PHYSICS, LEARNING & ENGINEERING

Research & engineering

Quantum systems.
Useful software.

I work across quantum memories, STEM reinforcement learning, and the tools that make research easier to do. Explore my experiments, published work, and software.

Education & experience 经历

01
ByteDance SeedSTEM × REINFORCEMENT LEARNING

2026.11–present · STEM reinforcement learning · Intern

Reinforcement learning
for STEM.

Reinforcement learning for STEM reasoning, spanning data development, algorithm research, and benchmark design.

Scientific reasoning, evaluation, and another attemptAn open geometric construction gives rise to three related candidate contours. One is compared with a reference; the hatched discrepancy sends feedback around a return ribbon. A subsequent contour changes shape, while a small discrepancy remains. A conceptual learning cycle.
  1. Frame a problem
  2. Explore solutions
  3. Evaluate
  4. Revise

A geometric problem opens into candidate solutions. Evaluation returns feedback that shapes the next attempt. Conceptual illustration.

01 / DATA

Make expertise learnable.

STEM tasks, data, and feedback.

02 / ALGORITHMS

Improve how models learn.

Reinforcement learning methods.

03 / BENCHMARKS

Measure what matters.

Scientific reasoning and evaluation.

STEM reinforcement learning · concept illustration

EXPERIMENTS & QUESTIONS

Research 研究

02
Complete author-provided experimental figure showing the spin-wave memory array, AOD-addressed write and read optical paths, detection, and timing control.
Multimode memory apparatus and write/read measurement configurations; complete panels a–c, supplied for the manuscript with Yu Hu.Author-provided manuscript figure

Cold atoms & optical control

Complete author-provided experimental figure showing the spin-wave memory array, AOD-addressed write and read optical paths, detection, and timing control.

Multimode memory apparatus and write/read measurement configurations; complete panels a–c, supplied for the manuscript with Yu Hu.

Author-provided manuscript figureOpen original image ↗
Cold atoms & optical control

Spatially addressed cold-atom memories for remote entanglement distribution.

Rare-earth crystal structure, Hamiltonian terms, and energy-level diagram.
Crystal-field and spin-Hamiltonian modelling view.Research portfolio / CV

Rare-earth quantum systems

Rare-earth crystal structure, Hamiltonian terms, and energy-level diagram.

Crystal-field and spin-Hamiltonian modelling view.

Research portfolio / CVOpen original image ↗
Rare-earth quantum systems

A modelling and parameterisation framework for rare-earth-doped Y₂SiO₅ quantum memories.

Open a topic for methods, figures & context.
01Multiplexed quantum memories & networksSpatially addressed cold-atom memories for remote entanglement distribution.

Spatially addressed cold-atom memories for remote entanglement distribution.

I work on AOD-addressed atomic ensembles, optical control, and multiplexed entanglement protocols. The experimental workflow connects waveform generation, FPGA timing, photon acquisition, and Python/C++ analysis.

Complete author-provided experimental figure showing the spin-wave memory array, AOD-addressed write and read optical paths, detection, and timing control.
Multimode memory apparatus and write/read measurement configurations; complete panels a–c, supplied for the manuscript with Yu Hu.Author-provided manuscript figure

Multiplexed quantum memories & networks

Complete author-provided experimental figure showing the spin-wave memory array, AOD-addressed write and read optical paths, detection, and timing control.

Multimode memory apparatus and write/read measurement configurations; complete panels a–c, supplied for the manuscript with Yu Hu.

Author-provided manuscript figureOpen original image ↗
Two-node quantum memory relay with FPGA feedforward and count traces.
Two-node relay architecture with FPGA feedforward.Research portfolio / CV

Multiplexed quantum memories & networks

Two-node quantum memory relay with FPGA feedforward and count traces.

Two-node relay architecture with FPGA feedforward.

