Flow Reasoning Models aim to solve reasoning problems by looping predictions
A researcher describes work that turns a flow model into a recurrent model for tasks such as Sudoku. Predictions are fed back as inputs and refined on successive passes.
TLDR
Flow Reasoning Models repeatedly reuse their predictions as inputs to refine them, according to a researcher describing the team’s work on reasoning problems such as Sudoku. The researcher says each pass gets its own local training loss, with no backpropagation through time.
Combined views
47K
4 Sources, first seen ago