Research

Study on Synthetic EHR Models for Subgroup-level Privacy and Fairness

2026 - present,

European Molecular Biology Laboratory (EMBL), University of Helsinki

Responsibilities:

Conduct large-scale benchmarking of existing synthetic-data generation methods for electronic health records (EHRs), comparing predictive utility and subgroup-level behavior.
Develop and evaluate new model architectures for EHR synthetic-data generation under mixed static and longitudinal clinical data settings.
Study subgroup-level fairness in synthetic EHR data, including whether generative models preserve or amplify performance disparities across sensitive groups.

Optimal Hyperparameters for Differentially Private Deep Transfer Learning

2025 - 2026,

Finnish Center for Artificial Intelligence (FCAI), University of Helsinki

Responsibilities:

Systematically studied how clipping norm, batch size, learning rate, privacy budget, training duration, model capacity, and task difficulty affect downstream utility in differentially private transfer learning.
Extended the broader clipping research program by characterizing interactions among optimization and privacy hyperparameters and deriving practical guidance for model tuning.
Contributed to experimental design, large-scale experiment execution, and analysis across image and language models.

Improve the Fairness in Differentially Private Deep Learning with Bounded Adaptive Clipping

2024 - 2026,

Finnish Center for Artificial Intelligence (FCAI), University of Helsinki

Responsibilities:

Investigated subgroup-level performance disparities in differentially private deep learning, with particular focus on subgroup bias induced by gradient clipping.
Developed and studied Bounded Adaptive Clipping, a clipping strategy designed to mitigate disparate impact while preserving the privacy guarantees of DP training.
Led the experimental design, research-method exploration, implementation, and empirical evaluation across image and tabular learning settings.

DPDL: Open-source Experimentation Framework for Differentially Private Deep Learning

Ongoing - Present,

Finnish Center for Artificial Intelligence (FCAI), University of Helsinki

Responsibilities:

Contribute to the development of DPDL, an open-source framework for reproducible experimentation in differentially private deep learning, including training and hyperparameter-optimization workflows.
Participate in research software development and evaluation aimed at making large-scale DP deep-learning experiments easier to configure, reproduce, and extend.
The project is being prepared for a software paper in the Journal of Open Source Software (JOSS). GitHub: https://github.com/TrustworthyMLHelsinki/dpdl

Spectral-based time series map convolutional neural networks traffic flow prediction

May 2022 - July 2022,

Hubei Natural Science Foundation Program

Responsibilities:

Cleaned the data jointly collected.
Research on laplacian matrix calculation and optimization method for graph convolution.
Design and build a graph convolutional network, which combined with GRU.
Train and test the model I build.
Draft paper preparation.

Electric vehicle battery failure prediction

Nov. 2021 - Apr. 2022,

Hubei Natural Science Foundation Program

Responsibilities:

Cleaned the data provided by other members.
Investigate the kinds of problems in failure data, using multiple machine learning methods.
Design and build a bidirectional deep GRU neural networks to predict failures.
Train and test the model I build.
Draft paper preparation.

Educational Quality Evaluation Indexes from the Perspective of UNESCO's Sustainability Goal

July 2021 - Mar. 2022,

National College Student Innovation and Entrepreneurship Training Program

Responsibilities:

Cleaned the data jointly collected
Build the model to fit data
Reviewed the relevant scientific research literature and shared it at the group meetings

Automated Agricultural Machinery based on Lightweight Neural Network and Machine Learning

Jun. 2020 - May 2022,

Internet+’ Innovation and Entrepreneurship Competition

Responsibilities:

Researched the characteristics of crops, and independently counted the scale characteristics and color value characteristics of crops using CenterNet, laying the foundation for follow-up work.
Cooperatively researched the communication and control of remote agricultural machinery manipulation. My responsibilities are mainly research on signal transmission, anti-interference, and encryption methods in a wireless environment.

Recognition Model for Targets in Ocean

Oct. 2020 - Dec. 2020,

Ocean Target Intelligent Perception International Challenge

Responsibilities:

Studied the characteristics of marine targets, and cooperated to study the data enhancement method, Albumentation; Due to large differences in target scales, SNIP was deployed to enhance multi-scale performance.
Built an object detection model, and independently used the SSD model to complete the identification of ocean targets.

Public Transportation Safety Monitoring System

Apr. 2020 - Sept. 2020,

Natural Science Foundation Program of Hubei Province and Siemens Cup China Intelligent Manufacturing Challenge

Responsibilities:

Researched face recognition methods and independently completed the TensorFlow model, implement the recognition, feature vector extraction, comparison, and clustering tasks.
Researched the neural network edge computing, cooperated to complete the development of the embedded terminal based on K210.
Researched graph databases and dual-database methods, independently applied graph database. Proposed the dual graph database method.
Cooperated to complete the system-level test, and my responsibility was mainly data security, neural networks, databases, visualization, etc.

