Projects
Research Topics and Approach
Conceptulization with stochastic sets: regimes, transitions and composition
- Atomic stochastic sets: the regimes of Gibbs models (high entropy) and Sparsity models (low Entropy)
- The spectrum of entropy regimes: information scaling and model transition
- Conceptualization by composition and stochastic grammar (middle entropy)
Representation: probabilistic And-or Graphs
- S-AoG: representing the spatial hierarchy---scenes, objects, parts and primitives.
such as, indoor and outdoor scenes, Human poses, Vehicles, and attributes. - T-AoG: representing the temporal hierarchy---event, action and motion primitive.
such as human vehicle interactions, group activities in surveillance video - Summary: unification on visual knowledge representation and reasoning
Inference: goal guided, cost sensitive computing
- The alpha, beta, gamma and omega processes in the AoG inference
- Optimal decision policy
- Tracking and online learning
- Efficiency of sensors and features for nodes in the AoG
- Scheduling the computational process: exploiting sensor efficacy and utility
- SEE Inference in video: PTZ tracking guided by utility, sensor efficacy and AoG
Theory of Learning: information projection, learnability and generalizability
- A unified learning theory based on information projection
- PAC-learning of the inductively learned sets for nodes in the AoG
- Generalizability: capacity and transportability of the S/T/C-AoGs
- Curriculum learning and energy landscapes
Text parsing and probabilistic query answering
- Lexical and syntactic processing
- Semantic tagging and information integration for Joint Parse Graph
- Text query parsing and updating utility functions from queries
- Evaluation tool for text query
Airborne imagery registration and understanding
- Geo-registration of WAMI and MAMI images
- Target tracking in Airborn images
- Integrating parse graphs from Airborne and ground sensors