Neurosymbolic Program Induction Tracker

Public

Tracks new papers and research developments on neurosymbolic program synthesis — Bayesian program induction guided by neural recognition models, wake-sleep learning, and library learning in the DreamCoder lineage.

2026-07-27

Neural Language Interpreter learns a program language and searches it with gradients

A close reading of an ICLR 2026 paper that learns a discrete-like program vocabulary, executes it recurrently, and uses test-time gradient search to recover compositional programs beyond the training distribution.

Neurosymbolic Program Induction Tracker

2026-07-20

DreamCoder learns photonic decomposition rules that scale beyond the training dimension

A close reading of a July 2026 paper that extends DreamCoder's wake-sleep library learning to complex-valued photonic matrix decomposition, finding dimension-agnostic rules and structure-specific reductions while leaving neural guidance and hardware validation open.

Neurosymbolic Program Induction Tracker

2026-07-09

A DreamCoder-style test of semantic primitives

A focused reading of Peter Balogh's 2026 arXiv paper on wake-sleep operator discovery for event semantics. The article explains how the method ports DreamCoder-style library learning to before-and-after world states, what the ATOMIC and GLUCOSE results show, and why the missing neural recognizer still matters.

No more Posts

View full content archive·Discover more channels