Our year-long series of features on the TOP500 continues with this profile of a fascinating spin-off site called TOP500 Predict.
Now in its 18th year, the semi-annual TOP500 list of the world’s most powerful supercomputers has always been the source of much speculation, analysis, and punditry. So when the new TOP500 Predict site launched recently, I was intrigued by the possibilities. I caught up with Martin Kolb, founder of TOP500 Predict, to learn more.
insideHPC: What is TOP500 Predict?
Martin Kolb: TOP500 Predict is a cooperation between TOP500 and delphit in order to give users the possibility to make use of their HPC insight knowledge, and to prove that our underlying algorithms work. Using virtual money, users can bet anonymously on certain questions (e.g. how big the entry level for new supercomputers will be) and win virtual money if they were right. All users are ranked by their amount of virtual money, which will be the basis for the distribution of prizes (such as iPads, iPods, etc.).
The actual predictions are created by our algorithm, which aggregates all bets with all users.
insideHPC: Why did you start TOP500 Predict?
Martin Kolb: Our main goal with TOP500 is to use the wisdom of the crowds to predict certain properties of the future releases of Top500 lists, and therefore prove that a crowdsourcing solution like this works under circumstances which top500 provides, namely distributed knowledge about certain industry-specific trends.
For the HPC Community, TOP500 Predict will be the perfect tool to prove themselves concerning their knowledge and insights, and even have the chance to win prizes. For TOP500, it’s a great opportunity to increase awareness of the project and compare crowd-sourced predictions with their own expectations.
For us (i.e. delphit), it’s kind of a research exercise in prediction markets, and a great reference for our “Social Business Intelligence” system for predicting relevant internal events that we offer to companies. The prediction market algorithms that we use in TOP500 Predict are the basis to our Social Business Intelligence (SBI) systems. The US-Market for prediction-market software is already in a later stage of evolution, but german companies especially still need to be convinced that Social Business Intelligence provides a major improvement to their business.
insideHPC: The HPC community is populated by some of the world’s top minds. Do you think TOP500 predict gives the participants a chance to compete, brain-against-brain?
Martin Kolb:Partially yes – it’s not just a matter of intelligence; the critical success factor is just information – even if it’s only some kind of gut feeling or instinct. Information can be transformed into a bet (which consists of a range of outcomes and a certain amount of money I’m willing to pay).
insideHPC: What does delphit do?
Martin Kolb: delphit is a spin-off of University of Mannheim (Germany) that performs research in the area of social business intelligence and develops prediction markets as part of this research.
delphit sells SBI solutions to companies to support them in operational and strategic decisions, similar to what classical business intelligence does. In comparison to classic BI, social business intelligence uses the knowledge of all employees as its primary source and for most companies, the knowledge that their own employees have is one of their most important resources. We’re helping them to use this resource for decision support in a systematic way, for instance in risk-management, product-development, supply-chain-management and several other fields.
insideHPC: Have you been surprised by any of the predicted trends?
Martin Kolb: We don’t have any clear predictions yet – but even if there would be more traffic right now, we assume that for this first run; surprising trends will only appear shortly before the actual release of the TOP500 List.
To again emphasize this: as “insider trading” (in a manner of speaking, because anyone is allowed to participate, all trading is completely anonymous and no real cash is exchanged) is possible here it might be the case that surprising trends- if information diffuses e.g. in shape of rumors.
insideHPC: I would like a good bet. Is there any wagering involved?
Martin Kolb: You can actually increase your winnings by actively observing and trading with your portfolio (the bets you already placed) and seeing if its value increases or decreases. As you have always the possibility to again withdraw your bet, you can make a profit even before the bet has officially ended.
Here is a simple example:
- Prediction for the Entry Level is at 35 PetaFlops.
- I’m placing a bet, believing the entry level is more likely to be between 36 and 38 PetaFlops. I invested 100 credits and the system tells me that I would win 450 Credits if I’m right.
- Another user now places a bet between 36 and 37 petaflops. Obviously, the overall prediction must somehow be changed into my direction – the value of my bet increases; in other words: if I were to place the same bet again now, investing the same money as before, I would win less than 450 credits. But if I’d withdraw my bet now, I can realize a winning.
In the end, you can act like a stock-trader on our platform: try to use information optimally and use chances to exploit arbitrage wherever possible.