We are at a crossroads where
- machine learning is at the heart of a technological and societal artificial intelligence revolution involving multiple sister disciplines1, with large implications for the future competitiveness of Europe,
- Europe is not keeping up: many of the top labs and top places to do a PhD are located in North America; moreover, AI investments in China and North America are significantly larger than in Europe, and
- the distinction between academic research and industrial labs is vanishing, with a significant part of the basic research now being done in industry (with substantial research freedom, and higher salaries), rapid commercialization of results, and academic institutions worldwide struggling to retain their best scientists (with negative implications not only for research but also for the education of future talent). This further weakens Europe since most of the companies doing top research in this field are controlled from the US (or China) – many European companies whose future business crucially depends on AI are not perceived as competitive.
There are still a few machine learning & perception research hotspots in Europe that play in the international top league. Virtually all of the top people in those places are continuously being pursued for recruitment by US companies. Even if we only wanted to retain these centers, we would need to increase our investments in line with what other countries are doing. To strengthen our position, we need to build on what is strong in Europe, think big and have the courage to try new models2. We believe our best bet is for the outstanding centers in Europe to join forces.
European strength currently lies in its academic culture and well-educated students. There are some top university departments in the field, as well as some centers of excellence at Max Planck and CNRS/INRIA. Large US players have started research labs in some of those places. While a major motivation for these labs is the competition for local talent, the labs also strongly contribute to the local ecosystems by rendering them more attractive for students and researchers, and educating a generation of high-level professionals, some of who subsequently form startups.
We should found a European Laboratory for Learning and Intelligent Systems (working title; abbreviated as “ELLIS”), involving the very best European academics while working together closely with basic researchers from industry.
The mission of ELLIS is to benefit Europe in two ways:
- we want the best basic research to be performed in Europe, to enable Europe to shape how machine learning and modern AI change the world, and
- we want to have economic impact and create jobs in Europe, and believe this is achieved by outstanding and free basic research, independent of industry interests.
This is how to make ELLIS competitive:
- Outstanding facilities and computing infrastructure.
- It is an inter-governmental organization (like EMBL, the European Molecular Biology Laboratory). While the initial ELLIS letter only listed a small set of countries, scientists from other countries have meanwhile expressed their interest. These are very much welcome, along with outstanding non-EU centers of excellence in countries such as Switzerland, the UK, and Israel.
- ELLIS comprises labs in the partner countries at the top academic sites for machine learning & perception research. This allows jump-starting ELLIS by means of (short or long term) co-affiliation and/or secondment of outstanding academics. Excellent researchers across each country may be connected via fellowships, and the links to local research institutions are vital for ELLIS to thrive
- It runs programs for visiting researchers (both from academia and industry), as well as workshops and summer schools for students, academics, and industrial participants. Mobility is facilitated by housing, childcare, and (international) schools at each site.
- It aims at building a European PhD and MSc program in cooperation with degreegranting universities. The participating degree-granting institutions will allow and encourage students in their MS and PhD programs to spend time in ELLIS partner sites, with no additional tuition charge, and co-supervision from researchers at these sites. ELLIS will provide fellowships to support this program
- ELLIS researchers can split their time between ELLIS and local university or industry research labs (creating an incentive for industry to co-locate). Collaboration with industry is encouraged and structured using transparent and simple IP rules that ensure that public funding is used in a way that benefits the public. Joint research involving industry and public funding is openly publishable.
- ELLIS researchers can found startups based on IP they generate. ELLIS does not aim to optimize short-term licensing income, and rather aims at sustained economic impact in Europe. To this end, it owns a modest share in those startups and claims no further rights as long as the startup is formed in a partner country, thus generating downstream impact (including jobs) in Europe. ELLIS supports startups in terms of (a) generous leave-of-absence rules, (b) temporary use of infrastructure, and (c) help with administration including legal/financial advice.
- ELLIS does not need a large headcount for personnel initially (since it strictly only recruits top notch academics), but it does need a long term funding commitmentincluding a plan how the funding ramps up. Each local lab could aim to reach at least the scale of a major Max Planck institute, i.e., around 100 Mio EUR for infrastructure and an annual budget increasing to 30 Mio EUR during the first ten years.
- Existing funding structures are too slow: ELLIS should start in 2018 3.
- In addition to researchers and faculty from the partner institutions, ELLIS will offer permanent employment to outstanding individuals early on and train them in both academic and non-academic skills. These researchers will receive an adjunct faculty position from one of the partner institutions. They will also be offered a complete career path within ELLIS, paralleling those found in tenure-track programs, from the equivalent of the rank of assistant professor to that of a full professor. This will be a major step towards avoiding brain drain to the US.
