The initial open DanKu agreement competition

21 views Leave a comment

Machine Learning algorithms are being grown and softened during an implausible rate, though are not indispensably removing some-more permitted to a broader community. That’s because currently Algorithmia is announcing DanKu, a new blockchain-based custom for evaluating and purchasing ML models on a open blockchain such as Ethereum. DanKu enables anyone to get entrance to high quality, objectively totalled appurtenance training models. At Algorithmia, we trust that widespread entrance to algorithms and deployment solutions is going to be a elemental building retard of a offset destiny for AI, and DanKu is a step towards that vision.

Since we’ve only announced a protocol, we suspicion that it would be also wise to emanate a initial open DanKu contract.

For this competition, we’ve motionless to use a 2016 U.S Presidential county choosing information as a dataset. Every county is represented with 3 information points: longitude, embodiment and inaugurated candidate.

For example, a information indicate can demeanour something like this: [047606200, 122332100, 0].
The initial dual values refers to a embodiment and longitude of Seattle.
The third value, 0 refers to a Democratic candidate, since 1 refers to Republican claimant in this information format

500 pointless information points are comparison as a dataset for this competition. 80% of this dataset will be used for training. The remaining 20% of a dataset will be used for evaluating a dataset. Since a training and contrast datasets are incidentally comparison by a contract, a integrity of a foe is assured.

Participants of a agreement are compulsory to sight a elementary brazen pass neural network where they can conclude a neural network structure in terms of layers, neurons and biases. After training, they are compulsory to contention their network definition, weights and biases to a contract.

The DanKu agreement was initialized on block 5121944 (Feb 19th, 2018). The analysis criteria will demeanour for models that has during slightest 50% correctness rate. The prerogative for a winning acquiescence is 5 Ether (ETH).

You can find a beam here to assistance we attend in a competition. After a foe ends, we’ll muster a winning indication to a Algorithmia marketplace!


Comment this news or article