← Back to all news
arXiv AI By Junru Zhang, Jinyong Wen, Bin Dou, Wuliang Huang, Baokun Wang, Yu Cheng, Weiqiang Wang, Cheng Peng

Rethinking User Encoders: Routing Diffusion States for Scenario-Adaptive User Representation

Read the original on arXiv AI →

The paper introduces FINDERS, a diffusion‑language‑model framework that replaces static user encoders with a dynamic routing system. FINDERS uses four operations—denoise, factorize, route, and compress—to create scenario‑adaptive user representations by organizing denoising states into transferable cross‑modal semantics and modality‑specific factors. The authors also build BehaviorQA, a large corpus with explicit behavioral rationales, and demonstrate through extensive experiments and online A/B tests that FINDERS outperforms existing methods across ten industrial scenarios.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv AI.

Related stories

arXiv AI
Jun 2

SimSD: Simple Speculative Decoding in Diffusion Language Models

arXiv:2606. 02544v1 Announce Type: cross Abstract: Diffusion large language models (dLLMs) have recently emerged as a promising alternative to autoregressive (AR) LLMs, offering faster inference through parallel or blockwise decoding.

By Junxia Cui, Haotian Ye, Runchu Tian, Hongcan Guo, Jinya Jiang, Haoru Li, Chaojie Ren, Yiming Huang, Kaijie Zhu, Zhongkai Yu, Kun Zhou, Jingbo Shang
llmsdiffusionefficiencybenchmarks
More like this →
arXiv AI
2d ago

TraceQuery: learning what to recompute from full-denoising trajectory in diffusion language models

TraceQuery proposes a new approach for efficient inference in diffusion large language models (dLLMs) by guiding computation using the full-denoising trajectory. Instead of broadly repairing representation errors, it predicts a small set of near-future trajectory-relevant positions, queries only those, and lets the backbone choose the best prediction. This lightweight sparse decoding framework achieves a superior quality‑computation trade‑off across tasks, model variants, and families.

By Yukai Jin, Zhankai Ye, Jiayue Yang, Fangyi Li, Shuoqiu Li, Xu Gao, Shibo Li, Xin Liu, Shangqian Gao
llmsdiffusion
More like this →
arXiv AI
Oct 2

Hierarchical Continuous Diffusion Language Models

Hierarchical Continuous Diffusion Language Models (HC‑DLM) combine discrete token generation with a continuous latent trajectory in a single denoising process, addressing the independence bottleneck of parallel decoding in discrete diffusion models. The training objective is derived from a variational bound on token likelihood, and the latent state is the sole persistent generative element, with tokens read out and fed back at each step. Experiments on Sudoku, Countdown, and LM1B show HC‑DLM outperforming both discrete and continuous diffusion baselines in puzzle accuracy and generative perplexity.

By Hui Ren, Zihan Li, Chang Liu, Huidong Liu, Alexander Schwing
llmsdiffusion
More like this →