Haoyu Gu

Hi, I'm Haoyu Gu!

Undergraduate Student in Artificial Intelligence at SCUT

My research focuses on generative AI and discrete representations of structured data, with a particular focus on symbolic music. I study how tokenization affects generation quality, controllability, and editing.

+86 18100607076 (WeChat: same number) | ghy20050104@gmail.com

Nanjing, Jiangsu, China | Last Update: September 11, 2026

About Me

Current Status

I'm a senior undergraduate student majoring in Artificial Intelligence at the School of Future Technology, South China University of Technology (SCUT).

Location

Based in Nanjing, Jiangsu Province. School located in Guangzhou, Guangdong Province.

Education

Sep. 2023 – Present

South China University of Technology

未来技术学院

Major: Artificial Intelligence

Sep. 2020 – Jun. 2023

Nanjing Ninghai High School

Sep. 2017 – Jul. 2020

High School Affiliated to Nanjing Normal University, Shuren School

Sep. 2011 – Jul. 2017

Nanjing Langya Road Primary School

Honors & Awards

Jun. 2025

Future Technology Taihu Innovation Award

学业创新一等奖

无锡市政府

May 2025

Lanqiao Cup C++ Programming Contest (Group A)

广东省二等奖

工业和信息化部人才交流中心

Nov. 2024

China Mathematics Competition

广东赛区二等奖

中国数学会

Sep. 2024

SCUT Mathematical Contest

一等奖

华南理工大学数学学院

Sep. 2024

China Undergraduate Mathematical Contest in Modeling

广东省二等奖

中国工业与应用数学学会

Research Interests

Discrete Representations of Structured Data

I study the design of token units and temporal structure, with existing work focused on symbolic music.

Controllable Generation and Editing

I study the relationship between discrete representations and generation methods for music continuation, streaming generation, and explicit editing.

Research Outputs

Dual-Track Piano Music Generation

AI-generated dual-track piano pieces with coherent structure and expressive harmony

音乐续写:Let It Go

给定《冰雪奇缘》Let It Go 前12秒,生成后续内容

原曲(更有节奏与情感):

生成(更加连贯和符合乐理):

长期语义连贯

全曲主题发展连贯,乐句结构清晰,旋律走向符合音乐逻辑

双声部协调

钢琴左右手配合默契,旋律对话清晰,声部走向和谐

创造性表达

和声进行富有新意,旋律转折出人意料,节奏型设计巧妙

调性转换

调式转换自然流畅,使用合理的转调和弦,不同调性之间过渡平滑

Multitrack Ensemble

Multiple instruments in coordination

String Quartet

Two violins, viola, and cello

Piano Quartet

Piano, violin, viola, and cello

Piano & Choir

Piano with SATB choir

Clarinet & Piano

Clarinet with piano accompaniment

Publications

BeatEdit
ACM MM 2026

BeatEdit: Symbolic Music Generation as Explicit Editing

Proposed BeatEdit, the first framework that formulates symbolic music generation as explicit editing. Starting from the prior that composing is revising rather than generating from scratch, it traces the absence of edit-based methods to the representation level, formalizes the required encoding properties, and designs three complementary editing mechanisms along an axis of edit density. It clearly outperforms conventional generative models in matching accuracy across tasks with inference two orders of magnitude faster, and reveals encoding-method interaction as an overlooked design lever.

BEAT
ICML 2026

BEAT: Tokenizing and Generating Symbolic Music by Uniform Temporal Steps

Proposed BEAT, a symbolic music tokenization built on uniform temporal steps. It compresses the temporal states of each pitch within a beat into a single token, matching event-based compactness while explicitly modeling the time grid, with transposition and time-shift invariance. Continuation tasks show it outperforming mainstream symbolic encodings in both subjective and objective evaluations, and it natively supports real-time accompaniment generation that event-based methods struggle to achieve.

Pianoroll-Event
ICASSP 2026

Pianoroll-Event: A Novel Score Representation for Symbolic Music

Proposed Pianoroll-Event, a symbolic music encoding that bridges the grid and discrete sequences. Addressing pianoroll being structure-preserving yet sparse, redundant, and hard to plug into autoregressive generation, it designs four complementary event types that losslessly compress pianoroll into compact discrete sequences, improving encoding efficiency by 1.36x to 7.16x over mainstream methods with generation quality consistently ahead across GPT-2, LLaMA, and LSTM architectures.

ACG
Findings of ACL 2026

Anchored Cyclic Generation: A Novel Paradigm for Long-Sequence Symbolic Music Generation

Addressing error accumulation of autoregressive models in long-sequence symbolic music generation, proposed the Anchored Cyclic Generation (ACG) paradigm and the hierarchical Hi-ACG framework, which calibrate each round of generation with anchor features drawn from already-generated content, significantly outperforming mainstream methods in both subjective and objective evaluations.

EmoMM
Findings of ACL 2026

EmoMM: Benchmarking and Steering MLLM for Multimodal Emotion Recognition under Conflict and Missingness

Built EmoMM, a benchmark for emotion recognition with multimodal large language models under modality conflict and missingness, uncovering the Video Contribution Collapse (VCC) phenomenon where models marginalize video evidence due to video token redundancy and modality preference. Proposed CHASE, which steers attention at inference time through a conflict-aware router, effectively mitigating decision bias.

Talks & Presentations

Sep. 21, 2025

AI Journey: From Study to Research

Event AI Association Freshman Seminar
Location SCUT, GZIC, F3-a101

As a member of the AI Association Academic Resources Department, I was invited to share insights on academic studies, competitions, research, and essential computational tools for computer science majors.

Sep. 7, 2025

Academic Planning & Experience Sharing

Event Jiangsu Students Welcome Meeting
Location SCUT, GZIC, F3-a101

Invited as a student representative to share academic insights and planning strategies with the incoming class of 2025 freshmen.

Sep. 1, 2024

University Life Guide for New Students

Event Jiangsu Students Welcome Meeting
Location SCUT, GZIC, F3-b101

Invited as a student representative to welcome and guide the class of 2024 freshmen, sharing insights on university academic life and providing practical guidance for their college journey.