Removing frequencies in red causes misclassification to “classical” despite the signal retaining blues perceptual qualities, revealing the model’s
reliance on narrow feature sets. Image from DMRN+20 paper.
Day/time: 23/04/2026 11am-12pm
Location: G2/Teams
Summary
In this meeting, Nico rehearsed his presentation scheduled for June at Kings related to the work:
García-Peguinho, N., Kelly, D., Navaratnarajah, M., Chockler, H., Xambó, A. (2025) “Validating Few-shot Bird Vocalisation Detection through Causal Explanations”, DMRN+20, 2025, King’s College London, London, UK.
We continued the meeting with updates on the ISMIR drafts and a final quick round of updates.
Agenda
- 11:00-11:05 Welcome
- 11:05-11:25 Nico’s talk rehearsal “Challenges for Perturbation style XAI for Audio Classifiers: Implications for Soundscape Research” - 10min presentation + 10min feedback (talk rehearsal for forthcoming talk at King’s College London University)
- 11:25-11:40 Shuoyang’s NIME video presentation rehearsal of his accepted NIME paper “Adapting Neural Audio Autoencoders as Design Materials in NIME” - (rehearsal + feedback)
- 11:40-11:48 ISMIR drafts updates
- 11:48 - 11:56 Quick round of news & updates
- 11:56 - 12:00 Closing
Follow-up
After the meeting, we went together for a quick bring-your-own-lunch at the Grad Cafe.