- ELLIS’ unique characteristic is outstanding academic quality as measured for instance by publications in the leading competitive conferences of the field5. It does not preclude other national and international activities that focus mainly on applied research and industry cooperation, but ELLIS’ pure mission of excellence in basic research must not be compromised.
- There is no shortage of funding for AI research, but it is extremely hard to attract outstanding researchers. However, it is the quality of the individual researchers that determines the strength of the overall lab, and only top people act as true talent magnets. US institutions and companies have recognized that money spent on those people pays off in multiple ways. In Europe, there are currently only few types of academic positions that allow us to attract such top people, e.g., Max Planck directorships or full professorships at ETH. Our only chance to attract such people to ELLIS is to offer positions with outstanding academic freedom and visibility (in cooperation with Max Planck, ENS, CNRS, INRIA, as well as participating top universities), with top packages6.
- Since the field holds great economical promise, there may be competition of different sites to be home to an ELLIS lab. The only criterion should be academic excellence.
- ELLIS will perform fundamental research in modern AI, attract top international industry research labs, and spawn startups that will become major players in the future. It will thus drive excellence in Europe’s research and use of machine intelligence to foster economic development and improve the lives of people.
- ELLIS will be a top employer in machine intelligence research, on par with Berkeley, Stanford, CMU, and MIT. It will also be a world class venue to get trained in the field: in conjunction with universities, it will develop a highly attractive European PhD program, and it will strive to retain the best graduates within ELLIS to groom them into the next generation of senior scientists.
- Taken together, this means that Europe will be able to play a major role in the scientific and societal revolution that is underway . The first and second industrial revolution not only transformed technology but also led to fundamental societal changes. These changes were managed by European democracies and values. The current revolution may be equally significant. Europe should benefit from it and European values should help shape its impact.
Note: The following text was added after the original ELLIS open letter (which was published on 24 April 2018). It incorporates discussions with CLAIRE, an initiative that published an open letter on 18 June 2018.
RELATIONSHIP BETWEEN ELLIS & CLAIRE
We have agreed with the CLAIRE consortium that we will endorse each other’s proposals, acknowledge their complementary nature, and coordinate efforts while preserving the unique characteristics of the two proposals. ELLIS and CLAIRE both emphasize the importance of excellence in fundamental research because economic leadership relies on technical leadership, especially in AI.
In recent times, machine learning has revolutionized many areas of AI, generating unprecedented momentum and huge interest in the field. The ELLIS open letter was the result of ongoing discussions among a group of machine learning researchers who got increasingly concerned that Europe is not keeping up in the area of machine learning/deep learning and its application areas such as computer vision, speech processing or natural language processing. Our letter was also read by researchers from the AI community at large, including researchers whose focus is on areas of AI not directly impacted by machine learning, and inspired them to call for a decisive investment across Europe, covering all areas of artificial intelligence. Their CLAIRE initiative aims to establish a network of research labs and organizations across Europe, along with a new, central facility without permanent scientific staff, providing cutting-edge infrastructure and support to top talent in AI from all of Europe. The ELLIS signatories believe that it is important to retain focus on the learning-based approaches that we feel (1) are the engine of the ongoing revolution and (2) are our best chance to approach human-level general intelligence.
In the time since publishing the ELLIS open letter, we have received a large number of messages, most of which strongly support the idea of establishing ELLIS. We below address some questions that came up across the board:
- How can I support ELLIS?
First of all, add your name to the list of endorsers. Also, feel free to speak up: talk to the press in your country, talk to decision makers in your country, and tweet or blog about ELLIS. You may also provide us with quotations that we can use, or send us a letter of endorsement. At a later stage, we will seek donations and industry sponsorship.
- How can we become an ELLIS site and/or how can my country become part of ELLIS?
Our plan is that each partner country provides their ELLIS facilities/infrastructure. Prerequisite for participating in ELLIS are long-term guarantees for the below:
Partner countries may choose to embed ELLIS facilities into larger infrastructures including researchers employed by local/national institutions, as long as such arrangements are deemed mutually beneficial. These researchers will usually be recruited nationally, however, all ELLIS recruitments are done solely by ELLIS to ensure a uniform bar of excellence across all of ELLIS (see below).
