[ContinualAI Seminars]: "TACOS: Task Agnostic Continual Learning in Spiking Neural Networks"

[Keiland Cooper]

Hi All,

This Thursday 11-11-2021, 5.30pm CEST, for the ContinualAI Seminar, Anurag Daram (University of Texas at San Antonio) will present the paper:

Title: “TACOS: Task Agnostic Continual Learning in Spiking Neural Networks” 

Abstract: Catastrophic interference, the loss of previously learned information when learning new informa- tion, remains a major challenge in machine learn- ing. Since living organisms do not seem to suffer from this problem, researchers have taken inspira- tion from biology to improve memory retention in artificial intelligence systems. However, pre- vious attempts to use bio-inspired mechanisms have typically resulted in systems that rely on task boundary information during training and/or explicit task identification during inference, infor- mation that is not available in real-world scenarios. Here, we show that neuro-inspired mechanisms such as synaptic consolidation and metaplasticity can mitigate catastrophic interference in a spik- ing neural network, using only synapse-local in- formation, with no need for task awareness, and with a fixed memory size that does not need to be increased when training on new tasks. Our model, TACOS, combines neuromodulation with complex synaptic dynamics to enable new learn- ing while protecting previous information. We evaluate TACOS on sequential image recognition tasks and demonstrate its effectiveness in reducing catastrophic interference. Our results show that TACOS outperforms existing regularization tech- niques in domain-incremental learning scenarios. We also report the results of an ablation study to elucidate the contribution of each neuro-inspired mechanism separately.

---------------

- YouTube link: [click here to set up reminders]

- Microsoft Teams: [ click here to join ] 

https://bit.ly/clai-seminars

- YouTube recordings of the previous sessions: https://www.youtube.com/c/ContinualAI

---------------

Feel free to share this email to anyone interested and invite them to subscribe this mailing-list here: https://groups.google.com/g/continualai 

Please also contact me if you want to speak at one of the next sessions!

Looking forward to seeing you all there!

All the best,

Keiland Cooper

University of California

ContinualAI Co-founding Board Member

http://kwcooper.xyz