Declarative Programming for Designing Neuro-Symbolic Learning Models
summary
DomiKnowS is a Python library that models tasks as a graph of concepts and relationships, connects arbitrary PyTorch or large language models as learners, and injects first‑order logical constraints either as hard constraints at decision time via ILP-in-the-loop or as training-time regularizers using sampling and dual loss options. Implemented algorithms include an ILP inference mask loss, sampling loss, and differentiable constraint surrogates; the authors publish a GLUCON benchmark and metrics (accuracy, consistency, data/time efficiency) and provide a pip-installable package with GitHub demo for entity–relation extraction.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 0:56Motivation
- 3:54Frameworks
- 7:24Why
- 8:14Program Components
- 9:52Algorithms
- 11:10Glucon
- 12:30Demo
- 21:28Questions
related talks
01h 10mNeurosymbolic Program Synthesis: Bridging Perception and Reasoning in Real-World(…)
18mActive Learning for Neurosymbolic Program Synthesis
01h 00mBetter Learning through Programming Languages
36mKeynote: Scientific Discovery: From the Lab Bench to the GPU
23mLise Getoor, Professor, Computer Science, UC Santa Cruz
11mCollaborative Natural Language Inference
16mA Domain-Specific Probabilistic Programming Language for Reasoning about Reasoning(…)
31mKeynote Q&A – Barbara Plank