Research portfolio / CVOpen original image ↗
  • Quantum repeater
  • Quantum memory
  • Entanglement distribution
  • Multiplexing
02AI for Science · CNN–BiLSTM entanglement classificationA hybrid deep-learning architecture for sample-efficient multipartite entanglement classification.

A hybrid deep-learning architecture for sample-efficient multipartite entanglement classification.

Quantum density matrices are converted into multi-channel image representations and classified with a CNN–BiLSTM network for 3–4 qubit systems. The work covered architecture co-design, the data pipeline, and benchmarking against MLP and CNN baselines.

Complete preprint Figure 1 contrasting feature flattening and sequence reshaping in two CNN–BiLSTM architectures.
Architecture: two routes from convolutional features to BiLSTM classification. Q. Sun, Y. Sun et al., preprint Fig. 1 (2026), CC BY 4.0.Preprint figure · arXiv ↗

AI for Science · CNN–BiLSTM entanglement classification

Complete preprint Figure 1 contrasting feature flattening and sequence reshaping in two CNN–BiLSTM architectures.

Architecture: two routes from convolutional features to BiLSTM classification. Q. Sun, Y. Sun et al., preprint Fig. 1 (2026), CC BY 4.0.

Complete preprint Figure 5 comparing classification accuracy of CNN, BiLSTM, MLP and CNN–BiLSTM models across training-set sizes for three- and four-qubit states.
Benchmarks: model accuracy across training-set sizes. Q. Sun, Y. Sun et al., preprint Fig. 5 (2026), CC BY 4.0.Preprint figure · arXiv ↗

AI for Science · CNN–BiLSTM entanglement classification

Complete preprint Figure 5 comparing classification accuracy of CNN, BiLSTM, MLP and CNN–BiLSTM models across training-set sizes for three- and four-qubit states.

Benchmarks: model accuracy across training-set sizes. Q. Sun, Y. Sun et al., preprint Fig. 5 (2026), CC BY 4.0.

  • Machine learning
  • Entanglement classification
  • CNN
  • BiLSTM
03SLM-based OAM multimode multiplexingExploring spatial and orbital-angular-momentum modes for optical multiplexing.

Exploring spatial and orbital-angular-momentum modes for optical multiplexing.

AOD beam routing and SLM-generated OAM modes provide complementary ways to address optical channels. My work connects mode characterisation, optical alignment, and Python analysis of coupling and crosstalk.

AOD and OAM optical multiplexing paths producing multiple spatial modes.
AOD and OAM paths for multimode multiplexing.Research portfolio / CV

SLM-based OAM multimode multiplexing

AOD and OAM optical multiplexing paths producing multiple spatial modes.

AOD and OAM paths for multimode multiplexing.

Research portfolio / CVOpen original image ↗
Optical mode preparation with an SLM, mode analysis, and coupling and crosstalk characterization.
From mode preparation to coupling and crosstalk characterization.Schematic · research methods

SLM-based OAM multimode multiplexing

Optical mode preparation with an SLM, mode analysis, and coupling and crosstalk characterization.

From mode preparation to coupling and crosstalk characterization.

Schematic · research methodsOpen original image ↗
  • OAM
  • SLM
  • AOD
  • Optical multiplexing
  • Laguerre–Gaussian modes
04Learning-assisted quantum control & inferencePhysics-informed models for quantum-state inference, noise analysis, and experimental feedback.

Physics-informed models for quantum-state inference, noise analysis, and experimental feedback.

I study how measurement information, physical models, and learning objectives interact in quantum-state estimation and control. Current work connects photon-count data, structured neural models, and reproducible validation workflows.

Signal and idler AODs addressing a selected region in a quantum memory array.
Signal/idler AOD control over a selected memory-array region.Research portfolio / CV

Learning-assisted quantum control & inference

Signal and idler AODs addressing a selected region in a quantum memory array.

Signal/idler AOD control over a selected memory-array region.

