✳flâneur — a map of the web's best reading

Clare Lyle | The state of plasticity in 2025

clarelyle.com · 4,463 words · saved by 1 readers

It’s been a while since I last posted about the state of plasticity in neural networks, so I thought I would give a very opinionated update on how I see the field as of Summer 2025. For a less opinionated take, I found that this survey paper did a nice job of giving an overview of various recent advances. In the world of neuroscience, people used the term plasticity to refer to the ability of a network to change and then hold its shape. This usage is analogous like how a plastic can be molded and then maintains the shape of its mold (as opposed to e.g. a liquid). Work on neural networks mostly focuses on the former part of this definition, I think in part because the ability to maintain a stable set of connections is trivial in deep learning (you just freeze the weights of your network) compared to biology, which makes the question of maintaining a new “shape” somewhat trivial Additionally, since plasticity is often contrasted with stability, e.g. in stability-plasticity trade-offs, it

Clare Lyle | The state of plasticity in 2025 The state of plasticity in 2025 - Clare Lyle Posted on September 6, 2025 The state of plasticity in 2025 A survey It’s been a while since I last posted about the state of plasticity in neural networks, so I thought I would give a very opinionated update on how I see the field as of Summer 2025. For a less opinionated take, I found that this survey paper did a nice job of giving an overview of various recent advances. Ancient History In the world of neuroscience, people used the term plasticity to refer to the ability of a network to change and then

Explore this link on the map →

saved by

related reading