OT: Breakthrough: "Liquid" neural networks

[Edward K. Ream]

From this article in Quanta magazine.

Attached is the cited paper. The maths are beyond me, but the summary is clear enough:

"...we obtain models that are between one and five orders of magnitude faster in training and inference compared with differential equation-based counterparts. More importantly, ...closed-form networks can scale remarkably well compared with other deep learning instances. Lastly, as these models are derived from liquid networks, they show good performance in time-series modelling compared with advanced recurrent neural network models."

Edward

[Thomas Passin]

Interesting!  And also hot off the presses, the complete neural structure of the brain of the fruit fly larva has been mapped.  It has  93 neuron types, 3016 neurons, with, get this, *548,000* interconnections.  The "liquid neural network" described in the Quanta article has 19 neurons and 253 connections.

[Edward K. Ream]

On Sat, Mar 11, 2023 at 8:19 AM Thomas Passin <tbp1...@gmail.com> wrote:

Interesting!  And also hot off the presses, the complete neural structure of the brain of the fruit fly larva has been mapped.  It has  93 neuron types, 3016 neurons, with, get this, *548,000* interconnections.  The "liquid neural network" described in the Quanta article has 19 neurons and 253 connections.

Yes. Early days. No telling where this will all lead...

Edward