Learning to (Learn at Test Time): RNNs with Expressive Hidden States
What this work does
Treats a small trainable model as recurrent hidden state, updated by a learned self-supervised loss, while outer training learns the surrounding projections and update behavior.
Connection to discovery
Provides a trainable memory mechanism that could consolidate observations during long discovery histories.
Evidence
The body distinguishes inner fast-state parameters from ordinary network parameters and derives minibatch forms for efficient sequence processing.
Scope and limitations
This is an architectural meaning of TTT, distinct from deployment-wide RL or adapting an entire pretrained LLM. Memory quality does not itself verify a scientific hypothesis.
THE RESEARCH CONVERSATION
Discussion & reactions
Share a question, an insight, or a replication experience. Sign in with GitHub to join the conversation.