1/2018

On the cusp of everyday AI

Assistant Professor (tenure track) Esa Rahtu joined the Laboratory of Signal Processing at TUT last autumn. He specializes in computer vision and machine learning and dreams of seeing intelligent machines hit the mainstream.

Esa Rahtu

 

"A machine that could replace humans or even accomplish a wide range of tasks is yet to be invented,” says Assistant Professor Esa Rahtu.

 

WHO: Esa Rahtu (born in Rovaniemi, Finland, in 1979)

  • Assistant Professor (tenure track). His area of expertise is intelligent machines.
  • Education:
    Doctor of Science in Technology 2007, University of Oulu
    Master of Science in Technology 2003, University of Oulu
  • Career:
    Adjunct Professor in the University of Oulu since 2014. Before joining TUT, Rahtu co-founded the indoor positioning company IndoorAtlas Inc. in 2012. He worked as the company’s CTO for several years and currently serves as an advisor. He has previously held a post-doc position funded by the Academy of Finland in the Machine Vision Group at the University of Oulu and worked as a visiting researcher in a group headed by Professor Andrew Zisserman at the University of Oxford for about two years. Rahtu has also worked as a research assistant and doctoral student in the Machine Vision Group at the University of Oulu.
  • Family and hobbies:
    Wife and two lively daughters aged 4 and 7. Enjoys sports, badminton and jogging.
 

Computer vision is one of the key areas of artificial intelligence (AI) research, because we live in an extremely visual world. For people and machines to work together, machines and systems must be able to interpret, understand and act on visual data. Modern devices already generate diverse data about their environment, but the lack of AI-powered methods for analysing this data is preventing its broader utilization. Esa Rahtu, the newest professor to join the tenure-track faculty in the Laboratory of Signal Processing, wants to contribute to making everyday AI a reality.

“Intelligent machines have long been on the horizon – everyone can see their potential but their practical deployment has largely remained unsolved. A machine that could replace humans or even accomplish a wide range of tasks is yet to be invented,” says Assistant Professor Esa Rahtu.

“There’s still a long way to go before machine learning technologies are so good that they can be widely utilized in everyday environments and not only to perform narrowly defined tasks in specific facilities, such as factories. It’s exciting to take part in developing methods that can genuinely help people at home and in the workplace.”

Both universities and companies are racing to develop AI, and the field is advancing rapidly. For example, deep learning, which was still unheard of while Rahtu was at university, has now come into the mainstream.

“The field of AI has undergone major changes even during my own career and is picking up even more speed as industry interest grows. A number of large corporations, such as Google, Facebook, Microsoft and Apple have invested heavily in the development of AI and machine vision. The future potential of AI has also attracted widespread interest in Finnish industry,” says Rahtu.

“There are only a few areas where both academia and industry are keen to pursue fundamental research. It is great to see that companies are also participating in AI conferences and sharing their knowledge.”

What will people do next?

The research conducted by Esa Rahtu focuses on the latest machine vision technologies that are necessary for the development of intelligent machines. His research interests cover, among others, image-based localization, motion analysis of mobile devices, automatic object recognition, video content analysis, and the development of AI-based image generation.

“These areas are present in our daily lives, for example, through self-operating machines, augmented reality, games, robotics, and intelligent production systems,” says Rahtu.

Machines must learn to anticipate future events and communicate better.

 

Self-driving cars and other autonomous machines must be equipped with sophisticated sensors and have the ability to interpret what they ‘see’ in order to interact with people.

“Machines must learn to anticipate future events and communicate better. While they are already capable of recognizing, for example, how many people, animals or objects are portrayed in a picture and even estimate the age of the people, they struggle to predict our actions and assess what happens next. Machines also have to learn how to communicate their intentions to people in a natural way,” describes Rahtu.

“People are amazingly adept at reading other people’s body language, for example, while driving a car: is that pedestrian too lost in thought to keep an eye on traffic, can they see my car, or are they just waiting on the side of the road? Robot cars are still not very good at this. Researchers who develop machine vision are looking for ways to teach machines to work more effectively and safely with people.”

Rahtu is among them, as one of his research interests is human pose estimation that teaches machines to detect our body posture. Some fine-tuning is still required, although the technology is already used, for example, in games.

