Spaces:

---

zadid6pretam

/

iSyncTab

Like 0

Running on Zero

Dear Hugging Face Team,

We would like to apply for a Community GPU Grant for our public iSyncTab Space:

Space: https://huggingface.co/spaces/zadid6pretam/iSyncTab

iSyncTab is an open-source, neural synchrony-guided multimodal learning framework accepted at the ECCV 2026 main conference. It introduces Neural Synchrony-guided Paired Feature Sequencing (NS-PFS) to align and sequence features across modalities, followed by an order-aware memory-augmented Transformer. The primary framework targets image-tabular learning, and we have also evaluated its extension to audio-video learning.

ECCV verification: https://eccv.ecva.net/virtual/2026/poster/5442

The Space provides an interactive HAM10000 image-tabular demonstration and allows users to explore small image-tabular and audio-video learning experiments. The complete implementation is publicly available through GitHub and PyPI:

• GitHub: https://github.com/zadid6pretam/iSyncTab
• Project webpage: https://www.zadidhabib.com/isynctab.html
• PyPI: https://pypi.org/project/isynctab/

The current deployment works on ZeroGPU, but its resource and execution limits require us to impose conservative caps on sample counts, feature dimensions, sequence lengths, training duration, and concurrent usage. These restrictions prevent the interactive training components from demonstrating iSyncTab at a more representative scale.

With a higher-capacity dedicated GPU allocation (ideally 1× L40S or an equivalent allocation), we plan to:

• Increase the supported sample and feature capacity for image-tabular experiments.
• Support larger audio-video datasets and longer multimodal sequences.
• Allow more substantial training runs, batch sizes, and model configurations.
• Improve responsiveness and reliability for concurrent public users.
• Add further reproducible examples and trained multimodal checkpoints for other example datasets.

Although the released inference checkpoint is relatively compact, the interactive training workflows perform modality encoding, GPU-based clustering, cross-modal NS-PFS alignment, and Transformer optimization for each experiment, making dedicated GPU access especially valuable.

We are committed to keeping the Space free, public, open source, and actively maintained for researchers, students, and practitioners working on multimodal learning, NeuroAI, medical imaging, and structured data.

Thank you for considering our application.