful from a learning point of view giving an interesting insight into product management. We learnt about Google's APM (Associate Product Management) program which is a two-year product management training program specifically designed for those who love managing engineers and coming up with ideas for new products; normally those who are not so good at programming and/or do not enjoy programming enter this line (with lots of those at undergraduate or graduate level). Mind you the product managers are not above engineers in hierarchy as they are simply the people who understand what products people need and then work with engineers to build that product. At the end of the session we were divided into six groups and each group had to work on one of four product ideas; my group got the School Diary idea which we had to chalk out as a product with various features. The following pictures were taken during the product design workshop:
The second day was full of more fun for all of us as most of it had been divided into parallel sessions based on the attendees' year of study and research interests. Following is a list of the parallel sessions with the ones attended by me in bold font:
09:00 - 11:00 Parallel session 1: Android coding challenge
09:00 - 11:00 Parallel session 2: UX web design
09:00 - 11:00 Parallel session 3: SRE Workshop
09:00 - 11:00 Parallel session 4: Natural Language Processing and Research at Google
11.30 - 12.30 Parallel session: Women in Computing
11.30 - 12.30 Parallel session: Mind the Gap
11.30 - 12.30 Parallel session: Employbility Session
14.45 - 16:15 Parallel session: Day in life of an Intern
14.45 - 16:15 Parallel session: Interview workshop
16.45 - 17.45 Career Panels: BSc students
16.45 - 17.45 Career Panels: MSc students
16.45 - 17.45 Career Panels: PhD students 1
16.45 - 17.45 Career Panels: PhD students 2
Perhaps the session on Natural Language Processing and Research at Google was one of the most awaited and popular one with most of the attendees opting for it. During the one hour Natural Language Processing session, Enrique Alfonseca who heads Natural Language Processing division at Google, Zurich gave a talk on his recently accepted ACL2013 paper in which a headline generative system is proposed that can augment Google's Knowledge Graph. The problem is motivated by the observation that news headlines are rarely objective and every news agency reports an event differently. From a computational perspective, such noisy headlines make it hard to detect events thereby making it a significantly challenging problem to augment event-based knowledge bases such as Google Knowledge Graph. The proposed model exploits event relatedness in news collections through dependency parsing on syntactic patterns using a Noisy-OR Bayesian network. Those interested can read the full paper here. Next up was a panel discussion on Research at Google with David Harper (one of Bruce Croft's PhD graduate). This was a highly interactive panel with research scientists (who were once renowned academics) giving insights into what it's like to work on real-world products/systems used by millions of users around the world; turns out it is a whole new experience with satisfaction far more different than joy of getting your research published. I asked two significant questions during this panel from point of view of my own plans of a research internship during PhD and my ambition to remain in academia. At the end of the panel session David Harper mentioned an important resource that gives a very detailed description of how Google approaches research; it is a Communications of the ACM article that can be accessed here.
We then entered the Women in Computing panel which was very interesting for women Computer Scientists. This mostly centered around the question of how women engineers at Google manage an engineering job in industry with kids. Google, Zurich has a flexible policy for mothers-to-be and up to 8 months of maternity leave are granted; along with that there is an option to opt for part-time work along with the option to work from home. Moreover, it is up to the woman herself how she manages the engineering role with her kids and it all comes down to priorities; for a woman kids are always the priority as a Google engineer very nicely put it, "Engineering work can be done by someone else but only I can be a mother to my child". Then another interesting perspective that came up was with respect to quality time being spent with your kids; according to one woman engineer at Google when you know you are always with your kids you take it bit lightly and the quality of the time you spend with them suffers whereas if you are working you know that all the time you spend with your kid has to be quality time. Moving the focus a bit I asked without taking names of course about the assertions by some women in CEO positions that very few women are in those roles and what were the thoughts of women engineers at Google on that to which they replied that it's all up to a person's priorities, CEO positions don't matter that much as long as you enjoy your work and life both.
In the panel session on Day in the Life of an Intern we were told about the work routine in various intern positions at Google. There are basically three intern positions at Google: APM (Associate Product Manager) which has to do with managing products at Google thinking of new features etc., SWE (Software Engineering) which has to do with programming behind Google products, and SRE (Site Reliability Engineering) which concerns site administration to keep the Google site up and running round the clock. A typical day of an APM intern involves loads and loads of meetings with engineers, discussions on certain features of products, a lot of email communications and among other things motivation boosters for the product team. A typical day of a SWE intern involves programming on the tasks assigned to him/her for the most part with little or no administrative stuff. A typical day in the life of a SRE intern involves being on wait and rushing to situations when a complaint arrives regarding the site being down.
The last session I took was Career Panel: PhD Students 2 which mainly centered around career options that PhD students can take once they are done with their PhD. There was a very interesting friction of academia vs. industry in this panel session with some of the panelists making honest confessions of missing academia specially interaction with students and the joy of getting research published while also accepting that one of the strongest motivations in moving from academia to industry is money. In an industry such as Google things are done differently with less freedom to work on things of your choice (like in academia) and the style of work is product-centric rather than research-centric; you cannot afford to solve a research problem in its entirety as the product release has a certain timeline which has to be met. Note that this is different from the other Web industry giants like Microsoft and Yahoo! which both have a separate research division while Google has merged research scientists with engineers in all of their product teams in order to meet the ambitious goal of "organizing the world's information and make it universally accessible and useful."