Ernestas Poškus

Technical blog

"We must view with profound respect the infinite capacity of the human mind to resist the introduction of useful knowledge." - Thomas R. Lounsbury

| github | linkedin | twitter |

ansible 2 / elasticsearch 2 / kernel 1 / learning 29 / linux 2 / mnemonics 1 / nginx 1 / paper 29 / personal 5 / research 28 / review 29 / rust 1 / scientific 29 / tools 2 /

Overcoming catastrophic forgetting in neural networks

WC 244 / RT 2min

Catastrophic forgetting in neural networks

The ability to learn tasks in a sequential fashion is crucial to the development of artificial intelligence. Neural networks are not, in general, capable of this and it has been widely thought that catastrophic forgetting is an inevitable feature of connectionist models.

It is possible to overcome this limitation.

Our approach remembers old tasks by selectively slowing down learning on the weights important for those tasks.

Achieving artificial general intelligence requires that agents are able to learn and remember many different tasks.

A multitask learning paradigm—deep learning techniques have been used to train single agents that can successfully play multiple Atari games.

The lack of algorithms to support continual learning thus remains a key barrier to the development of artificial general intelligence.

Elastic weight consolidation

A deep neural network consists of multiple layers of linear projection followed by element-wise non-linearities. Learning a task consists of adjusting the set of weights and biases θ of the linear projections, to optimize performance.

In order to justify this choice of constraint and to define which weights are most important for a task, it is useful to consider neural network training from a probabilistic perspective.


catastrophic forgetting: knowledge of previously learnt task(s) (e.g. task A) to be abruptly lost as information relevant to the current task (e.g. task B) is incorporated.

elastic weight consolidation: this algorithm slows down learning on certain weights based on how important they are to previously seen tasks.