If you are looking to access or implement this "deep paper," you can find the primary materials here:
is an approximate Bayesian method used to embed object concepts into a high-dimensional vector space based on human behavioral data (specifically "triplet odd-one-out" tasks).
: The LukasMut/VICE GitHub repository contains the implementation for producing these embeddings. ViceHD
: Research by related groups (such as Ranzato et al. ) focuses on producing "deep feature hierarchies" by stacking unsupervised modules, similar to how deep belief networks function.
The request "" likely refers to the research paper "VICE: Variational Interpretable Concept Embeddings" , which introduces a deep learning framework for modeling mental representations of object concepts. 1. The Core Technology: VICE If you are looking to access or implement
: It has been shown to rival or outperform previous models like SPoSE in predicting human behavior and is more consistent across different initializations. 2. Related High-Definition (HD) or Deep Concepts
: Unlike standard deep learning "black boxes," VICE is designed to provide interpretable dimensions, allowing researchers to understand which specific features (e.g., "living thing," "edible," "metal") define a concept. ) focuses on producing "deep feature hierarchies" by
While "ViceHD" isn't a standard academic term, it may be a shorthand for high-fidelity or "deep" applications of this technology: