AI research competing with giants

“Giants such as Google, Facebook and Apple are buying up top-flight researchers in artificial intelligence. In the academic world we find it difficult to compete”, warned Professor Michael Felsberg from LiU at the WASP4All conference.

Close to 300 attended WASP4All Michael Felsberg, professor of computer vision at LiU, giving his presentation at WASP4All. David Isaksson

WASP, the Wallenberg AI Autonomous Systems and Software Program, held its first large open conference with invited speakers and around 300 participants from research and industry. Michael Felsberg, professor of computer vision at LiU and a leading Swedish AI researcher, described the development of the computer vision research field, which has taken huge strides forward in the past few years. Before the systems can be used in everyday life, however, several advances are required.

“We want systems that learn from experience as new events occur, such as autonomous cars that can distinguish between a car, a cyclist and a wheelchair, and that later learn how new types of road-user, such as e-scooters, behave.

We construct our environment model primarily on what we see and experience, and if robots are to function together with people they must have the same ability”, he said.

Segmentation

For object tracking, which is the process in which a computer recognises and follows a moving object, LiU researchers in computer vision have until now used rectangular regions in an image, a bounding box. The computer can use deep learning to learn how to recognise the different parts in the box, with respect to both colour and shape.

“We can now manage this in so close to real-time that we believe that the computer has immediately learnt to follow an object”, says Michael Felsberg.
Research is now, however, starting to increasingly use a technique known as “segmentation” in which pixels are classified as belonging to different objects. For example, the computer should be able to distinguish a sleeping dog from the sofa on which it’s sleeping and the cushions around it. The segments are also simple to visualise with the aid of Michael FelsbergMichael Felsbergcolour.

All of the various methods require huge amounts of computing power, preferably using computers that consist of graphics processors, GPUs, which are advanced parallel processors with large memories. Such a national resource is in the process of being established at Chalmers University of Technology, financed by SNIC (the Swedish National Infrastructure for Computing).

“We need computing resources to be able to compete. A doctoral student in computer vision requires, on average, GPU resources that cost around half of what the student receives in salary. Until now, we have been compelled to purchase expensive computer power from suppliers stipulated in our framework agreement”, says Michael Felsberg.

Top-flight researchers bought up

One danger that he sees is that huge multinationals such as Google and Facebook are buying up top-flight researchers in AI. Also Apple, whose purchases previously tended to revolve around technology (such as the purchase of the Linköping company C3technlogies for 3D maps), has started to employ leading scientists from the academic world. Another example is the Uber company, which has its own research-based subsidiary, Uber AI.

“The giants, of course, have resources that we in academia cannot compete with. So far, researchers employed in industry have been allowed to publish their results, but I wonder how long that will continue. What happens as competition becomes stiffer?” asks Michael Felsberg.

“Academia must have the resources required to educate undergraduates, postgraduates and young researchers – both for its own needs and those of industry. We need to discuss which research questions and areas in which the academic world can compete, given the limited resources we have.”

WASP investment

Professor Anders Ynnerman, from next year programme director in WASP, gave the opening speech at the conference. He took up the problem of securing the resources required for competitive AI research in his opening speech, and referred to the report he has authored, together with colleagues from LiU, the Royal Institute of Technology, Lund University and Umeå University, which discusses the infrastructure needed for AI research. Later, he gave further details:

“The extensive investment under the auspices of WASP enable us in Sweden to lie in the global forefront of the field, which is developing rapidly. We can greatly advance Swedish competitiveness through recruiting international experts, organising graduate schools, and other WASP initiatives. We have also taken on the responsibility of ensuring that the computing power needed in the AI field is available, and powerful computing resources will come online during 2020.”
Anders YnnermanAnders Ynnerman

WASP is the largest individual research programme ever to be carried out in Sweden, with a budget of SEK 4 billion in the period until 2026. The Knut and Alice Wallenberg Foundation has contributed SEK 3 billion. Its objective is to develop world-leading knowledge – research and education – within automation, artificial intelligence and software in order to promote Swedish industry.

“The programme has several components: research, recruitment, graduate schools, collaboration with the best universities in the world, and arenas in which research and industry meet in common projects”, said Anders Ynnerman.

Extensive recruitment

Since it started in 2015, seven professors within autonomous systems have been recruited; 10 Wallenberg scholars within artificial intelligence have been selected; 13 associate professors have taken up posts within artificial intelligence; and 238 doctoral students have been employed. Forty companies are involved in the programme in various ways.

“We are planning an announcement for vacant positions as industry-based doctoral students within AI, with the application period to open in January 2020. We have an organisation to deal with innovations that are developed within the programme, and we have projects to study the human and societal aspects of artificial intelligence. These are collected in WASP HS – humanities and the social sciences – a separate initiative with which we collaborate closely”, said Anders Ynnerman.

A steady stream of speakers from the academic world and industry gave presentations at the conference. They spoke on such topics as the development of graphics processors, traditional processor technology, quantum computers, supercomputers with a capacity of exaflops, and quantum cryptography. The influence of digitalisation on society was also discussed.

Coffee break WASP4AllWASP4All: an opportunity for many interesting discussions. Photo credit David Isaksson

Translated by George Farrants


Contact
Show/Hide content

News about AI infrastructure
Show/Hide content

Research
Show/Hide content

Latest news from LiU
Show/Hide content