Research portfolio / CVOpen original image ↗
Complete author-provided experimental figure showing the spin-wave memory array, AOD-addressed write and read optical paths, detection, and timing control.
Multimode memory apparatus and write/read measurement configurations; complete panels a–c, supplied for the manuscript with Yu Hu.Author-provided manuscript figure

Learning-assisted quantum control & inference

Complete author-provided experimental figure showing the spin-wave memory array, AOD-addressed write and read optical paths, detection, and timing control.

Multimode memory apparatus and write/read measurement configurations; complete panels a–c, supplied for the manuscript with Yu Hu.

Author-provided manuscript figureOpen original image ↗
  • Quantum control
  • Noise analysis
  • Feedback
  • ESM/CSM
  • Physics-informed ML
05Rare-earth solid-state quantum memory and repeater modellingA modelling and parameterisation framework for rare-earth-doped Y₂SiO₅ quantum memories.

A modelling and parameterisation framework for rare-earth-doped Y₂SiO₅ quantum memories.

Built full Hamiltonian models spanning free-ion, crystal-field, Zeeman, and hyperfine terms down to low-energy spin Hamiltonians. The work included ZEFOZ analysis and comparison of AFC, spin-wave storage, NLPE-DLCZ, and single-ion repeater schemes.

Rare-earth crystal structure, Hamiltonian terms, and energy-level diagram.
Crystal-field and spin-Hamiltonian modelling view.Research portfolio / CV

Rare-earth solid-state quantum memory and repeater modelling

Rare-earth crystal structure, Hamiltonian terms, and energy-level diagram.

Crystal-field and spin-Hamiltonian modelling view.

Research portfolio / CVOpen original image ↗
Schematic of crystal-field Hamiltonian modelling, effective spin models, and their connection to transitions, ZEFOZ analysis and memory protocols.
From the microscopic Hamiltonian to an effective spin model and memory-protocol analysis.Schematic · research methods

Rare-earth solid-state quantum memory and repeater modelling

Schematic of crystal-field Hamiltonian modelling, effective spin models, and their connection to transitions, ZEFOZ analysis and memory protocols.

From the microscopic Hamiltonian to an effective spin model and memory-protocol analysis.

Schematic · research methodsOpen original image ↗
  • Rare-earth ions
  • YSO
  • Crystal field
  • ZEFOZ
  • AFC
  • Quantum repeater
06Er³⁺:YSO single-ion node · interface and nanofabricationAn optical–microwave interface for communication-band Er³⁺:YSO single-ion repeater nodes.

An optical–microwave interface for communication-band Er³⁺:YSO single-ion repeater nodes.

Optical and microwave interface design for Er³⁺:YSO single-ion repeater nodes. The work connects electromagnetic simulation, grating coupler design, coplanar waveguides, and nanofabrication workflows.

SEM images of fabricated photonic structures for an Er3+:YSO interface.
SEM views from the nanofabrication workflow.Research portfolio / CV

Er³⁺:YSO single-ion node · interface and nanofabrication

SEM images of fabricated photonic structures for an Er3+:YSO interface.

SEM views from the nanofabrication workflow.

Research portfolio / CVOpen original image ↗
Schematic of optical and microwave interfaces to an Er3+:YSO ion, with a design, simulation, fabrication and inspection workflow.
Optical and microwave interfaces, alongside the device-development workflow.Schematic · research methods

Er³⁺:YSO single-ion node · interface and nanofabrication

Schematic of optical and microwave interfaces to an Er3+:YSO ion, with a design, simulation, fabrication and inspection workflow.

Optical and microwave interfaces, alongside the device-development workflow.

