Results

Results - see a publication submitted to Biological Cybernetics -

Challenge A

In the above table we see the ranking of participants along with their average performance (middle column) and bootstrap estimation of standard deviation (right column). There were 3 statistically equivalent solutions:

- Shinomoto & Kobayashi (Autoregressive exogenous model with moving threshold),

- Badel (Exponential integrate-and-fire with dynamic threshold),

- Mensi (spike response model with moving threshold)).

The winner submission is decided according to the best average performance:

2nd prize : Shigeru Shinomoto.

You can download the ‘answers’ of the test set here: V.txt

--------------------------------------------------------------------------------------------------

Challenge B

Split 2nd prize: Tie between Kramer (hand tuned Traub model), Hirschi and Naud (adaptive Exponential Integrate-and-Fire), Acker (Izhikevich’s simple model) and Druckmann (Hodgkin and Huxley type model tuned with genetic algorithm).

This challenge was difficult to win since the winning solution had to achieve the best cost for each feature of the test set. To display the results we have kept the best submission of each participant. We can make a global ranking using the criteria of last year’s challenge (http://icwww.epfl.ch/~gerstner//QuantNeuronMod2007/). Briefly, we take C to be the sum over i of ci*(0.5)i with ci being the cost associated with each feature in decreasing order of cost (c1 is the cost of the worse feature, c2 the second worse, etc.). With this ranking scheme, the results are:

1. C = 9.5 Hirschi - Naud (adaptive Exponential Integrate and Fire)

2. C = 11 Druckmann (Hodgkin-Huxley type model)

3. C = 38 Acker (Izhikevich’s simple model)

4. C = 112 Kramer (Traub model)

You can download the test set ‘answers’ here: ChallBTestV.txt.

----------------------------------------------------------------------------------------------

Challenge C:

1. Ryota Kobayashi C1 = 0.499 C2 = 0.427

2. Shigeru Shinomoto C1 = 0.476 C2 = 0.379

3. Richard Naud C1 = 0.309 C2 = 0.168

Challenge D:

1. Richard Naud D1 = 0.408 D2 = 0.446