Combining distributed and localist representations, the fast learning algorithms of Deep Learning with interpretable symbolic approaches.

Neuro-symbolic Approach

<h1>Laplace</h1>
<h2>Overview</h2>
<p><strong>laplace</strong> is an end-to-end recommendation engine framework for large-scale graphs.</p>
<p>The pipeline is designed for self-supervised edge prediction on heterogenous graphs.</p>
<p><img src="/assets/images/pipeline_overview.png" alt="Pipeline Overview"></p>
<h3>Features</h3>
<ul>
<li>Multi-step, hybrid recommendation pipeline:
<ol>
<li>Candidate Selection:
<ul>
<li>Integrating LightGCN recommendations (can be ran on its own aswell)</li>
<li>Multiple, custom heuristics</li>
<li>Strategies can be mixed and matched</li>
</ul>
</li>
<li>Ranking: GraphConvolutional network prediction on candidate edges</li>
</ol>
</li>
<li>Works on <strong>Heterogenous</strong> graphs
<ul>
<li>User based training, validation and test splitting</li>
<li>N-hop neighborhood aggregation</li>
<li>Node Features</li>
<li>Works on any number of node types</li>
</ul>
</li>
<li>Advanced preprocessing of tabular data into graphs
<ul>
<li><strong>Neo4j</strong> integration for better visualization and handling of large graphs.
No newline at end of file</li>
</ul>
</li>
</ul>


Large scale recommendation leveraging Graph Neural Networks

Dream Faster

Studio

Large scale recommendation leveraging Graph Neural Networks

Laplace

Overview

Features

Neuro-symbolic Approach

Combining distributed and localist representations, the fast learning algorithms of Deep Learning with interpretable symbolic approaches.