Schematic · research methodsOpen original image ↗
  • Nanophotonics
  • Grating coupler
  • FDTD
  • COMSOL
  • CPW
  • EBL
  • Er³⁺:YSO

PAPERS & PREPRINTS

Publications 论文

Google Scholar ↗
Nature Communications 17, 697Published · 2026

Metropolitan-scale ion-photon entanglement via a quantum network node with hybrid multiplexing enhancements

Z.-B. Cui, Z.-Q. Wang, P.-C. Lai, et al., including Y.-D. Sun

Co-authorDOI ↗
Complete Figure 1 showing a conceptual multiplexed trapped-ion network, the implemented ion–photon node highlighted within it, the time-bin protocol, and multiplexing enhancement scaling.
Fig. 1 · Implemented ion–photon node within a network concept; timing and scaling. Cui et al. (2026), CC BY-NC-ND 4.0.Published figure · Nature Communications (2026) ↗

Metropolitan-scale ion-photon entanglement via a quantum network node with hybrid multiplexing enhancements

Complete Figure 1 showing a conceptual multiplexed trapped-ion network, the implemented ion–photon node highlighted within it, the time-bin protocol, and multiplexing enhancement scaling.

Fig. 1 · Implemented ion–photon node within a network concept; timing and scaling. Cui et al. (2026), CC BY-NC-ND 4.0.

Fig. 1 · Implemented ion–photon node within a network concept; timing and scaling. Cui et al. (2026), CC BY-NC-ND 4.0. Published figure · Nature Communications (2026) ↗

Nature Communications 15, 10306Published · 2024

Fast delivery of heralded atom-photon quantum correlation over 12 km fiber through multiplexing enhancement

Sheng Zhang, Jixuan Shi, Yibo Liang, Yuedong Sun, et al.

Co-authorDOI ↗
Complete Figure 1 showing the quantum-repeater concept and the implemented multiplexed atom–photon heralding protocol over 12 km of fiber.
Fig. 1 · Repeater concept and the implemented atom–photon protocol. Zhang et al. (2024), CC BY 4.0.Published figure · Nature Communications (2024) ↗

Fast delivery of heralded atom-photon quantum correlation over 12 km fiber through multiplexing enhancement

Complete Figure 1 showing the quantum-repeater concept and the implemented multiplexed atom–photon heralding protocol over 12 km of fiber.

Fig. 1 · Repeater concept and the implemented atom–photon protocol. Zhang et al. (2024), CC BY 4.0.

Fig. 1 · Repeater concept and the implemented atom–photon protocol. Zhang et al. (2024), CC BY 4.0. Published figure · Nature Communications (2024) ↗

ONGOING RESEARCH

Manuscripts & collaborations

Research manuscriptIn preparation

Measurement-Efficient Characterization and Control of Multipartite Quantum States via Physics-Informed Learning

Y.-D. Sun*, Y. Hu*, et al.

Co-first author
Author-provided manuscript figure showing a spin-wave memory array, AOD-addressed write and read paths, photon detection, electronic timing control, and the write and read measurement configurations.
Experimental platform and measurement interface for multimode quantum-state characterization and control. Figure supplied by Yuedong Sun for the manuscript with Yu Hu; complete panels a–c.Author-provided manuscript figure

Measurement-Efficient Characterization and Control of Multipartite Quantum States via Physics-Informed Learning

Author-provided manuscript figure showing a spin-wave memory array, AOD-addressed write and read paths, photon detection, electronic timing control, and the write and read measurement configurations.

Experimental platform and measurement interface for multimode quantum-state characterization and control. Figure supplied by Yuedong Sun for the manuscript with Yu Hu; complete panels a–c.

Author-provided manuscript figureOpen original image ↗

Experimental platform and measurement interface for multimode quantum-state characterization and control. Figure supplied by Yuedong Sun for the manuscript with Yu Hu; complete panels a–c. Author-provided manuscript figure

Further publications & manuscripts

Research in print and in progress
04

Towards Sample-Efficient Entanglement Classification for 3–4 Qubit Systems: A Tailored CNN-BiLSTM Approach

