Neuromorphic Computing: towards event-based (cognitive) sensing and control

Event: Neuromorphic Computing Workshop 2019 · Duration: 24 min · ▶ Watch on YouTube

Abstract

The presentation explores the potential of neuromorphic computing for event-based sensing and control, addressing the inherent inefficiencies of processing asynchronous event streams with traditional Von Neumann architectures. It introduces various neuromorphic hardware platforms designed to better match the event-based nature of data. The speaker discusses methods for programming these devices, including adapting deep neural networks and introducing Dynamic Neural Fields (DNF) as a neurally inspired framework. The talk emphasizes the limitations of current machine learning approaches for complex cognitive tasks and advocates for learning from biological principles, particularly the role of memory, recurrence, and feedback in intelligent processing. Examples of DNF architectures for action selection, sequence learning, and spatial language processing in robotics are presented.

Speakers

  • Yulia Sandamirskaya — Institute of Neuroinformatics, University of Zurich and ETH Zurich

Talks (1)

  • 00:00:05 — Yulia Sandamirskaya: Neuromorphic Computing: towards event-based (cognitive) sensing and control
    • This talk introduces neuromorphic computing as a solution for event-based sensing and control, highlighting the limitations of traditional computing architectures and exploring neurally inspired frameworks for building cognitive systems.

Key Takeaways

  • Event-based vision fundamentally requires event-based processing to overcome the Von Neumann bottleneck and leverage the asynchronous nature of sensor data.
  • Neuromorphic hardware offers a promising alternative to traditional computers for event-based processing, providing better efficiency and performance for neurally inspired algorithms.
  • While deep learning can be adapted for neuromorphic devices, it faces limitations in transparency, post-training learning, and handling complex cognitive tasks like decision-making, planning, and knowledge representation.
  • Learning from biology suggests that intelligence is deeply tied to memory, recurrence, and feedback, which are core components of Dynamic Neural Fields (DNF) architectures.
  • The interface between event-based sensors and processors remains a significant challenge, requiring further development in hardware and computational frameworks to enable truly autonomous and adaptive cognitive systems.

Methods / Models / Datasets Mentioned

  • Loihi
  • DYNAp
  • aiCtx
  • SpiNNaker
  • UniMan
  • DNN
  • CNN
  • SNN
  • SCAMP
  • Dynamic Neural Fields (DNF)
  • Braitenberg vehicle

Topics

Neuromorphic computing · Event-based sensing · Event-based processing · Spiking Neural Networks (SNN) · Dynamic Neural Fields (DNF) · Cognitive architectures · Robotics · Memory · Sensor-processor interface · Machine learning limitations

Notes

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