Vicehd

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).

: VICE: Variational Interpretable Concept Embeddings (arXiv) . ViceHD

: 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 is an approximate Bayesian method used to embed

: It uses variational inference (a technique often used in deep generative models) to obtain sparse, non-negative representations of concepts. ViceHD

Latent Diffusion Model for Video Counterfactual Explanations