Conference Presentation (After Call)

FZJ-2024-02523

http://join2-wiki.gsi.de/foswiki/pub/Main/Artwork/join2_logo100x88.png Neuromorphic Hyperdimensional Visual Scene Factorization

Renner, A. (Corresponding author)FZJ* ; Supic, L. ; Danielescu, A. ; Indiveri, G. ; Olshausen, B. A. ; Sandamirskaya, Y. ; Sommer, F. T. ; Paxon Frady, E.

2023

This record in other databases:

Please use a persistent id in citations: doi:10.34734/FZJ-2024-02523

Abstract: In this talk at NNPC 2023, I will present a modular neuromorphic algorithm leveraging recent advancements in hyperdimensional computing/ Vector Symbolic Architectures (VSAs).VSAs have been proposed as a framework for symbolic reasoning, spatial, and graph operations on neuromorphic hardware. They make use of a small set of computational primitives that are robust, efficient, and compatible with diverse hardware. Our algorithm approaches scene understanding as a factorization problem and employs the resonator network to extract object identities and transformations. This is achieved by reflecting the algebraic structure of 2d rigid transforms (translations and rotation) in the neural VSA representation. Finally, we use a spike-timing-based implementation of phasor neurons to show an efficient proof of concept implementation on neuromorphic hardware and employ the model in a robotics task for visual odometry (visual SLAM).

---

Contributing Institute(s):

1. Neuromorphic Software Eco System (PGI-15)

Research Program(s):

1. 5234 - Emerging NC Architectures (POF4-523) (POF4-523)

Appears in the scientific report 2024

---

Database coverage:

---