Source author record

Huichao Zhang

Huichao Zhang appears in the imported research catalog. Authorship, coauthor and topic links are available while profile ownership is still unclaimed.

ResearcherUnclaimed source record

cond-mat.mtrl-sci Computer Vision physics.optics Artificial Intelligence cond-mat.dis-nn Machine Learning

Catalog footprint

What is connected

5works

6topics

4close collaborators

Actions

Connect this record

Open graph Browse works

Inspect adjacent papers, topics, institutions and collaborators without losing the researcher page.

Building this map preview

BZPEER is loading the nearby papers, people, topics and institutions for this page.

preprint2026arXiv

NextFlow: Unified Sequential Modeling Activates Multimodal Understanding and Generation

We present NextFlow, a unified decoder-only autoregressive transformer trained on 6 trillion interleaved text-image discrete tokens. By leveraging a unified vision representation within a unified autoregressive architecture, NextFlow natively activates multimodal understanding and generation capabilities, unlocking abilities of image editing, interleaved content and video generation. Motivated by the distinct nature of modalities - where text is strictly sequential and images are inherently hierarchical - we retain next-token prediction for text but adopt next-scale prediction for visual generation. This departs from traditional raster-scan methods, enabling the generation of 1024x1024 images in just 5 seconds - orders of magnitude faster than comparable AR models. We address the instabilities of multi-scale generation through a robust training recipe. Furthermore, we introduce a prefix-tuning strategy for reinforcement learning. Experiments demonstrate that NextFlow achieves state-of-the-art performance among unified models and rivals specialized diffusion baselines in visual quality.

preprint2026arXiv

VAR RL Done Right: Tackling Asynchronous Policy Conflicts in Visual Autoregressive Generation

Visual generation is dominated by three paradigms: AutoRegressive (AR), diffusion, and Visual AutoRegressive (VAR) models. Unlike AR and diffusion, VARs operate on heterogeneous input structures across their generation steps, which creates severe asynchronous policy conflicts. This issue becomes particularly acute in reinforcement learning (RL) scenarios, leading to unstable training and suboptimal alignment. To resolve this, we propose a novel framework to enhance Group Relative Policy Optimization (GRPO) by explicitly managing these conflicts. Our method integrates three synergistic components: 1) a stabilizing intermediate reward to guide early-stage generation; 2) a dynamic time-step reweighting scheme for precise credit assignment; and 3) a novel mask propagation algorithm, derived from principles of Reward Feedback Learning (ReFL), designed to isolate optimization effects both spatially and temporally. Our approach demonstrates significant improvements in sample quality and objective alignment over the vanilla GRPO baseline, enabling robust and effective optimization for VAR models.

preprint2019arXiv

Molecular and polymeric amorphous forms in dense SO$_2$

We report here a study of reversible pressure-induced structural transformation between two amorphous forms of SO$_2$: molecular at pressures below 26 GPa and polymeric above this pressure, at temperatures of 77 - 300 K. The transformation was observed by Raman spectroscopy and x-ray diffraction in a diamond anvil cell. The same phenomenon was also observed in ab initio molecular dynamics simulations where both forward and reverse transitions were detected, allowing to analyze in detail the atomic structure of both phases. The high-pressure polymeric amorphous form was found to consist mainly of disordered polymeric chains made of 3-coordinated sulfur atoms connected via oxygen atoms, and few residual intact molecules. The simulation results are in good agreement with experimental data. Our observations suggest a possible existence of molecular liquid - polymeric liquid transition in SO$_2$ and show a case example of polyamorphism in system consisting of simple molecules with multiple bonds.

preprint2016arXiv

Slow Auger Recombination of Charged Excitons in Nonblinking Perovskite Nanocrystals without Spectral Diffusion

Over the last two decades, intensive research efforts have been devoted to the suppressions of photoluminescence (PL) blinking and Auger recombination in metal-chalcogenide nanocrystals (NCs), with significant progresses being made only very recently in several specific heterostructures. Here we show that nonblinking PL is readily available in the newly-synthesized perovskite CsPbI3 (cesium lead iodide) NCs, and their Auger recombination of charged excitons is greatly slowed down, as signified by a PL lifetime about twice shorter than that of neutral excitons. Moreover, spectral diffusion is completely absent in single CsPbI3 NCs at the cryogenic temperature, leading to a resolution-limited PL linewidth of ~200 μeV.

preprint2015arXiv

Single Photon Emission from Single Perovskite Nanocrystals of Cesium Lead Bromide

The power conversion efficiency of photovoltaic devices based on semiconductor perovskites has reached ~20% after just several years of research efforts. With concomitant discoveries of other promising applications in lasers, light-emitting diodes and photodetectors, it is natural to anticipate what further excitements these exotic perovskites could bring about. Here we report on the observation of single photon emission from single CsPbBr3 perovskite nanocrystals (NCs) synthesized from a facile colloidal approach. Compared with traditional metal-chalcogenide NCs, these CsPbBr3 NCs exhibit nearly two orders of magnitude increase in their absorption cross sections at similar emission colors. Moreover, the radiative lifetime of CsPbBr3 NCs is greatly shortened at both room and cryogenic temperatures to favor an extremely fast output of single photons. The above findings have not only added a novel member to the perovskite family for the integration into current optoelectronic architectures, but also paved the way towards quantum-light applications of single perovskite NCs in various quantum information processing schemes.