TechBites

30.5. TechBites:
Robotics and AI for Industry

Grab your TechBites of the future of robotics and AI technology and its applications! Build contacts and join discussions with key people from companies and research teams. The event is organised by Tampere University of Technology.

Who’s it for?

Private businesses and decision makers who would like to benefit of the ongoing digitalization for industry and society.

How does it all happen?

Registration opens soon! Register online by 21 May 2018. Check the full details techbites.fi

 

“Machines can now easily identify the position of our hands, but what about our fingers? In many cultures, the hands are almost as expressive as words. To really understand human behaviour, it is important for machines to notice these little details.”

Ordinary or exceptional?

Another area where further development is needed is video content analysis, which is currently a subject of intense scientific interest. Computers are taught to identify people in videos and recognize what they are doing.

“The amount of video content available online is mind-blowing. What if we could search for specific video content using keywords, the same that we do with texts? The amount of available information would grow exponentially. The technology could also be used to automatically create informative video summaries,” describes Rahtu.

One of the challenges is that machines need to be taught a great deal of things that we humans take for granted.

“It’s no simple task to teach AI to interpret situations that are immediately anomalous or surprising to us humans. For example, if a photo depicts a mouse sitting on top of a cat's head, people can guess what happens next, but machines cannot see that anything is out of the ordinary.”

Broader research questions can only be solved through collaboration

Besides Esa Rahtu, the Laboratory of Signal Processing welcomed two new tenure-track professors last year, namely Taneli Riihonen and Heikki Huttunen, who was already a familiar face on campus. Rahtu’s Artificial Intelligence and Vision Group maintains close collaboration with AI researchers from the other laboratories at TUT. Rahtu is also keen to collaborate with robotics researchers and, for example, with Roel Pieters who took up a tenure-track professorship in the Laboratory of Automation and Hydraulic Engineering last autumn.

“TUT is genuinely committed to developing and investing in important research areas,” Rahtu says.

Research visits to the Visual Geometry Group at the University of Oxford, one of the world’s leading groups in the fields of vision and AI, have also made a lasting impression on Rahtu. The visits offer first-hand insights into what other researchers are doing and where the field of AI is headed.

“Collaboration with colleagues and researchers who represent other fields is crucially important. It enables us to look beyond our own narrow field and address broader research questions. This helps turn AI from a vision into reality.”

Text: Sanna Kähkönen
Photo: Mika Kanerva

 
Tell a friend
Machines outperform the human ear
Machines outperform the human ear
1/2018
Machines outperform the human ear
More devices, more data – and critical research problems
More devices, more data – and critical research problems
1/2018
More devices, more data – and critical research problems
Pure entertainment and critical safety
Pure entertainment and critical safety
1/2018
Pure entertainment and critical safety
Open a door to assist in the development of smart homes
Open a door to assist in the development of smart homes
1/2018
Open a door to assist in the development of smart homes
What does the future hold for developers of intelligent systems and software?
What does the future hold for developers of intelligent systems and software?
1/2018
What does the future hold for developers of intelligent systems and software?
First thought in the morning: Let’s do it!
First thought in the morning: Let’s do it!
1/2018
First thought in the morning: Let’s do it!
Keep your feet on the ground when you go to the cloud
Keep your feet on the ground when you go to the cloud
1/2018
Keep your feet on the ground when you go to the cloud
Serious injuries should not be overlooked
Serious injuries should not be overlooked
1/2018
Serious injuries should not be overlooked
School groups attended the European Robotics Forum: “The best day ever!”
School groups attended the European Robotics Forum: “The best day ever!”
1/2018
School groups attended the European Robotics Forum: “The best day ever!”
New ultrafast measurement technique shows how lasers start from chaos
New ultrafast measurement technique shows how lasers start from chaos
1/2018
New ultrafast measurement technique shows how lasers start from chaos
Campus IoT network now available
Campus IoT network now available
1/2018
Campus IoT network now available
Immersive Media goes Disruptive!
Immersive Media goes Disruptive!
1/2018
Immersive Media goes Disruptive!

Tampere University of Technology is at the leading edge of technology development and a sought-after collaboration partner among the scientific and business communities. The University produces competent graduates who enter careers in the different sectors of society.

Visiting address
Korkeakoulunkatu 10,
FI-33720 Tampere
Finland

Mailing address
PO Box 527, FI-33101 Tampere
Finland

Switchboard:
+358 3 311 511