Event-based attention and tracking on neuromorphic hardware

Event: CVPR 2025 Workshop · Duration: 2 min · ▶ Watch on YouTube

Abstract

This work explores event-based attention and tracking on neuromorphic hardware by interfacing a neuromorphic sensor, the DAVIS camera, with Intel’s Loihi spiking processor. The researchers implemented a recurrent network, termed a Dynamic Neural Field, on Loihi. This network effectively filters noise from event-based streams and performs blob detection in its initial layer. Subsequently, it demonstrates selective attention by focusing on and tracking a single moving object while ignoring others, showcasing a proof-of-concept for attentional processing in neuromorphic computing.

Speakers

  • Alpha Renner — Institute of Neuroinformatics, University of Zurich and ETH Zurich
  • Mathew Evanusa — University of Maryland, MD, USA
  • Yulia Sandamirskaya — Institute of Neuroinformatics, University of Zurich and ETH Zurich

Talks (1)

  • 00:00:00 — Alpha Renner: Event-based attention and tracking on neuromorphic hardware
    • This talk presents a neuromorphic network implemented on Intel Loihi for event-based attention, selection, and tracking, demonstrating noise filtering and object selection capabilities.

Key Takeaways

  • A neuromorphic sensor (DAVIS camera) can be successfully interfaced with a neuromorphic spiking processor (Intel Loihi) for real-time event-based processing.
  • A recurrent network architecture, specifically a Dynamic Neural Field, can be implemented on neuromorphic hardware to achieve attentional selection and tracking.
  • The system effectively filters noise from event-based data and performs blob detection, followed by selective tracking of a single object.
  • This proof-of-concept demonstrates the potential of neuromorphic hardware for efficient and low-power event-based vision tasks, serving as a fundamental computing primitive for more complex architectures.

Methods / Models / Datasets Mentioned

  • DAVIS camera
  • Intel Loihi
  • Dynamic Neural Field

Topics

Neuromorphic computing · Event-based vision · Attention · Object tracking · Intel Loihi · Dynamic Neural Field · DAVIS camera · Spiking neural networks · Recurrent neural networks

Notes

Open for commentary — connections to other work, critiques, follow-up reading.