Ye Sun · 孙烨

Beihang University · Computer Science and Technology

Bachelor of Engineering, School of Computer Science and Engineering. GPA: 3.67/4.00; major-course scores include Mathematical Analysis (100), Software Engineering (98), and Deep Learning (94).

Beihang University · Computer Science and Technology

Integrated master’s–PhD program, School of Computer Science and Engineering. Master’s stage: Sep 2021–Jun 2023; PhD stage: Sep 2023–Jun 2027 (expected). GPA: 3.87/4.00; rank: 13/143. Major-course scores include Data Mining (100), Algorithm Design and Analysis (94), Software Engineering (95), and Pattern Recognition (93). Research focuses on knowledge graph reasoning, data mining, and visual analytics; related results have appeared in multiple CCF-A papers.

GeneticPrism · Hierarchical graph layout and scholarly evolution visualization

• Addresses the joint representation of within-topic citations and cross-topic influence in multi-topic scholarly evolution by proposing a hierarchical graph representation and IFHL (Integrated Flow Hierarchical Layout) layout.
• Leads the design of the GeneticPrism overview and GeneticScroll detail views for topic overlap, temporal evolution, and cross-topic influence, enabling linked multi-scale exploration and deeper analytical insight.

eXpath · Explainable knowledge graph reasoning framework

• Addresses the black-box nature of embedding-model decisions by independently proposing an explanation framework that combines ontological closed-path rules and relational-path evidence.
• Designs a structured explanation-generation mechanism that coordinates rule mining and path search, integrating ontological closed-path rules with relational-path evidence into a quantitatively evaluable framework.
• On standard benchmarks, improves two explanation-quality metrics by approximately 20% and reduces explanation-generation time by 61.4% against the best comparison method, demonstrating gains in both explanation quality and computational efficiency.

RuleDep · Dependency-aware rule aggregation

• Leads the design of sparse second-order corrections and a two-stage training framework to model complementarity and redundancy in rule aggregation through dependencies among jointly triggered rules and signed gains in log-failure evidence space.
• Across seven datasets, raises average MRR from 0.382 to 0.396, achieves 21 best or tied-best metrics among interpretable methods, and improves MRR by approximately 10.7% on dependency-rich subsets.

INSPIRE · Expressive Piano Performance Generation

• Builds a note-level structured performance-generation framework using PyTorch and supervised fine-tuning of a Transformer (T5), generating expressive performances from discrete symbols through relative-timing modeling and continuous distribution prediction.
• Designs typed encoding that fuses pitch, duration, dynamics, and other note attributes into a single Transformer timestep, with score-relative timing deviations for fine-grained expressive control.
• On ASAP, reduces PP HR by 15.2% against PianistTransformer; architecture and inference-flow optimization reduces the reported test-task inference cost by 95.3%, improving deployability and iteration efficiency.

CueIR · Long-term agent memory

• Designs a traceable long-term memory system for agents, organizing cross-session information with Episodes, Cues, and a relation graph for hybrid retrieval, graph expansion, and evidence reconstruction.
• Implements memory-graph construction, agent tool calls, and on-demand evidence reconstruction with evidence-sufficiency checks, verification, and provenance tracking.
• Evaluates accurate retrieval, long-range understanding, and selective forgetting on LoCoMo and MemoryAgentBench.

MaterAgent · Agent Harness for materials R&D

• Understands the PI-Agent framework through agent interaction, including the agent loop, runtime, plugin model, and lifecycle, and participates in Agent Harness architecture design for end-to-end materials R&D.
• Designs MCP tools for materials R&D, writes tool specifications, and uses agent interaction to implement, integrate, and validate them.
• Designed and developed a Paper Browser MCP tool for automated retrieval and question answering over several thousand materials-science papers.
• Participates in the advisor's industry-academic collaboration with MaterBrain, moving agent execution infrastructure and supporting tools into materials R&D workflows.

GeneticFlow V2.0 · Large-scale scholarly literature visualization system

• Leads development of a multidimensional visualization system for million-scale scholarly literature, integrating graph-layout computation with visual interaction; the system has served 102,129 users, supported 507,827 visits, and received software copyright registration.
• Compiles Graphviz C++ sources into browser-side layout components, connecting graph-layout computation with front-end interaction through independent full-stack development and operations.

Research-agent-based data classification

• Developed LLM entity-extraction tools and a RAG-based data classification pipeline for sensitive-data annotation and refinement.

National Key R&D Program: Rare-earth catalysis knowledge graph and full-process digital R&D platform

• Led construction of the database and materials knowledge graph, participating throughout the project from proposal to acceptance and ensuring scheduled delivery reviews.
• Designed and built a multimodal materials database at the 10^5 scale by integrating heterogeneous data to support synthesis-condition recommendation and knowledge discovery.
• Implemented synthesis-condition recommendation and knowledge-discovery functions based on the knowledge graph to support intelligent materials R&D workflows.

SenseTime · AI full-stack engineering intern

• Contributed deeply to the Miaohua AI-drawing system, designing and implementing back-end APIs for efficient and stable core drawing calls.
• Led development of a Discord-based intelligent bot that connected efficiently to the AI-drawing core APIs and served over 1,000 users.

SenseTime · Model toolchain intern

• Contributed to training-cluster scheduling toolchains and led development of the spring.remote command-line tool for remote cluster login and management, as well as system monitoring tools.
• Received an Outstanding Intern award for performance on the project.

Technical University of Munich · Research visit

Worked with Prof. Stephen Kobourov on knowledge graph visualization and graph drawing; subsequently collaborated with Maribel Acosta on rule-based reasoning.