Department of Computer Science • Carl R. Ice College of Engineering • Kansas State University
A Founding Laboratory of the K-State Center for Artificial Intelligence and Data Science (CAIDS)
🌐 KDDResearch.org • 🤗 Hugging Face • 🦋 Bluesky (@kddresearch.bsky.social) • 𝕏 X (@kstate_bigdata) • 📦 Public Releases
🧬 Core Research Divisions (Public Release Directories)
💬 Natural Language [NL] | 👁️ Computer Vision [CV] | 🤖 Reinforcement Learning [RL] | 🛡️ Secure & Trustworthy AI [ST] | 🌾 Other Applied AI [OA]
Established in 1999 by Prof. William H. Hsu (@banazir), the Laboratory for Knowledge Discovery in Databases (KDD Lab) is the longest-running continuous artificial intelligence and machine learning research group in the Department of Computer Science at Kansas State University, and one of the five founding laboratories of the Center for Artificial Intelligence and Data Science (CAIDS).
Our research centers on applied artificial intelligence (AI), machine learning (ML), data mining and knowledge discovery from large spatial, temporal, and multimodal databases, neurosymbolic and probabilistic reasoning, human–computer intelligent interaction (HCII), and knowledge-based software engineering (KBSE) for decision support and autonomous systems.
In our research, we systematically decompose complex analytical learning and reasoning problems using information-theoretic, probabilistic, and structural criteria so that the most appropriate inductive, symbolic, or hybrid methods can be applied to the resulting transformed representations:
- Representation Learning, Problem Decomposition & Multimodal Fusion: A central challenge in modern AI is designing mostly-unsupervised, weakly-supervised, and active learning methods—alongside hyperparameter and inductive-bias optimization—to decompose high-dimensional tasks effectively. By addressing the high-level control of inductive learning in a statistically and logically sound fashion, we advance techniques for both model selection and model integration (including multimodal sensor fusion and cross-modal constraint satisfaction) across computation-intensive problems in uncertain reasoning, open information extraction, and control automation.
- World-Model Grounding & Epistemic Trust: Rather than treating machine learning models as opaque black boxes or relying on scalar confidence heuristics, we investigate how adaptive artifacts—whether grounded in Bayesian networks and probabilistic graphical models, deep neural and foundation models, symbolic ontologies, or evolutionary computation—interact with human users and domain constraints. We treat information trust as a composite function of evidence corroboration, provenance traceability, logical consistency, and counterfactual robustness.
- End-to-End Intelligent Systems: Important applications of this paradigm across our work include open-domain and scientific document understanding, multimodal knowledge graph construction, explainable visual perception, causal and multi-objective reinforcement learning, intelligent visualization, and high-performance agentic systems for research and instruction.
The KDD Lab is organized into five interlocking research divisions that connect foundational theory in language, perception, control, and epistemic trust to deployable real-world systems. (Note: The majority of our active scientific infrastructure is maintained in private GitOps ledgers; vetted artifacts, models, and code are exported to the public_releases hub).
| Division | Core Mission & Role in the Lab Vision | Representative Research Thrusts |
|---|---|---|
💬 Natural Language [NL](Semantic Backbone) |
Focuses on representation, extraction, reasoning, trust, and longitudinal modeling of meaning in text, speech, and multimodal language artifacts. Provides mechanisms for stabilizing meaning from unstructured sources, propagating provenance and uncertainty, and grounding language models in symbolic and domain knowledge. | • Open Information Extraction & Learning to Rank • Knowledge Graph Construction & Ontology Grounding • Code-Mixed & Multilingual Speech Recognition • Topical Knowledge Maps & Meta-Research |
👁️ Computer Vision [CV](Perceptual Grounding) |
Focuses on structured perception, hierarchical part–whole interpretation, and explainable visual understanding tightly coupled with language and decision-making. Treats vision not merely as pattern recognition, but as perceptual grounding for interpretable world models. | • Ecological & Pollinator Computer Vision • Biomedical & Cellular Image Analysis • Biomechanics & Behavioral Video Analytics • Agricultural & Livestock Vision |
🤖 Reinforcement Learning [RL](Decision & Control Layer) |
Focuses on learning, reasoning, and sequential decision-making under uncertainty where state, reward, and policy representations are structured, multi-objective, or non-stationary. Emphasizes causality over correlation, counterfactual reasoning, and human-interpretable policies. | • Inverse Reinforcement Learning & Behavioral Modeling • RL for Precision Agriculture & Autonomous Control |
🛡️ Secure & Trustworthy ML & Reasoning [ST](Epistemic & Ethical Scaffolding) |
Treats reliability, robustness, interpretability, fairness, and value alignment as first-class modeling objectives rather than post-hoc constraints. Emphasizes epistemic humility, verifiable provenance, and graceful failure in open-world and high-stakes environments. | • Cross-Divisional Alignment, Safety & Verification • Latent Objective State (LOS) & Multi-Objective Alignment |
🌾 Other Applied AI [OA](Translational & Systems Layer) |
