mirror of
https://github.com/hanzoai/papers.git
synced 2026-11-02 13:32:30 +00:00
zeekay
f9db4f5737
Bylines, paper authorship, correspondence addresses and the founder blurb under each post named an individual. They name the organisation that published the work instead: the brand on posts and pages, the research byline on papers, and research@ in place of a personal address. Both are what these repositories already used elsewhere — nothing new was coined. Co-authors are untouched. Where a paper named someone else alongside, they stay named; only the one byline changed. Bibliography entries, running heads, PDF metadata and affiliations move with it, so a citation still resolves to whoever the cited paper names. Sample and fixture data uses the placeholder name these tests already use elsewhere, so no assertion changed meaning.
64 lines
2.5 KiB
TeX
64 lines
2.5 KiB
TeX
\documentclass[11pt,twocolumn]{article}
|
|
\usepackage{../shared/hanzocover}
|
|
\usepackage[utf8]{inputenc}
|
|
\usepackage{CJK}
|
|
\usepackage{amsmath,amssymb,amsthm}
|
|
\usepackage{graphicx}
|
|
\usepackage{booktabs}
|
|
\usepackage{algorithm}
|
|
\usepackage{algorithmic}
|
|
\usepackage{hyperref}
|
|
\usepackage{xcolor}
|
|
\usepackage{listings}
|
|
\input{../shared/lstlang}
|
|
\usepackage{tikz}
|
|
\usetikzlibrary{shapes,arrows,positioning,fit,backgrounds}
|
|
|
|
\definecolor{codegreen}{rgb}{0,0.6,0}
|
|
\definecolor{codegray}{rgb}{0.5,0.5,0.5}
|
|
\definecolor{codepurple}{rgb}{0.58,0,0.82}
|
|
\definecolor{backcolour}{rgb}{0.95,0.95,0.92}
|
|
|
|
\lstdefinestyle{mystyle}{
|
|
backgroundcolor=\color{backcolour},
|
|
commentstyle=\color{codegreen},
|
|
keywordstyle=\color{codepurple},
|
|
numberstyle=\tiny\color{codegray},
|
|
stringstyle=\color{codegreen},
|
|
basicstyle=\ttfamily\footnotesize,
|
|
breakatwhitespace=false,
|
|
breaklines=true,
|
|
captionpos=b,
|
|
keepspaces=true,
|
|
numbers=left,
|
|
numbersep=5pt,
|
|
showspaces=false,
|
|
showstringspaces=false,
|
|
showtabs=false,
|
|
tabsize=2
|
|
}
|
|
\lstset{style=mystyle}
|
|
|
|
\title{Jin: Unified Multimodal AI Architecture for Vision-Language, Audio, and Cross-Modal Understanding}
|
|
\author{Antje Worring, Hanzo AI Research\thanks{[email protected]} \\ \textit{Hanzo Industries \quad Lux Industries \quad Zoo Labs Foundation} \\ \texttt{[email protected]}}
|
|
\date{April 2023}
|
|
|
|
\begin{document}
|
|
|
|
\hanzocoverpage
|
|
|
|
\begin{abstract}
|
|
We present Jin, a hypermodal AI architecture that unifies text, vision, audio, and extensible custom modalities through a joint embedding predictive framework. Jin combines three key innovations: (1) modality-specific encoders projecting heterogeneous inputs into a shared latent space with learnable modality embeddings; (2) a diffusion transformer backbone with Mixture of Experts (MoE) routing for efficient cross-modal reasoning; (3) hierarchical zooming JEPA (z-JEPA) for multi-scale visual understanding. Our experiments demonstrate that Jin achieves 82.1\% accuracy on ImageNet with zero-shot transfer, outperforms single-modality baselines by 15\% on AudioCaps retrieval, and enables novel cross-modal generation tasks. The architecture supports runtime modality extension without retraining the core model, enabling deployment flexibility for diverse applications.
|
|
\end{abstract}
|
|
|
|
\input{sections/01-introduction}
|
|
\input{sections/02-background}
|
|
\input{sections/03-jin-architecture}
|
|
\input{sections/04-training}
|
|
\input{sections/05-experiments}
|
|
\input{sections/06-applications}
|
|
\input{sections/07-discussion}
|
|
\input{sections/99-conclusion}
|
|
|
|
\end{document}
|