Gast ::
| Hauptseite > Publikationsdatenbank > Training-to-Learn with Memristive Devices |
| Hauptseite > Publikationsdatenbank > Training-to-Learn with Memristive Devices |
| Conference Presentation (Invited) | FZJ-2024-02778 |
; ;
2023
This record in other databases:
Please use a persistent id in citations: doi:10.29363/nanoge.neumatdecas.2023.013
Abstract: Memristive crossbar arrays are promising non-von Neumann computing technologies to enable real-world, onlinelearning in neural networks. However, their deployment to real-world learning problems is hindered by their non-linearitiesin conductance updates, variation during operation, fabrication mismatch and the realities of gradient descent training. In thiswork, we show that, with a phenomenological model of the device and bi-level optimization, it is possible to pre-train the neuralnetwork to be largely insensitive to such non-idealities on learning tasks. We demonstrate this effect using Model Agnostic Meta Learning (MAML) and a differentiable model of the conductance update on the Omniglot few-shot learning task. Since pre-training is a necessary procedure for any on-line learning scenario at the edge, our results may pave the way towards real-world applications of memristive devices without significant adaption overhead.
|
The record appears in these collections: |
