From Mars to machine learning
A project that Aalto University PhD Student Tuomas Kynkäänniemi was working with at the summer 2017 has reached worldwide attention, as part of the Perseverance rover that just landed on Mars on Thursday night.
In 2016 as an undergraduate student at Aalto university studying engineering physics, Kynkäänniemi got a summer research job at the Finnish Meteorological Institute, working in what is now the Planetary Research and Space Technology Group at FMI. Kynkääniemi’s job was to help with the calibration of the temperature and pressure sensors that make up MEDA a Martian weather station that takes a wide range of atmospheric measurements that FMI helped develop. ‘The sensors give an electrical signal, and my job was to take that electrical signal and turn it into a ‘real world’ unit, like degrees celsius’
Kynkäänniemi’s work now is “on another planet” as he puts it, from his engineering work from his student internship days. He now works in the Computer Science research group of Jaakko Lehtinen, developing machine learning methods called generative models for generating highly realistic images. ‘When I finished my job with FMI, the field of machine learning, and generative models were very exciting and interesting to me. I got an internship at Nvidia Finland and started working in this area instead, so when it came to choosing what to focus on for my Master’s project, I swapped from engineering physics to computer science.’ Professor Lehtinen also worked at Nvidia as a principal research scientist, and Kynkäänniemi did his master’s project with him developing a method for evaluating the performance of AI that generates images.
‘A model that generates images needs to meet two criteria,’ explain Kynkäänniemi, ‘they need to be realistic looking, but they also need to be highly variable. It’s no good if your model can only turn out one realistic looking image and they all look the same. It needs to make a wide range of realistic looking pictures.’ Kynkäänniemi's master's thesis was given the MSc Thesis Award of 2018-2019 by the Finnish Society for Computer Science, and was also published as a paper for the NeurIPS 2019 conference.
Kynkäänniemi describes his research path from undergraduate to now as a bit of a ‘random walk’ but he hopes it inspires other students, and people thinking about studying technology at university. ‘I hope my example shows that if you are open-minded and curious, you can get the opportunity to work on lots of very different and fascinating projects’
Read more
For more information about the FMI involvement with the Mars Perseverance lander, you can read about it on their website
You can read Kynkäänniemi’s master’s thesis here: https://aaltodoc.aalto.fi/handle/123456789/39830
Link to research article:https://proceedings.neurips.cc/paper/2019/file/0234c510bc6d908b28c70ff313743079-Paper.pdf
Forbes AI Award to research on image-creating GAN models
The magazine pays tribute to the development of a model that starkly reduces the amount of data needed for generating new, artificial images
A student satellite project launched alumna Maria Hieta’s career in the space industry
Alumna Maria Hieta majored in Space Science and Technology.
Read more news
Join Unite! Seed Fund Ideathon in Austria
Travel and accommodation are covered for accepted students. Apply by 6 October.
‘Off to a flying start’ – a new research team explores bacteria that thrive in extreme conditions
Jarno Mäkelä joined Aalto’s Department of Neuroscience and Medical Engineering as Assistant Professor of Biophysics in early September, together with research fellow Taras Redchuk, postdoctoral researchers Dan Noel, Eveliny Nery, doctoral researchers Ada Pajari and Aakeel Wagay, and research assistant Chris Hayes. They were accompanied by equipment, funding from the Academy of Finland and an ERC Starting Grant from the European Research Council.
Best Student Paper Award for Eloi Moliner – again!
The award-winning paper shows how speech recordings can be improved by removing the room reverberation effect using unsupervised machine learning