Author: Sai Rama Raju Penmatsa, San Jose State University with input from Ankur Singh, San Jose State University
A couple of weeks ago, I participated in my first-ever hackathon, which Intel hosted. My friend from the research team informed me about the hackathon and we both registered as a team.
Day 0 (Hackathon)
This hackathon was aimed toward real-world problems that can be solved or aided through the power of AI. There were three tracks that a team can select from, which include Machine Learning, Natural Language Processing and Computer Vision. There were prizes for the top 3 teams from each of the tracks and a prize for the best team out of all the 3 tracks. My teammate and I chose the computer vision track. So we were given a binary classification problem - given an image, the model should be able to predict if the image is of a plant or a weed, to implement a targeted pesticide attack on weeds. So with this hackathon being organized by Intel, we had to utilize oneAPI and perform model training on the CPU.
This experience was quite a shocker - the acceleration that was obtained by oneAPI really changed my perspective of what can be actually done on just CPUs. The event kicked off starting with demos from Habana Labs, and later we were given reference kits which we could start with, along with oneAPI-based software for the hackathon. The hackathon lasted for 8 hours. The entire event has not only given me a hands-on experience in problem-solving but also helped me identify the areas where I need to focus more in the future. Moreover, we had the opportunity to meet some amazing people from Intel, who guided us throughout the process. We were able to use some pre-trained models for our track using fastAI, and we finalized our best model and ended the day at 8 PM, hoping for the results to be announced by midnight.
The leaderboard was getting updated frequently after the competition ended as the models were getting evaluated. We couldn’t find our team name on the leaderboard for quite some time, which led us to question if our model had been registered properly for the evaluation. Then finally, our model got evaluated and rose to the top of the leaderboard. Following this, we received an email, where we were asked to reach the venue early the next day, as we would get a chance to have an interview with the Intel CTO, Greg Lavender. My teammate and I were ecstatic upon hearing this news, as this was never expected.
Day 1 (Keynote)
Still in disbelief and euphoric, we reached the convention centre and had some wonderful conversations with Scott Apeland, Rahul Nair, Eduardo Alvarez, Paula Ramos, and Benjamin Consolvo before proceeding to “The MOMENT”, meeting Greg Lavender.
The interview went great, which included how we approached the problem that was given and how amazed we were with the computational efficiency achieved with the usage of Intel oneAPI in combination with Intel CPUs. Post-interview, we were each awarded an Intel® NUC® Dragon Canyon by Greg Lavender.
Following this, my teammate and I attended the keynote by the Intel CEO, Pat Gelsinger. This was my first in-person keynote, and I loved the atmosphere and the nerdiness of the CEO and the audience. The keynote itself was quite packed, with Intel launching their Raptor Lake processors. My favourite moment of the keynote was when the Intel CEO stated the famous quote “Moore’s Law: Alive and Well”, following the announcement of their entry into the GPU segment.
This is a reference to the NVIDIA CEO stating that “Moore’s law is dead”, when asked about the increasing prices of the GeForce RTX® 40 series. This for me is big, as all three major chipmakers (Intel, AMD, and NVIDIA) are in competition now, and we as end-consumers get better products in terms of performance and cost-efficiency. Another standout from the keynote for me was Intel® Geti®, which helps accelerate the model development process, right from data labeling to model training.
In the evening, all the winning teams were awarded their prizes and were given the opportunity to talk about the models they built.
Following this, a networking party was organized by Intel, wherein we got to meet and have some candid discussions with people from the same domain with all levels of experience. We had some amazing food along with games as ice-breakers, giving us an opportunity to talk with others.
Overall, it has been an incredible experience, regardless of the win. I would recommend everybody to participate in hackathons, which can help build self-confidence, provide an opportunity to network with like-minded, and put the previous learnings to practice.
The link to the code has been uploaded as a Kaggle Notebook here.
This post originally appeared on Medium, where you can follow the authors.
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.