Student Challenge winners develop neural net for images

The winning team from Sweden’s KTH University, comprising (left to right): Sandra Picó, Miquel Larsson, Quentin Lemaire and Daniel Del Castillo.
5 February 2019
The winning team from Sweden’s KTH University, comprising (left to right): Sandra Picó, Miquel Larsson, Quentin Lemaire and Daniel Del Castillo.

Winners of the 2018 Huawei European Student Challenge developed a neural net to identify images using Artificial Intelligence algorithms, at a hackathon organised by Huawei Sweden at KTH University in Stockholm in December. The team put its success down to “keeping it simple”. Here we interview them about their project and what they think the future holds with AI.

Could you provide a short description of your winning project for the layman – which problem did your project solve, and how? How is it using AI to achieve this?

Miquel: One of the common tasks AI algorithms are used for is to identify images. To do this, a neural net (inspired by the way the brain works) is trained by showing it lots of examples of images and what they contain (labels). The proposed challenge was: how can we use a large number of unknown images (without any labels) in order later to train this neural net on fewer examples, without its effectiveness being compromised? 
Our solution consisted first in training a neural net on the unlabelled images, from which we were capable of extracting a compressed representation of the images (shapes, lines, etc.). This first part of the neural net can be taken advantage of when training with the labelled examples, reducing the amount of these we need to create a working neural net able to identify the content of images.
    
How did you benefit from the competition? Was it a useful training/learning experience? Did you enjoy the collaboration with your peers?
       
Sandra: The whole competition was an incredible experience. Because time was limited, you needed to solve the challenge posed in a short time and efficiently, together with your colleagues. Then, the competition was a completely learning experience really useful to learn how to solve problems as quickly as possible. 

Daniel: It was my second hackathon and my first one related to Machine Learning, so I got a whole new dimension of practical experience, such as new tools and platforms, new prototyping techniques, new ways of quickly finding solutions, etc. Also, a quite useful insight into how these competitions are organised and executed. We are all good friends in our team, and we were in a great mood that day, so we worked as a well-oiled machine. It was lots of fun!

What are your plans for the next months and years?

Miquel: At the moment, to finish the Master in machine learning I am currently pursuing at KTH (Stockholm), after which I expect to find a job in the field.

Sandra: First, to finish my Master in Machine Learning at KTH next June. Then, try to find a job related to AI and improve and learn as much as possible about it. 

Daniel: I plan to write my Master thesis on Language Identification from Raw Audio Files with Deep Learning at a company in Stockholm. After that, my idea is to find an NLP-related job that fulfills my expectations either in Stockholm or in a different and exciting city!

Why did you get into coding?

Miquel: I rapidly fell in love with programming after I discovered it in my first year of university. It actually feels like playing around with digital Lego blocks.

Sandra: I discovered programming during the first year of my Bachelor in Electronics in Communications. 

Daniel: I wouldn’t say I’m a coder in the narrow sense of the term. But when I first learned C programming during my first year of university, I felt like coding was an amazing and useful way of exploiting my imagination and creativity.

In your opinion, how can AI change the world? How can we make sure it’s a force for good, triggering positive changes?

Daniel: The current AI we are experiencing, or “narrow AI”, has already proved itself to be a successful solution provider for very specific and hard problems that were abandoned in the past, such as machine translation, self-driving cars, or image recognition. But I believe it’s the integration of all these concrete capabilities into unified and more general systems what will change and even reshape the world as we know it. 

Sandra: AI has enough power to change the world so that it becomes a better place. However, without regulations and proper laws, the effect could be the other way around. We need to set up and define proper rules to ensure that AI is used as a force for good. 

In which areas do you see AI as having the greatest impact on people’s daily lives in the near future?

Miquel: AI is already having an impact in all aspects of our lives, but probably the most significant and visual one will be when autonomous cars are fully developed and deployed.

Sandra: AI will have impact in most areas. However, in my opinion, medicine and self-driving cars will be the areas to feel the greatest impact. 

What would be your advice to tech students who consider getting involved in the challenge?

Miquel: Two quick rules for programming competitions: keep it simple (things will always go wrong), and; get an initial basic pipeline with a benchmark solution working as soon as possible.