Yan Liu (刘燕) is a distinguished computer scientist whose scholarly pursuits lie at the intersection of advanced machine learning techniques and real‑world applications that span climate science, transportation systems, digital communication platforms, and biomedical research. She holds the Fletcher Jones Foundation Endowed Chair Professorship across three departments—Computer Science, Electrical and Computer Engineering, and Biomedical Sciences—within the USC Viterbi School of Engineering at the University of Southern California. Her appointment reflects a commitment to interdisciplinary inquiry, blending algorithmic innovation with domain‑specific expertise to address complex, data‑rich problems.
Academic Position and Institutional Context
USC Viterbi School of Engineering
The Viterbi School, named after the eminent engineer and professor Andrew Viterbi, is renowned for its emphasis on engineering research that tackles societal challenges. Located in Los Angeles, it offers a robust ecosystem for cross‑disciplinary collaboration, particularly between computer science, electrical engineering, and biomedical engineering.
Departments of Computer Science, Electrical and Computer Engineering, and Biomedical Sciences
Yan Liu’s professorship spans three distinct yet interrelated departments:
- Computer Science: Focuses on algorithmic theory, artificial intelligence, and data analytics.
- Electrical and Computer Engineering (ECE): Provides a foundation in signal processing, systems engineering, and hardware‑accelerated computation.
- Biomedical Sciences: Embraces computational biology, medical imaging, and translational research that leverages data for clinical insights.
By holding positions in all three, Liu is uniquely positioned to translate theoretical advances into tangible solutions across multiple scientific fronts.
Fletcher Jones Foundation Endowed Chair
The Fletcher Jones Foundation, a philanthropic organization supporting higher education and research, endows the chair that Liu occupies. Endowed chairs are typically reserved for faculty who demonstrate exceptional promise and leadership in their fields. This recognition underscores Liu’s standing as a pioneer in machine‑learning research with broad societal impact.
Research Focus Areas
Yan Liu’s research portfolio centers on three core technical themes, each of which underpins a suite of application domains. The following sections provide a conceptual overview of these themes, the types of problems they address, and why they matter in contemporary science and industry.
Machine Learning for Time Series
What is Time‑Series Data?
Time‑series data are sequences of observations indexed by time. They appear in countless contexts—from sensor readings in autonomous vehicles to daily temperature records used in climate studies. The temporal ordering introduces unique challenges: autocorrelation, non‑stationarity, and the need for forecasting.
Why Time‑Series ML Matters
Traditional statistical methods often fall short when data exhibit complex, nonlinear dynamics or when the dimensionality is high. Machine learning models—especially deep neural networks—have proven adept at capturing such intricacies. Applications include:
- Predictive Maintenance: Forecasting equipment failures in industrial settings.
- Energy Consumption Forecasting: Optimizing grid operations.
- Financial Modeling: Predicting stock price movements.
Yan Liu’s work focuses on developing algorithms that can learn from limited or noisy time‑series data, ensuring robust predictions even when observations are sparse or irregular.
Explainable Machine Learning
The Explainability Imperative
As machine learning models grow in complexity, their decision processes become opaque. In high‑stakes domains—healthcare, finance, autonomous driving—stakeholders require transparent reasoning to trust and audit model outputs.
Techniques in Explainable AI (XAI)
Common approaches include:
- Model‑agnostic explanations: Methods like SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model‑agnostic Explanations) provide post‑hoc insights into model predictions.
- Interpretable model design: Building simpler models (e.g., decision trees, rule lists) that are inherently transparent.
- Visualization tools: Graphical representations that reveal feature importance and decision boundaries.
Liu’s research explores how to embed explainability into models dealing with time‑series data and physics‑constrained systems, ensuring that predictions are not only accurate but also interpretable by domain experts.
Physics‑Informed AI
Integrating Physical Laws
Physics‑informed AI (PI‑AI) incorporates known physical relationships—such as conservation laws, differential equations, or symmetry constraints—into machine learning frameworks. This hybrid approach offers several advantages:
- Data Efficiency: Leveraging physics reduces the amount of data needed for training.
- Generalization: Models respect underlying laws, improving performance on unseen scenarios.
- Safety: Enforcing physical constraints mitigates risk in safety‑critical applications.
Typical PI‑AI Frameworks
- Physics‑informed neural networks (PINNs): Neural networks that embed differential equations into loss functions.
- Hybrid models: Combining mechanistic simulations with data‑driven components.
- Constraint‑based regularization: Penalizing deviations from known physical laws during training.
Liu’s work harnesses PI‑AI to tackle problems where data alone are insufficient or where adherence to physical realism is paramount, such as climate modeling or biomedical signal interpretation.
Applications of Liu’s Research
While the technical themes are broad, Liu’s research explicitly targets four application areas, each benefiting from the synergy of time‑series ML, explainability, and physics constraints.
Climate Modeling
The Challenge
Climate systems are governed by complex, multiscale physical processes. Traditional numerical models require massive computational resources and can be limited by coarse spatial resolution.
Machine Learning Contributions
Data‑driven approaches can:
- Accelerate simulations: Surrogate models approximate expensive climate dynamics.
- Enhance resolution: Super‑resolution techniques reconstruct fine‑scale details from coarse data.
- Integrate heterogeneous data: Combining satellite observations, ground stations, and model outputs.
