• Summary
• Introduction
  • Background & Motivation
  • Gap
  • Goal
• Details
  • MCMC algorithm:
  • Model Selection based on model.
  • Hybrid Inference

Summary

This paper proposes a way to integrate the MCMC and Gibbs sampling algorithm using SQL language. Further, it analyzes those models with some characterizations and proposes a rule to choose the different algorithms for different documents in a single query.

Introduction

Background & Motivation

In-database inference offers significant speed-up.

Gap

Existing work can only deal with simple CPF models, not non-linear ones such as skip-chain CRF models.

Goal

This paper tries to explore the in-database implementation of a few inference algorithms, such as MCMC, Gibbs, and MCMC-MH.

Details

Implement various algorithms and identify a set of parameters and rules for choosing inference algorithms.

MCMC algorithm:

Using some UDF can implement the MCMC. But it iteratively calls UDF a million times, leading to a slow process.

The paper tries to re-implement the MCMC such that one query can finish all iterations.

Model Selection based on model.

The paper model the characterizations of a few algorithms and then propose a set of rules to choose from among various inference algorithms.

Hybrid Inference

Then the paper shows the inference algorithm choice is not only model-dependent but also query and text-dependent. Thus it uses a hybrid approach to choose the different algorithms for different documents in a single query.

---

• ← Previous Post
• Next Post →