The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) Semantic Folding for Natural Language Understanding with Francisco Weber - #451
Jan 29, 2021
Francisco Weber, CEO and co-founder of Cortical.io, shares insights into his company's innovative use of semantic folding for natural language processing. He discusses the evolution from traditional NLP methods to a biologically inspired approach, emphasizing data efficiency. The conversation critiques GPT-3's inefficiencies in business applications and highlights the development of contract intelligence tools for lawyers. Francisco also touches on advancements in sentiment analysis and the challenges of automating machine learning workflows.
AI Snips
Chapters
Transcript
Episode notes
Semantic Folding
- Cortical.io uses Semantic Folding for NLP, inspired by neuroscience and Jeff Hawkins' work.
- It prioritizes efficient representation of words as bitmaps, contrasting with statistical methods.
Sparse Distributed Representations
- Semantic Folding uses sparse, distributed representations, contrasting with dense structures in deep learning.
- It grounds semantics by linking bitmaps to real-world references, improving meaning representation.
Efficiency over Precision
- Semantic Folding acts like a language model, focusing on efficiency over pure precision.
- It excels in real-world scenarios with limited, messy data by training on general language, not specific tasks.