Bridges foundational research in language, vision, RL, and trustworthy reasoning to real-world deployment constraints—addressing compute efficiency, memory, latency, containerized reproducibility, and human-in-the-loop interaction. | • Hydrological & Agricultural Decision Support • Applied AI & Engineering Extension |
Across three decades of research, the KDD Lab's publications, grants, and dissertations are organized around 16 cross-cutting methodological themes:
| Theme ID | Research Theme | Core Scientific Focus |
|---|---|---|
KDD-01 |
Sustainable Machine Learning via Compression | Automatic synthesis of compression techniques, feature selection wrappers, model distillation, quantization, pruning, and energy-/memory-aware inference under real-world resource constraints. |
KDD-02 |
Information Trust & World-Model Grounding | Bayesian network inference, probabilistic graphical models, and structured uncertainty representation—modeling trust as a composite of evidence agreement, logical consistency, and provenance. |
KDD-03 |
Mostly-Unsupervised Dynamic Topic Modeling | Hierarchical Bayesian and continuous-time topic models (including HDP-based architectures) for tracking semantic drift and evolving concepts across scientific and social corpora under limited supervision. |
KDD-04 |
Meta-Research Pipelines from Scientific Literature | End-to-end pipelines integrating layout detection, structural parsing, metadata extraction, and named entity recognition for automated synthesis of large scientific corpora. |
KDD-05 |
Trust-Weighted Aggregation & Evidence Fusion | Principled ensemble learning, probabilistic graphical fusion, class-imbalance-aware modeling, and cross-modal corroboration for transparent belief updating from heterogeneous sources. |
KDD-06 |
Hierarchical Part–Whole Reasoning | Relational graphical models and compositional representations of components and relations across biological, ecological, social, and engineered systems. |
KDD-07 |
Vision-Based Structured Representation | Interpretable visual understanding spanning document layout analysis, anomaly detection, agricultural remote sensing, behavioral scoring, and fine-grained species classification. |
KDD-08 |
Multimodal Ontology Learning | Unifying symbolic ontologies and knowledge graphs with neural embeddings to support cross-domain generalization, weak supervision, and verifiable inference. |
KDD-09 |
End-to-End Active Learning Systems | Layered learning and adaptive perception–reasoning–action pipelines for robotics, crisis monitoring, query synthesis, and high-performance computing optimization. |
KDD-10 |
Value Alignment & Robustness | Adversarial reinforcement learning, safety benchmarking, fairness interventions, and latent objective preservation under constrained or adversarial operating conditions. |
KDD-11 |
Sentic & Affective Computing | Modeling sentiment, stylistic variation, code-switching pragmatics, and affective signals across natural language, speech, and multimodal human–computer interaction. |
KDD-12 |
Neurosymbolic & Relational Reasoning | Hybrid architectures bridging statistical/neural representation learning with formal logic, ontologies, and probabilistic relational models for explainable inference. |
KDD-13 |
Causal & Counterfactual Reasoning | Bayesian decision networks, structural causal models, and spatio-temporal/epidemiological simulations for counterfactual evaluation and robust decision support. |
KDD-14 |
Multimodality as Constraint | Exploiting cross-modal consistency across text, audio, video, and structured metadata as a self-supervised and structural constraint for extraction and generation. |
KDD-15 |
Rich Representations for Policy Learning | Structured policy representations, inverse reinforcement learning, hyperbolic temporal discounting, and multi-horizon / multi-objective Markov decision processes. |
KDD-16 |
Multiresolution Explanation & Interpretability | Feature attribution, task-referenced quality critiquing, intelligent visualization, and structural explanation across perception, language, and control layers. |
The KDD Lab integrates graduate and undergraduate research training with reproducible, artifact-first software and machine learning engineering:
- Courses & Open Curriculum: Prof. Hsu instructs core and advanced courses in the K-State Department of Computer Science, including CIS 530/730 (Artificial Intelligence), CIS 531/731 (Foundations of Data Science & Analytics / Knowledge Discovery), CIS 732 (Machine Learning & Pattern Recognition), CIS 830 (Advanced Topics in AI / Large Language Models), and CIS 837 (Reinforcement Learning for Robotics & Autonomous Agentic AI), as well as intensive short courses on Applied LLM Engineering. Public course templates and starter materials are available via
public_releases. - Lab Colloquium & Reading Group: The lab hosts a weekly research colloquium and graduate reading group.
- Prospective Students: Kansas State University graduate (Ph.D. and M.S.) and undergraduate students interested in thesis/dissertation research or independent study in
[NL],[CV],[RL],[ST], or[OA]are encouraged to review our active repositories and publications at KDDResearch.org before reaching out with a research statement and CV.
- Official Lab Portal: https://www.kddresearch.org
- Hugging Face Organization: https://huggingface.co/KDDResearch
- Bluesky: https://bsky.app/profile/kddresearch.bsky.social
- X (Twitter): https://x.com/kstate_bigdata
- Department of Computer Science, Kansas State University: https://cs.k-state.edu