Signal Timing Optimization System for Ecological Intersections based on Edge Computing

Nov. 2019 - Aug. 2020,

National University Student Social Practice and Science Contest on Energy Saving & Emission Reduction

Responsibilities:

Proposed a mathematical model of traffic intersections, and cooperated with classmates to apply fuzzy control and sparrow particle swarm optimization methods, and wrote related programs.
Researched vehicle recognition methods to enhance robustness, independently completed the vehicle recognition TensorFlow model and completed the system construction and code writing of vehicle recognition, statistics, and signal cycle calculations.
Deployed edge computing nodes and configured the operating environment, and independently completed the environment configuration and model deployment on NVIDIA Jetson AGX Xavier, enabling the model to be run on edge computing nodes.
Developed a cloud platform and cooperated with classmates to complete the construction of the web platform and completed the communication between the edge computing node and the cloud platform.
Cooperated to complete the system-level test, and my responsibility was mainly the edge computing node, neural network, and communication parts.

Human Vital Signs Detection Equipment with Optical Fiber Sensing

Aug. 2019 - May. 2020,

Challenge Cup National Undergraduate Curricular Academic Science and Technology Works by Race

Responsibilities:

Studied the separation technology of human heartbeat and respiratory signals and working with graduate students to find a good signal separation plan, finally deployed the Hilbert transform and completed the matrix iterative operation.
Studied the communication between embedded components and servers, server-side programs and databases, and independently developed server-side programs to receive and process human vital signs.
Developed apps and desktop applications, enabling the results to be visually accessed in a wide range using the server-side program in conjunction.
Studied the relationship between signals and health, and summarized the relationship between BCG signal and health with graduate students; The recognition rate of sinus tachycardia, sinus breadiness, and sinus irregularity is almost 100%, and the recognition rate of the sinus with premature atrial is more than 80%.
Independently established machine learning models to analyze human health and provided suggestions.

		University of Helsinki 2024-present Doctoral Researcher in Computer Science Extracurricular Activities Funded by the Finnish Center for Artificial Intelligence (FCAI)
		University of Helsinki 2022-2024 Master's Programme in Data Science Extracurricular Activities Scholarship from the University of Helsinki Scholarship from Open Philanthropy

Moi, I am Linzh

Linzh

Skills

Machine Learning

Deep Learning

Distributed Systems

Signal Processing

Linux

Git

SQL and Non-SQL

Python

R

Flutter

JVM supported languages

Go

Rust

C/C++

Publications

On Optimal Hyperparameters for Differentially Private Deep Transfer Learning

Mitigating Disparate Impact of Differentially Private Learning through Bounded Adaptive Clipping

Design of Remote Driving Wheeled Robot

A method of vehicle flow training and detection based on ResNet50 with CenterNet method

Study on intersection traffic lights timing optimization system based on sparrow algorithm

Neural Network Training in Distribution Computing Method

Research on bus elastic departure interval based on Wavelet Neural Network

Research on Optimization of Intersection Traffic Light Timing Based on Firefly Algorithm

Cassava disease detection method based on EfficientNet

Public transportation safety monitoring system based on edge computing

Research

Study on Synthetic EHR Models for Subgroup-level Privacy and Fairness

Responsibilities:

Optimal Hyperparameters for Differentially Private Deep Transfer Learning

Responsibilities:

Improve the Fairness in Differentially Private Deep Learning with Bounded Adaptive Clipping

Responsibilities:

DPDL: Open-source Experimentation Framework for Differentially Private Deep Learning

Responsibilities:

Spectral-based time series map convolutional neural networks traffic flow prediction

Responsibilities:

Electric vehicle battery failure prediction

Responsibilities:

Educational Quality Evaluation Indexes from the Perspective of UNESCO's Sustainability Goal

Responsibilities:

Automated Agricultural Machinery based on Lightweight Neural Network and Machine Learning

Responsibilities:

Recognition Model for Targets in Ocean

Responsibilities:

Public Transportation Safety Monitoring System

Responsibilities:

Signal Timing Optimization System for Ecological Intersections based on Edge Computing

Responsibilities:

Human Vital Signs Detection Equipment with Optical Fiber Sensing

Responsibilities:

Work

Wuhan QK Information Technology Development Co., Ltd.

Machine Learning Researcher and Engineer

Mobile App Engineer

Backend Engineer

Education

University of Helsinki

Doctoral Researcher in Computer Science

Extracurricular Activities

University of Helsinki

Master's Programme in Data Science

Extracurricular Activities