Liu’s focus on physics‑informed AI ensures that surrogate climate models remain faithful to fundamental atmospheric and oceanic laws, while explainable components help climatologists interpret model predictions.
Transportation Planning
The Problem Space
Modern transportation networks must accommodate fluctuating demand, limited infrastructure, and environmental constraints. Predictive models inform route optimization, traffic signal timing, and fleet management.
ML Applications
- Demand forecasting: Predicting ridership or traffic volumes.
- Anomaly detection: Identifying unusual congestion patterns.
- Optimization: Planning routes that minimize travel time or emissions.
By applying time‑series ML to traffic data and embedding physical constraints such as road capacities or vehicle dynamics, Liu’s research contributes to smarter, more sustainable mobility solutions.
Social Media
Data Characteristics
Social media platforms generate vast streams of user interactions, content, and network dynamics. These datasets are high‑dimensional, noisy, and temporally evolving.
Analytical Goals
- Sentiment analysis: Tracking public opinion over time.
- Information diffusion: Modeling how content spreads.
- Anomaly detection: Identifying coordinated misinformation campaigns.
Explainable AI is particularly valuable in this domain, allowing platform moderators and policymakers to understand the drivers behind algorithmic content curation or moderation decisions.
Biomedicine
Biomedical Data Challenges
Biomedical datasets—such as electrocardiograms, genomic sequences, or medical imaging—often contain irregular time‑series signals and are subject to stringent privacy and regulatory constraints.
ML in Healthcare
- Disease progression modeling: Forecasting patient trajectories.
- Diagnostic support: Assisting clinicians with automated image interpretation.
- Personalized treatment: Tailoring interventions based on individual data patterns.
Physics‑informed AI can incorporate physiological constraints (e.g., cardiac electrophysiology) into predictive models, while explainability ensures clinicians can trust algorithmic recommendations.
Significance and Impact
Yan Liu’s research sits at the nexus of algorithmic innovation and real‑world problem solving. By advancing methods that are both powerful and interpretable, her work helps bridge the gap between abstract machine‑learning theory and actionable insights across diverse domains.
- Societal Benefit: Improved climate predictions inform policy; smarter transportation reduces congestion and emissions; better biomedical models enhance patient care.
- Scientific Advancement: Integrating physics into AI fosters more robust, data‑efficient models that can tackle previously intractable problems.
- Educational Leadership: Her cross‑departmental role promotes interdisciplinary training for graduate students, preparing a new generation of researchers capable of tackling complex, multi‑disciplinary challenges.
Interdisciplinary Collaboration
The breadth of Liu’s research naturally encourages collaboration across multiple fields:
- Computer Science & Electrical Engineering: Joint efforts on algorithm design, high‑performance computing, and hardware acceleration.
- Biomedical Sciences: Partnerships with clinicians and biologists to apply machine‑learning models to patient data and experimental studies.
- Climate and Environmental Sciences: Collaborative projects with atmospheric scientists and geophysicists to improve climate model fidelity.
These collaborations exemplify the modern scientific paradigm, where breakthroughs often emerge from the confluence of expertise.
Current Projects (Conceptual Overview)
While specific project details are not publicly disclosed in the source, typical endeavors aligned with Liu’s stated interests might include:
- Physics‑Constrained Forecasting for Urban Heat Islands
A project that blends satellite temperature data with atmospheric physics to predict heat‑island intensity, aiding city planners in designing mitigation strategies.
- Explainable Traffic Flow Prediction
Developing a time‑series model that forecasts congestion on major highways, accompanied by interpretable visualizations highlighting key contributing factors (e.g., weather, incidents, construction).
- Surrogate Climate Models for Extreme Event Prediction
Building surrogate neural networks that approximate full climate simulations, allowing rapid exploration of scenario space while maintaining physical consistency.
- Biomarker Discovery in Longitudinal Health Studies
Applying explainable ML to irregularly sampled patient data to identify early indicators of disease progression.
Each of these projects would leverage Liu’s core methodological strengths—time‑series analysis, explainability, and physics integration—to produce actionable insights.
Future Directions
The research landscape in machine learning is evolving rapidly, and Liu’s work is poised to adapt and influence several emerging trends:
- Causal Inference in Time‑Series: Moving beyond correlation to uncover causal mechanisms in dynamic systems.
- Federated Learning for Sensitive Domains: Enabling collaborative model training across institutions while preserving data privacy, especially relevant in biomedicine.
- AI‑Driven Policy Simulation: Using explainable models to assess the potential impact of climate or transportation policies before implementation.
- Hybrid Quantum‑Classical ML: Exploring how quantum computing might accelerate physics‑informed neural network training.
By staying at the forefront of these developments, Liu’s research will continue to shape how data‑driven insights are integrated into complex, real‑world systems.
Conclusion
Yan Liu (刘燕) exemplifies the modern interdisciplinary scientist, leveraging cutting‑edge machine‑learning techniques to address some of the most pressing challenges in climate science, transportation, social media, and biomedicine. Her appointment as the Fletcher Jones Foundation Endowed Chair across three departments at USC Viterbi underscores both her academic excellence and the transformative potential of her research. While the specific outcomes of her projects are not detailed in the source, the breadth of her focus areas suggests a career dedicated to building models that are accurate, interpretable, and grounded in physical reality—qualities essential for responsible AI deployment in society.