- subscribing to ELLIS' overall philosophy, i.e., open basic research benefitting the public good, adoption of ELLIS governance and excellence standards
- outstanding lab/office space, provided free of charge (incl. for ELLIS visitors)
- guest housing (for individuals and families), childcare and international schools, all free of charge for ELLIS visitors
- local university courses are offered free of charge to visiting ELLIS students (including course credit)
- How can I join an ELLIS lab? ELLIS recruits fellows and students, and hosts visitors. By default, ELLIS positions are not associated with universities, however, co-affiliations (e.g., as adjunct faculty) are encouraged.
Full-time appointments range from tenure-track positions right after the PhD7 all the way to director positions (comparable to full professor). All full-time employees must have their main work location in an ELLIS institute, and they are encouraged to spend time as visitors at other ELLIS institutes. For tenure track appointments, the tenure decision is taken within four years after recruitment. Senior positions are immediately tenured, unless they are secondments from other positions, in which case limited term contracts can be offered.
Part-time fellows are outstanding researchers whose main employment is outside ELLIS (and anywhere in Europe), either in an academic or industrial lab, and who normally spend at least one month per year in an ELLIS institute. Part-time fellows are not necessarily tenured. IP generated during their residence at ELLIS generally falls under the same rules as IP generated by full time ELLIS members, and all results are openly publishable. If part-time fellows are in parallel working with their home institution, exceptions will be made for such projects, provided the fields of work are transparently defined and delineated.
PhD or MSc students working at ELLIS get their degree from a European university. PhD students spend at least one year at an ELLIS site different from their main location (which will often, but not necessarily, be an ELLIS site itself).
In addition, ELLIS hosts short- and long-term visitors. This includes the possibility of secondment, during which researchers join ELLIS temporarily (paid either by ELLIS or their home institution). We anticipate making heavy use of this at the beginning to help cold-start ELLIS.
ELLIS has an international advisory committee (AC) consisting of outstanding scholars in the field who are not themselves members of ELLIS. ELLIS fellows and students are selected by means of an international search process. For ELLIS fellows in any given country, the recruiting committee consists of peer researchers located in other countries, the majority of which should be ELLIS fellows (as well as, during the start-up phase, AC members). Since all candidates will thus be evaluated by fellows located in other countries, this mechanism will ensure that the highest quality standards are applied in recruiting. It will also ensure that no ELLIS fellow faces political pressure from his national colleagues to be made an ELLIS fellow. The ELLIS initiative is open to all researchers in Europe, but will hold everybody to a high bar. Since part-time fellows can be located anywhere in Europe and sponsorship will be sought from the EU, we expect that ELLIS will raise the level of excellence across all of Europe, having an effect akin to the outstanding LMB (Learning in Machines and Brains) program of the Canadian Institute for Advanced Research.
The host country pays for local ELLIS infrastructure and running costs. For the fellows' program and studentships, we will seek funding from the EU, host countries, foundations, and industry donations.
- What are the next steps?
To enable an agile path forward we currently pursue two tangible ways of collaboration with which we can incrementally build a European research center of excellence:
- Following the highly successful Canadian CIFAR program “Learning in Brains and Machines” (LMB7) we are going to establish an ELLIS fellowship program that aims at building a community of researches that is maximally visible at NIPS, ICML, CVPR, and similar conferences.
- In addition to fostering collaboration and providing some unrestricted funds to facilitate agility, we work towards establish a joint PhD Fellowship program following the model of the Center for Learning Systems and the Cambridge-Tübingen PhD Fellowship program. That is, we will start to jointly recruit PhD students for collaborative projects supervised or co-supervised by ELLIS fellows working at different locations. The students will divide their time between two locations and spend at least one year at each of them.
We use the term machine learning to include areas of AI that are strongly influenced and driven by machine learning, such as much of computer vision, natural language and speech understanding, and parts of robotics.
European governments are beginning to realize this, as shown by the recent establishment of the Alan Turing Institute as well as the new French AI strategy. Countries like Canada and Japans are taking action to address the challenge of retaining top AI researchers; and Canada’s Vector Institute is an exciting model of what can be done.
It can build on institutional links between some of the top European ML sites ( http://learning-systems.org/, http://mlg.eng.cam.ac.uk/?page_id=1458 ), as well as a cooperation between MPI-IS/Cyber Valley and PRAIRIE.
For Germany, such an analysis has been performed in the recent EFI expertise, see https://www.e-fi.de/fileadmin/Gutachten_2018/EFI_Gutachten_2018.pdf
EMBL offers attractive packages that come with special conditions ( https://www.embl.de/jobs/work-at-embl/ ), but the field of machine learning & perception is more competitive. Co-appointments with industry will help significantly.
The term ‘tenure’ in this document does not refer to a university process, but to an analogous process to turn temporary ELLIS appointments into permanent ones.