Meet the Team that Won the SP Cup in IEEE ICASSP 2020

You are here

Inside Signal Processing Newsletter Home Page

Top Reasons to Join SPS Today!

1. IEEE Signal Processing Magazine
2. Signal Processing Digital Library*
3. Inside Signal Processing Newsletter
4. SPS Resource Center
5. Career advancement & recognition
6. Discounts on conferences and publications
7. Professional networking
8. Communities for students, young professionals, and women
9. Volunteer opportunities
10. Coming soon! PDH/CEU credits
Click here to learn more.

10 years of news and resources for members of the IEEE Signal Processing Society

Meet the Team that Won the SP Cup in IEEE ICASSP 2020

By: 
Nitin Jonathan Myers

The signal processing cup, called the SP Cup, is an annual competition in the IEEE ICASSP conference. The goal of this competition is to encourage teams of 3-10 undergraduate students to solve real world problems using tools from signal processing. Teams that secure the first, second, and third spot win a prize money of $5000, $2500, and $1500. This year’s SP Cup was sponsored by Mathworks Inc. Interested to participate in the next year’s SP Cup? Let’s hear from the winning team of SP Cup 2020.

The Winner: A team of five people from the Signal and Image Processing Lab (SIPL) at Technion - Israel Institute of Technology won the SP Cup 2020. This team was supervised by Yair Moshe and was tutored by Pavel Lifshits. Undergraduate students Theo Adrai, David Ben-Said, and Samuel Sendrowicz participated in this project.

 

ICASSP 2020 SP Cup Winning Team

 
You can find their project presentation on the ICASSP website. Yair and Pavel were kind to give an interview on their path to winning the contest.
 
Q1. Can you provide an overview of this year’s contest and its stages?
 
This year’s problem has been unsupervised anomaly detection in autonomous systems. Our task was to develop a technique for detecting abnormalities in the behavior of ground or aerial systems based on embedded video and IMU (Inertial Measurement Unit) sensors. We were required to compute a score value for each timestamp of a drone, indicating the degree of abnormality at that timestamp. Since the problem is in an unsupervised setting, no precise labels of anomalies or performance metrics have been provided in advance.
 
The students were required to submit a full conference paper, a technical report, and a reproducible code. The top three teams were selected based on the quality of their project, novelty of the solution, performance of their results, and clarity of presentation. The three finalist teams were invited to participate in the final competition at ICASSP 2020. The final competition was virtual this year – the students recorded a short presentation highlighting their contribution and were asked questions by the judges.

Q2. Congratulations on winning the contest! We’d be happy to hear about your solution on anomaly detection.

Most of the other teams have proposed solutions based on deep neural networks. Such solutions have recently become very popular, but they have several drawbacks, especially when dealing with a small amount of data, such as in this competition. Our solution stands out as it takes a different approach. We embed the noisy high-dimensional multimodal data in a low dimension representation and utilize the correlation between sensors as an indicator of abnormality. First, appearance features are extracted from each image by a pre-trained ResNet-18 deep neural network. Then, appearance and motion features are fused into a multivariate time series. Each time window is embedded in a known cone manifold, using a Gaussian kernel function. The resulting kernel matrix captures the pairwise similarity between each pair of windows. We then exploit the geometric properties of the reduced dimension manifold and use its intrinsic distance as a measure of abnormality. The proposed method is attractive as it is fully unsupervised, data agnostic, and noise-robust. As an example, we have shown that our solution can not only detect abnormal events of different types in the competition dataset, but also in another inherently different dataset.

 

Block Diagram

Block diagram of the anomaly detection technique by Yair, Pavel and their team.

 
Q3. That’s an interesting approach. In your opinion, what were the challenging components of your project? 
 
Participation in the competition was challenging because the given problem is difficult to solve and the students, who are not experts in the field, had to cover a lot of background and professional knowledge quickly while meeting short deadlines. Keep in mind that at the same time, the students continued to carry out all their other academic duties, so they had to be highly motivated and very committed to the competition.
 
All this in addition to the COVID-19 pandemic that has affected the normal course of life around the world, and in particular the participation in the competition. For example, one of our team members went into lockdown, and later all of us stayed home in a lockdown and another team member went to Belgium to visit his family and got stuck there. You can imagine that when we were in a lockdown with our families, it was not easy to work from home and to meet the strict deadlines.
 
Q4. We’re glad that your team won the contest amidst all these challenges. Can you provide insights into your teaching methodology which enabled undergraduates understand complex concepts related to your project?
 
The Signal and Image Processing Lab (SIPL) was established 45 years ago at the Electrical Engineering Faculty of the Technion. It has been since active in research and teaching in a wide range of signal processing topics with many dozens of graduate students over the years and over 50 undergraduate projects each year – many of them in collaboration with industry. So, we’ve got a lot of experience guiding undergraduate projects. Of course, this competition is unique because the timeframe was different, and the effort required was significantly higher than for other projects.
 
To keep a constant progress, we set up weekly or bi-weekly meetings with the students and set a well-understood schedule. We started with a literature survey, read about different approaches to solving the problem, and got to know the given dataset. We then divided the work between the three students and took a fail-fast approach, where the students quickly explored different techniques to allow us to quickly focus on productive approaches. Communication between all members of the team was very important and has taken place in a variety of ways. We also showed our intermediate results to a number of researchers in our lab and asked for their advice.
 
We think the reason we won is first and foremost the motivation, determination and abilities of the students. All of this, together with proper guidance, has created a winning combination.
 
We would like to thank Yair, Pavel and their team for providing information about the SP Cup and discussing their cool solution. Watch out for the next SP Cup’s problem statement which is expected to be announced in October 2020. Good luck for the competition and follow the IEEE SPS on twitter for more updates.

Table of Contents:

SPS on Twitter

SPS Videos


Signal Processing in Home Assistants

 


Multimedia Forensics


Careers in Signal Processing             

 


Under the Radar