Harvest Programme

PASCAL is very successful in shaping and stimulating research in Machine Learning: we want to make it as successful also in interacting with the outside world. Machine Learning has many uses, and many results from PASCAL researchers could have great applications. While the vast majority of them is published, a paper in a scientific journal or in some conference proceedings is not always the best way to give ideas the appropriate exposure. Understanding a NIPS paper requires highly specialized training, and people with needs and opportunities that could be addressed by such results could be lacking the right skills. Moreover, many published results are very abstract: going from them to actual applications requires knowledge of the domain and of practical aspects of the problem the author of a paper is rarely aware of.

The main idea behind Harvest is: put together in a room a team for long enough to produce an innovative software for a real application. PASCAL2 will pick up the bills.

A Harvest project proposal should address the following points in 4-8 pages:

• Problem description: what is the project going to do? Why is it interesting? Why is Machine Learning relevant?
• Specification and validation: How will performance be assessed?
• Expected staffing, location and duration. How many persons will be needed? With what skills? For how long?
• Milestones, timeline.
• Requested funding.
• Content of the training, if any, that will be delivered to participants.

The team will meet and work full-time on the project for 30-90 days. To make sure no time is wasted, though, it is important that everyone is clear about what needs be done from Day 1. This in turn requires that some preparation is done to define objectives, assign roles, read relevant literature etc. Here is what the life cycle of a typical Harvest project could look like:

1. Anytime: proposals in search of participants (and possibly location) are published on the Wiki. Project leaders conduct team selection.
2. M-3.5: Call opens.
3. M-2.5: Call closes.
4. M-2.3 to M-2: proposals are reviewed by independent experts.
5. M-2: funding decisions are notified.
6. M-2 to M: The teams prepare for the on-site project execution.
7. M to M+1/M+3: On-site project execution
8. M+1/M+3 to M+5: The project is reviewed
9. M+5 to M+?: The project team prepares a demonstration
10. M+?: Projects executed in the year are demonstrated either in a dedicated Harvest workshop or in margin of another PASCAL event.

The execution phase should take place in the European Union, or in a country hosting a PASCAL 2 member. Participants who are not normally located where the execution takes place will be refunded a return flight in economy class from the place where they live, and will be entitled to a per-diem for accommodation, meals and local travel. This per-diem will vary from country to country, and will be indexed on the per-diem granted by the EC to refund its contractors. As of 2011, this means (in euros):

Australia: 86 Austria: 113 Belgium: 116 Bulgaria: 114 Czech Republic: 115 Cyprus: 119 Denmark: 135 Estonia: 92 Finland: 122 France: 123

Germany: 104 Greece: 111 Hungary: 111 Ireland: 127 Israel: 101 Italy: 115 Latvia: 106 Lithuania: 92 Luxembourg: 119 Malta: 103

Netherlands: 132 Poland: 109 Portugal: 102 Romania: 111 Slovak Republic: 103 Slovenia: 90 Spain: 106 Sweden: 129 Switzerland: 119 United Kingdom: 138

Who will have the right to do what with the outcome of the project? When designing the Harvest Programme we had three "scenarios" in mind:

1. Open-source project. In this scenario the "user" side of the consortium is, for instance, some public research group in a field clearly separate from the scope of PASCAL 2 itself (e.g. Astronomy, Medicine, ...). Participants have their salary paid by their employers, and the Harvest programme covers the whole cost of travel and subsistence. For this scenario the objective should be to give maximal exposure to all results, possibly releasing the software to the general public.
2. Joint research and development of an application or a component. In this scenario, a combination of industrial and academic partners contribute complementary skills to the development of a component or an application. Participants have their salary paid by their employers, and the Harvest programme covers the whole cost of travel and subsistence. In this case it is important to ensure that the industrial partner can dispose of the output for further developing it, but also that the academic partners retain rights for research and education purpose, and that scientific results can be published.
3. Sponsored research. In this scenario one or more industrial partners fund the project at a significant level, possibly covering part of the salaries of all involved personnel. In this case more exclusive rights on the produced software could be envisaged, as well as some control over the publication of research results.

In order to limit as much as possible the burden on prospective coordinators, the Harvest programme can provide support in drafting the necessary "Harvest Project Agreement".

So, why should you participate in, or -better still- propose, a Harvest Project? Here are some possible answers:

• You can obtain resources for pursuing ambitious applied projects
• You'll work with focus and critical mass on a project you care about
• You'll see your research results applied in the real world
• You'll get better known
• You'll receive good training from the best experts
• You'll get in touch with potential employers
• You'll acquire hands-on experience on an advanced research project
• You'll show the impact of your research
• You'll improve your attractiveness over talented students
• You'll establish durable relations
• You'll come out with some cool software, and maybe some patents
• You'll get financial support on risky research projects
• You'll get in touch with great people to hire
• Maybe this one will become the next Facebook

We were inspired by a number of sources:

• Johns Hopkins University CLSP summer workshops
• The NineSigma innovation broker
• Intel Research Network of labs
• Available information on the European Institute of Technology
• The Newton Institute
• WebValley
• The Netflix Prize
• Google's Summers of Code
• Henry Chesbrough's book « Open Innovation »