Munich Diaries #6 – Joining TUM.ai1805 words • 9 minute read
After saying goodbye to Bori on Wednesday, it was straight back into being productive.
On Thursday, an email landed in my inbox: “Congratulations, you made it into TUM.ai!”. I was above and beyond to become an official member. I couldn’t wait to join the Kick-Off weekend next week and meet all of the team.
The rest of the week flew by really fast. I caught up on some (pre-recorded) lectures, completed my weekly deep learning assignment, met up with my study group, and wrote a (very-very basic) file system in C.1
On a side note, we also had the first “dropout” (no pun intended) from our Introduction to Deep Learning study group, although we were lucky to find a perfect new member for our group instead. Other than this, I feel extremely lucky to happen upon such great, motivating, and knowledgeable teammates for all of my groups I happened to be a part of out of pure chance.
A workshop about interculturality
Next Monday, I participated in the first part of a two-part Online Intercultural Communication Workshop I signed up for earlier. It is organized by TUM in cooperation with ENSEA Paris, a leading engineering university in France. During the 2-hour workshop session, we learned about different aspects of cultural diversity, dimensions, and frameworks through which to assess how interculturally competent a person is. The educators actively facilitated everyone’s participation, starting with turning on cameras and microphones. It resulted in insightful breakout-room discussions about cultural diversity. It was great to listen, participate and gain insights into for example how decision-making in French companies usually work, and what the role and image of a leader traditionally are there.
At the end, we also assessed James Bond’s intercultural competence based on a short clip from “The Spy Who Loved Me”, which proved to be a fun little exercise.
All in all, it was actually one of the most enjoyable online workshop sessions I took part in.
My second consultation
I had my second consultation almost a month after my first one. During this one month, I tried to familiarize myself with all the basic concepts on the topic of rendering participating media. As I already wrote about it, it’s mainly about rendering phenomena such as smoke, fog, clouds, and fire. To be more precise, it’s about modelling the interaction of light rays and particles that make up these volumes. Current methods model these interactions as a change in the light ray’s direction and radiance (or “color”) as it tries propagating through these media. This change in the light ray’s direction and/or radiance is attributable to collisions with tiny-tiny particles making up the smoke/fog/cloud/fire/etc. In essence, as these modified light rays end up reaching our eyes (or more precisely, our virtual camera), we end up seeing the phenomena at hand.2
Of course, as you start formalizing this, you pretty quickly have to utilize many tools and methods of calculus, linear algebra, and probability theory3. This time around I had so many questions that we couldn’t fit all of them in our allocated time and settled on meeting in a weekly manner going forward. It is always extremely motivating to have my questions answered and to be able to receive intuition and guidance in even the simplest things, like understanding advanced notation. Overall, I always feel extremely thankful to have the time and attention of a knowledgeable domain expert and adept professor during consultations.
In cooperation with KNUST, the biggest university in Ghana, TUM.ai organized the AI4Good Workshop. University students from Ghana could register to participate in the project. Although some of them had a background in Telecommunications, Informatics, or related fields, most of them had completely “unrelated” backgrounds, which really helped to widen the horizons of the potential domain of solvable problems considered by the teams.
During the first “Education” week, they received a lecture series about an overview of the current capabilities of AI, evaluation of AI systems, and developing real-life solutions to problems present around us.
Starting this Monday, the workshop rolled into its second “Venture” week, where teams were formed amongst the participants, each working out a solution to a real-life problem in Ghana utilizing AI technologies. Although an implementation was not required, it was important to develop a feasible end-to-end solution, which included outlining things like where the data comes from, how reliable it is, whether certain types of sensors are needed to install, and what drawbacks or potential difficulties might arise during the implementation of the project.
TUM.ai members acted as “AI Helpers” for these teams, helping the ideation process, discussing the feasibility of the proposed solutions, and overall steering the brainstorming phase towards a realizable AI solution that solves a real problem in Ghana.
As a new member of TUM.ai, I felt extremely grateful for the opportunity to be able to act as an AI Helper for two of the teams. We set up 1-hour AI Helper sessions via Zoom for both teams, helping to bring the members of each team together under the same concept, and provide feedback on their ideas, outlining the current reach of AI technologies. During my 2 AI Helper discussions, we covered a great variety of topics with both teams, ranging from outlier detection to deep learning techniques in computer vision and the feasibility of implementing robots and beyond. One of the teams settled on detecting the misconduct of drivers on the roads, and the other one steered from the pollination of cocoa flower4 to automating a part of the exam grading process at universities.
