Overcoming the Modality Gap in Context-Aided Forecasting
Introduces CAF-7M and DoubleCast for context-aided probabilistic forecasting with textual information.
Research Output
Publications
Earlier and ongoing research on agents, context-aware forecasting, probabilistic ML, and transportation systems.
4 publications
Selected
Representative work across agents, context-aware prediction, and probabilistic forecasting.
Probabilistic Traffic Forecasting with Dynamic Regression
Extends error-correlation modeling to spatiotemporal forecasting with a matrix-variate autoregressive process and non-isotropic training loss.
Multivariate Probabilistic Time Series Forecasting with Correlated Errors
Introduces an efficient parameterization of cross-covariance matrices for multivariate probabilistic forecasting.
Better Batch for Deep Probabilistic Time Series Forecasting
Uses generalized least squares in the temporal domain to account for autocorrelated errors in deep probabilistic forecasting.
2 publications
Agents, LLMs, and Context
Work on context-aided forecasting, LLM-based forecasting, and foresight-driven agents.
Dr-CiK: A Testbed for Foresight-Driven Agents
Builds a testbed for evaluating foresight-driven agents with context and forecasting-oriented tasks.
Beyond Naive Prompting: Strategies for Improved Zero-shot Context-aided Forecasting with LLMs
Studies prompting, correction, in-context examples, and routing strategies for LLM-based context-aided forecasting.
5 publications
Probabilistic Time Series
Forecasting methods for calibrated uncertainty, correlated errors, and structured prediction.
Probabilistic Time Series Forecasting with Correlated Errors
Dissertation on probabilistic forecasting methods that model temporal, multivariate, and spatiotemporal error correlation.
Frequency-Constrained Learning for Long-Term Forecasting
Adds spectral priors through frequency initialization and constrained optimization for long-horizon forecasting.
Scalable Dynamic Mixture Model with Full Covariance for Probabilistic Traffic Forecasting
Models time-varying matrix-variate error distributions with dynamic Gaussian mixtures for probabilistic traffic forecasting.
Dynamic Modes as Time Representation for Spatiotemporal Forecasting
Uses dynamic mode decomposition to build data-driven time embeddings for long-range seasonal dependencies.
MVG-CRPS: A Robust Loss Function for Multivariate Probabilistic Forecasting
Proposes a robust CRPS-style loss for multivariate Gaussian probabilistic forecasting.
10 publications
Transportation Systems
Earlier work on traffic forecasting, sensing, metro systems, and transportation networks.
Understanding the Predictability of Path Flow Distribution in Urban Road Networks Using an Information Entropy Approach
Uses information entropy to analyze predictability limits for urban path flow distributions.
Traffic Speed Estimation Based on Multi-Source GPS Data and Mixture Model
Estimates urban traffic speed by combining multi-source GPS data with mixture modeling.
A Data-Driven Urban Metro Management Approach for Crowd Density Control
Develops a data-driven metro management method for crowd density control during large crowding events.
Hybrid Model for Predicting Anomalous Large Passenger Flow in Urban Metros
Combines machine learning and complex network modeling for anomalous metro passenger-flow prediction.
Traffic Congestion Spreading Analysis Based on Causal Nexus
Analyzes traffic congestion spreading through causal relationships in transportation networks.
Statistical Analysis of Traffic-Related Social Media Data of Multiple Cities in China
Analyzes traffic-related social media data across multiple Chinese cities.
Understanding Coupling Dynamics of Public Transportation Networks
Studies multiplex public transportation networks and trip reconstruction from smart card data.
Framework for Fusing Traffic Information from Social and Physical Transportation Data
Fuses social media and physical transportation data to improve traffic sensing and analysis.
Traffic Anomaly Detection Method Based on Travel Time of Path
Detects traffic anomalies using path travel-time information.
A Traffic Sensing and Analyzing System Using Social Media Data
Builds a traffic sensing and analysis system using traffic-related information from social media.