Aug 8, 2019 · We propose a novel approach to continual learning by approximating a true loss function using an asymmetric quadratic function with one of its sides ...

This paper presents a novel continual learning frame- work based on asymmetric loss approximation with single- side overestimation (ALASSO), which effectively ...

This paper presents a novel continual learning frame- work based on asymmetric loss approximation with single- side overestimation (ALASSO), which effectively ...

We propose a novel approach to continual learning by approximating a true loss function using an asymmetric quadratic function with one of its sides ...

This work proposes a novel approach to continual learning by approximating a true loss function using an asymmetric quadratic function with one of its sides ...

Continual learning by asymmetric loss approximation with single-side overestimation. This repository ( https://github.com/dmpark04/alasso ) contains code to ...

Intuitively, the proposed method consolidates the changes in important parameters to monitor past modes, while allowing the unimportant parameters to learn ...

View recent discussion. Abstract: Catastrophic forgetting is a critical challenge in training deep neural networks. Although continual learning has been ...

공동 저자 ; Continual learning by asymmetric loss approximation with single-side overestimation. D Park, S Hong, B Han, KM Lee. Proceedings of the IEEE/CVF ...

Continual Learning by Asymmetric Loss Approximation with Single-Side Overestimation (ICCV2019) [paper]; Lifelong GAN: Continual Learning for Conditional ...