Finally, on Friday, all of the teams pitched their ideas. Although only of the teams I helped ended up in the final round, it was great to see all of the teams bring forth all these realizable project ideas utilizing already available AI technologies. The winning team presented a solution for automatically tracking the spread of malaria. It was a once again great eye-opener to the unbound potential of applications of not only AI systems but creative, innovative, and solution-oriented thinking in general.
At the end of the day, it was a great first-hand personal introduction to how TUM.ai facilitates, motivates, and investigates the interdisciplinary use of AI technologies, while also lowering the barrier of entry for the opportunity to involve more non-tech-savvy domain experts – a really important aspect which I came to appreciate even more after later discussions with other TUM.ai members. And not to mention upkeeping a dependable, efficient, and joyful community of members in the background.
A “still great” (but actually awesome) Kick-Off Weekend
“I can‘t believe that a 5 hours long online event was actually fun throughout."
-Someone on the online side of an online event
The original plan was to get away somewhere for the whole weekend together, but as is the case with many good things these days, the concept of the weekend had to be reconsidered. It was finalized as a hybrid gathering. In the morning, we had an online start together, and then met up with our respective departments to get to know each other and start our joint work.5
After that, as we made the event 2G+ (which means that you have to be either vaccinated or have proof of recovery and have a recent negative test result), I went to do an official rapid test (which is provided for free) at many sites. Luckily it was negative, so I made my way to our “department house” where we met in person for the first time with most people from our departments. Some could only join online.
It turned out to be quite an intercultural gathering of students from Turkey, Brazil, Albany, Germany, Vietnam, and Hungary as far as nationalities were concerned. We also covered quite a variety of disciplines, approaching Informatics from vastly different perspectives. We were students of Robotics, Politics, Electrical Engineering, Business Administration, Human-Computer Interaction, and Computer Engineering. It was a fun experience to get to know the different viewpoints, knowledge, and experiences each of us was bringing to the table. At one point we realized that the usual circle of introduction stretched out to multiple hours, steering off in the most interesting and thought-provoking ways possible while talking about really fascinating subjects and aspects of AI and related possibilities.
By the end of the day, I felt like I just met some of the loveliest, most energized, motivated (and motivating) teammates. I’m really excited to work and learn with them, and I’m looking forward to what will come out of all of this.
We closed the Saturday with an in-person part. We went to a very traditional-looking beer house. It was a huge hall, with broad wooden tables laid down. It was a really fun night of getting to know new people, hanging out, and having a lovely time overall.
Remember how I wrote earlier about taking on the 10 ECTS worth of “Advanced Systems Programming in C/Rust”? Yeah, starting with this exercise, the true meaning of this statement really started to sink in. Although it requires a lot of patience (and tolerance for pain), it is actually truly fascinating to get more familiar with a specific aspect of the inner workings of the system(s) I use on a daily basis week after week. I might gather some of the more interesting things I got to familiarize myself with in a separate blog post. Anyways, for the interested reader (and just to be precise), the file system assignment was about writing a very basic, in-memory one (no messing with kernel modules) with FUSE (Filesystem in Userspace). ↩︎
If you are interested in the topic in more detail, I can recommend the “Production Volume Rendering SIGGRAPH 2017 Course” and the 2018 report “Monte Carlo Methods for Volumetric Light Transport Simulation”. Or just drop me an email. I’m always happy to discuss anything I write about here. ↩︎
Probability theory comes into the picture because we model the particles inside the volume as a probability field, denoting either a collision or scattering event happening. This works by assuming the collision events to be independent of each other. This model of course has its limitations though. It works only as long as the size of the particles is much-much bigger than the space between them. We can’t model this way for example granular media such as sand sugar or even properly shaped snowflakes. ↩︎
I actually just learned during this discussion that Cocoa accounts for almost 20% of Ghana’s total exports. ↩︎
TUM.ai is organized into Functional and Mission-based Departments. Functional Departments are IT & Infrastructure, Legal & Finance, Community, Marketing & PR, and Partners & Sponsors. On the other hand, Mission-based Departments are the Makeathon, Venture, Industry, and Education departments. Without going into too many details, each department contributes a crucial aspect to what makes TUM.ai a great and fresh student initiative. ↩︎