公开(公告)号：US20210232930A1

公开(公告)日：2021-07-29

申请号：US16972427

申请日：2019-10-11

Applicant: Google LLC

Inventor： Jyrki Alakuijala ,  Iulia-Maria Comsa ,  Krzysztof Potempa

IPC: G06N3/08 ,  G06N3/04

Abstract: Spiking neural networks that perform temporal encoding for phase-coherent neural computing are provided. In particular, according to an aspect of the present disclosure, a spiking neural network can include one or more spiking neurons that have an activation layer that uses a double exponential function to model a leaky input that an incoming neuron spike provides to a membrane potential of the spiking neuron. The use of the double exponential function in the neuron's temporal transfer function creates a better defined maximum in time. This allows very clearly defined state transitions between “now” and the “future step” to happen without loss of phase coherence.

公开(公告)号：US20210256388A1

公开(公告)日：2021-08-19

申请号：US17169740

申请日：2021-02-08

Applicant: Google LLC

Inventor： Thomas Fischbacher ,  Luca Versari ,  Krzysztof Potempa ,  Iulia-Maria Comsa ,  Moritz Firsching ,  Jyrki Antero Alakuijala

IPC: G06N3/08 ,  G06F17/16 ,  G06N3/04

Abstract: The present disclosure proposes a model that has more expressive power, e.g., can generalize from a smaller amount of parameters and assign more computation in areas of the function that need more computation. In particular, the present disclosure is directed to novel machine learning architectures that use the exponential of an input-dependent matrix as a nonlinearity. The mathematical simplicity of this architecture allows a detailed analysis of its behavior.

公开(公告)号：US20210248476A1

公开(公告)日：2021-08-12

申请号：US17170025

申请日：2021-02-08

Applicant: Google LLC

Inventor： Thomas Fischbacher ,  Iulia-Maria Comsa ,  Luca Versari

IPC: G06N3/08 ,  G06N3/04 ,  G06N20/00 ,  G06K9/62 ,  G06F17/16

Abstract: The present disclosure proposes a model that has more expressive power, e.g., can generalize from a smaller amount of parameters and assign more computation in areas of the function that need more computation. In particular, the present disclosure is directed to novel machine learning architectures that use the exponential of an input-dependent matrix as a nonlinearity. The mathematical simplicity of this architecture allows a detailed analysis of its behavior, providing stringent robustness guarantees via Lipschitz bounds.