July 19th Announcement to Participants

We would like to express our gratitude to all participants for their involvement and enthusiasm. We have received several promising solutions during this time. However, upon conducting a preliminary analysis of the submissions, we would like to draw your attention to important information regarding some submissions.

The datasets consist of pairs, with x0 representing increasing straight lines and x1 representing decreasing straight lines. This feature was chosen to more fully capture the nature of the acoustic data, which inspired us to create this challenge, as mentioned in the Starter Notebook. Additionally, as you can see in the code, we have included a section in the loss function for control loss reconstruction, not only for y but also for x, to ensure that the CVAE generates x0/x1 similar to the input samples. The desired outputs should resemble the examples presented below.

During our preliminary analysis, our team has noticed that certain submissions included reconstructed samples where x0 (as well as x1) combined both increasing and decreasing lines, as shown below.

It is our preference to have solutions that maintain the inherent characteristics of the inputs during reconstruction, as this aligns with the desired feature of CVAE.

We would like to remind all participants that the use of any additional techniques apart from CVAE, including post-processing, is strictly prohibited. For the Final Evaluation, we expect all solutions to include the complete code required to accurately reproduce your solution.

Thank you for your understanding and cooperation. We eagerly anticipate seeing your innovative solutions in the competition!

Xeek Team


I thought in the second case the Paerson check should filter them out and give 0 score.

In all fairness, your overview says "a decreasing straight line (X0), an increasing straight line (X1), "
BUT your training set has the opposite “an increasing straight line (X0), decreasing straight line (X1)” ?!?

Thank you for clarifying which case you actually want.

You’re absolutely right - there’s a bit of confusion between the description and the examples we’re using for training. Thank you for pointing that out. What we really want to do is follow the pattern shown in the dataset. Consequently, it is crucial for one of the signals, (let’s call it X0 for instance), to consistently represent an increasing straight line, while the other signal, X1 in this case, should consistently represent a decreasing straight line. This alignment with the dataset structure will ensure the fidelity of our approach.

Thanks for noticing this, and I appreciate your understanding.

Best regards, Xeek Team

1 Like