Qian Sun†, Yuedong Sun†, Yu Hu, Yihan Ma, Runqi Han, Nan Jiang · † Equal contribution

arXiv:2601.22562Preprint · 2026Co-first authorarXiv ↗
Explore the preprint figure
Complete Figure 1 comparing two CNN–BiLSTM architectures: convolutional features are either flattened or reshaped into a sequence before entanglement classification.
Preprint Fig. 1 · Two ways to connect convolutional features with BiLSTM classification. Q. Sun, Y. Sun et al. (2026), CC BY 4.0.Preprint figure · arXiv (2026) ↗

Towards Sample-Efficient Entanglement Classification for 3–4 Qubit Systems: A Tailored CNN-BiLSTM Approach

Complete Figure 1 comparing two CNN–BiLSTM architectures: convolutional features are either flattened or reshaped into a sequence before entanglement classification.

Preprint Fig. 1 · Two ways to connect convolutional features with BiLSTM classification. Q. Sun, Y. Sun et al. (2026), CC BY 4.0.

05

500 Hz atom–atom entanglement distribution over metropolitan-scale fibres

J.-X. Shi*, Y.-D. Sun*, et al.

Research manuscriptManuscriptCo-first author

CODE, TOOLS & EXPERIMENTS

Selected projects 项目

Tools in detail ↗
8 / 8 repositories
Agent infrastructure
Pre-alpha

VibeMux

A local workspace for coordinating coding agents, isolated worktrees, and durable task histories.

Agent infrastructure
v0.2.0 preview

AI Agent Gateway

One local entry point for model providers, coding tools, routing, and fallback.

Scientific benchmarks
M0 · Scaffold

Quantum Optics Inference

A task framework for agent-based parameter inference from quantum-optical observations.

Workflow sketch
Synthetic countsInfer θphysical modelθ₁θ₂θ₃Reference check
From photon counts to parameters.
Scientific benchmarks
M0 · Scaffold

Rare-earth Crystal Fields

Scientific computing tasks for crystal-field and effective spin Hamiltonians.

Workflow sketch
Ion + crystal siteHfree ion + crystalZeeman + hyperfineProjectg · A · Q
From an ion site to an effective spin.
Scientific benchmarks
M0 · Scaffold

Quantum Circuit Design

A task framework for circuit generation, simulation, and resource-aware evaluation.

Workflow sketch
HRzCircuit codeSimulatestate evolutionEvaluatefidelity + cost
Design a circuit. Check its behavior.
Scientific benchmarks
M0 · Scaffold

QEC & Error Mitigation

A task framework for quantum error correction, decoding, and error mitigation.

Workflow sketch
Syndrome dataDecodeerror modelEvaluatelogical errorspecified seeds + shot budget
Turn error signals into a decoding task.
Developer tools
Open source

Codex State Monitor

Inspect model response metadata, SSE streams, and repeatable API diagnostics.

Workflow sketch
SSE responserequested modelserved modelRecordJSONL
Make response metadata inspectable.
Research workflow
Fork

Zotero–arXiv Daily

A daily reading workflow that ranks new papers against a Zotero library.

Workflow sketch
ZoteroarXivRankresearch fitReading list
A literature library becomes a reading list.

Based on TideDra/zotero-arxiv-daily

Browse all repositories ↗

IN THE LAB & AT THE KEYBOARD

Working toolkit 工具

05

SOFTWARE

Languages & tooling
  • Python
  • Rust
  • TypeScript
  • C++
  • Verilog HDL
  • LaTeX

SYSTEMS

Infrastructure & operations
  • Git worktrees
  • SQLite
  • Linux
  • Mesh networking
  • Agent tooling

RESEARCH

Computation & control
  • Data analysis
  • Scientific computing
  • FPGA control
  • ML pipelines
  • Simulation

OPTICAL

Experimental practice
  • AOD multimode addressing
  • SLM and OAM
  • Beam shaping
  • Crosstalk calibration

LANGUAGES

Communication
  • Chinese · native
  • English · CET-4/6

GET IN TOUCH

Let’s compare notes.

Quantum networks, scientific software, and ideas worth building.