About initializing learnable parame φi and ηi

[zjy359]

hello! in Algorithm 1 -Lines3,I want to kown how to update φi and ηi by φi-1 and ηi-1,could you enjoy a pecific methods？
And others, In i-th phase,plastic blocks will learn new parameter by novel data,so this new paramter how to merge with stable blocks in i+1-th phase?

Thank you!!!

[yaoyao-liu]

Hi @zjy359

Thanks for your interest in our work!

For your Q1: At the beginning of each phase, we directly use the old model to initialize the current-phase model. i.e., letting φi=φi-1 and ηi=ηi-1.

For your Q2: we don't merge the stable and plastic branches. We train them separately.

If you have any further questions, please do not hesitate to contact me.

Best,

Yaoyao

[zjy359]

Thank you for your reply!!! So in the i-th phase,the plastic branches learned new knowledge will be keep in plastic branches rather than transfer to stable branches in the i+1-th phase(But plastic branches learned new knowledge in i-th phase will become old knowledge in i+1-th phase,and this konwledge still in plastic branches rather transfer to stable branches,this is my question).In all class-incremental-learning process,the plastic branches learn all new-class knowledge ,and the stable branches only keep base-class knowledge。 Is my understanding correct?

…

------------------ 原始邮件 ------------------ 发件人: "yaoyao-liu/class-incremental-learning" ***@***.***>; 发送时间: 2022年8月18日(星期四) 晚上11:14 ***@***.***>; ***@***.******@***.***>; 主题: Re: [yaoyao-liu/class-incremental-learning] About initializing learnable parame φi and ηi (Issue #45) Hi @zjy359 Thanks for your interest in our work! For your Q1: At the beginning of each phase, we directly use the old model to initialize the current-phase model. i.e., letting φi=φi-1 and ηi=ηi-1. For your Q2: we don't merge the stable and plastic branches. We train them separately. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: ***@***.***>

[yaoyao-liu]

@zjy359

In each phase, we update both plastic and stable branches with new data and old exemplars. The stable branch tends to keep more knowledge because it uses a different architecture.

If you have any further questions, please do not hesitate to contact me.

Best,

Yaoyao

[zjy359]

Thank you very much!!!

…

------------------ 原始邮件 ------------------ 发件人: "Yaoyao ***@***.***>; 发送时间: 2022年8月19日(星期五) 凌晨1:55 收件人: ***@***.***>; 抄送: ***@***.***>; ***@***.***>; 主题: Re: [yaoyao-liu/class-incremental-learning] About initializing learnable parame φi and ηi (Issue #45) @zjy359 在每个阶段，我们用新的数据和旧的范例更新塑料和稳定的分支。稳定分支倾向于保留更多的知识，因为它使用不同的体系结构。 如果您有任何进一步的问题，请不要犹豫与我联系。 最好的， 瑶瑶 — 直接回复此邮件，在GitHub上查看，或取消订阅. ***@***.***与>.