Hardware Podcasts

Embracing the Efficiency of the Neuromorphic Hairball

EE Times Current · 6 Dec 2024 · 47m

What this episode covers

Dr. Katie Schumann of the University of Tennessee explains her work evolving highly recurrent, sparse spiking neural networks, which she calls 'hairballs,' using the EONs framework to achieve high efficiency with very few neurons. She discusses the trade-offs between evolving network structure and tuning parameters via local learning rules like STDP, and the challenges of training these complex topologies compared to feed-forward models. The conversation with Dr. Giulia D’Angelo and Professor Ralph Etienne-Cummings of Johns Hopkins University explores the partitioning between evolutionary algorithms and learning mechanisms, emphasizing the need for meta-learning and hardware-in-the-loop co-design with memristive technologies.

Most relevant to

early-stage companies

Topics

Mentioned

Companies
University of Tennessee, Oak Ridge National Laboratory, Johns Hopkins University, University College London, Czech Technical University
Chips
FPGA
Materials
memristive technologies
Also
EONs, STDP, SNNPy, Tonic

Related episodes from other shows

Other shows on the same ground, closest first.

  • The Circuit · 51m

    Episode 70!!! NPUs!!!!!!!!

    Ben Bajarin and Jay Goldberg host Austin Lyons and Paul Karazuba to discuss the architecture, market, and future of Neural Processing Units (NPUs). The group debunks the metric 'TOPS per watt' as meaningless, arguing that real-world performance depends on specific network workloads and latency. They explore how chipmakers like Qualcomm, Intel, and Apple are designing NPUs, often by licensing IP or evolving DSPs, and predict that NPUs will become ubiquitous in devices from smartphones to microcontrollers within three to four years.

    Microcontrollers & dev platformsChips, FPGAs & semiconductors
    8 chapters

  • The Next Byte Podcast · 23m

    110. The Future of Neurology: Neural Interfaces Countering Neural Diseases

    Hosts Daniel and Forbode discuss NeuralTree, a neurochip developed by researchers at EPFL (École Polytechnique Fédérale de Lausanne) designed to detect and counter neural disorders like epilepsy and Parkinson's disease. The chip features 256 input channels, an area of 3.5 square millimeters, and an energy-aware machine learning algorithm that allows it to operate efficiently on the edge without cloud dependency. The hosts highlight the device's ability to distinguish between healthy and irregular neural impulses, its potential for self-evolving algorithms to adapt to changing brain signals, and its broader implications for neurological research and treatment.

    Embedded firmwareMicrocontrollers & dev platforms
    7 chapters

The summary, chapters and topics are generated from the episode’s audio by this directory, not written by EE Times Current. The episode itself belongs to its publisher.