October 10, 2025
Deniz Kucukahmetler Wins Third Place at Algonauts Challenge
Side by side with Semih Eren, Deniz Kucukahmetler achieved third place in the Algonauts Challenge, earning a monetary prize of €3,000. The Challenge is a brain encoding challenge where participants compete using different machine learning models to predict the brain activations of humans as they watch naturalistic movies. The outcomes of the competition were subsequently presented at the Cognitive Computational Neuroscience Conference 2025 (CCN), hosted at the University of Amsterdam.
This year's Algonauts Challenge tasked participants with predicting brain activity (fMRI time series) in 1,000 cortical regions for four subjects watching 80 hours of multimodal movies. The model of Semih and Deniz incorporated multimodal features from Vision (SlowFast, VideoMAE, Swin, CLIP), Audio (HuBERT, WavLM), and Text (CLAP, BERT, Longformer with previous-episode context). Their software architecture consisted of modality-specific bidirectional LSTMs (capturing future and past context), followed by simple average fusion, a cross-modal recurrent layer (LSTM/GRU), and subject-specific linear heads. The main difference in their approach was that, rather than relying on transformer-based architectures, they used a lightweight recurrent neural network (RNN) with simple fusion to combine different modalities and subject-specific heads to produce personalized outputs. Including a brain-inspired learning curriculum also contributed them to achieve performance comparable to larger transformer systems.
They competed against more than 60 teams, and as one of the three winning teams, they were given the opportunity to present their paper at the CCN Conference 2025 in Amsterdam. CCN is an annual forum for discussion among researchers in cognitive science, neuroscience, and artificial intelligence, dedicated to understanding the computations that underlie complex behavior. The goal is to deepen interactions between these disciplines and to discover ways that the communities can benefit one another and leverage each other’s successes.
About the Team
Deniz Kucukahmetler is a PhD student of the SECAI Graduate School in the Neural Data Science and Statistical Computing Group at the Max Planck Institute for Human Cognitive and Brain Sciences (MPI CBS), supervised by Fellow Peter Stadler, Fellow Ivo Sbalzarini, and Nico Scherf. Semih Eren is a M.Sc. student in Computational Modelling and Simulation at TU Dresden and a student assistant at MPI CBS under the supervision of Nico Scherf.
