aiqbench takeMachine Learning
aiqbench analysis
The work addresses a practical constraint: photonic and neuromorphic coprocessors can't yet run whole neural networks, and their internal gradients are often inaccessible. The proposed method trains hybrid models without that gradient access, which could make exotic hardware usable alongside conventional electronics. The results so far come from digital simulations of photonic layers across three task classes, so readers should wait for validation on physical devices before drawing conclusions about real-world speed or accuracy.
1
Source coveringGDELT
What this isA model's reading of the article and related coverage, written for aiqbench. Not the publisher's words.
Algorithm Developed for Training AI on Photonic and Neuromorphic Chips
GDELT3 min read
Researchers tested the approach on six digital models of photonic computing layers and circuits, and on three classes of tasks. MOSCOW, October 2. /TASS/. Russian scientists have developed an approach that makes it possible to quickly train hybrid neural networks that use alternative computing systems in their work, including photonic, neuromorphic,…
Excerpt shown under the publisher's RSS terms · 51 words · full text and figures are on the publisher's site
