Statistical ML background with focus on industrial applications. Interested in bringing research to practice.
Probability theory, optimization, and statistical inference. Theoretical grounding for reliable ML systems.
Transformers, computer vision, time-series analysis. Published research and practical implementations.
Applying ML research to real-world systems. From theory to production.
Research publications, academic programs, and invited presentations.
Enhancing AI with Advanced Sampling and Data Splitting Techniques for Big Data
Proposed a robust data splitting methodology for ML based on statistical ML theory.
Theoretical framework for reliable train-test splits that preserve statistical properties.
International gathering of statisticians and data scientists.
Quantization-aware matrix multiplication through the lens of rate-distortion theory for efficient model compression.
Deriving theoretical lower bounds for Kernel Ridge Regression (KRR) and Gradient Descent optimization.
Statistical analysis in high-dimensional regimes where classical theory breaks down.
Intensive program with leading researchers in theoretical ML.
Detecting 'bradykinesia' - subtle movement hesitations in Parkinson's patients.
LSTM-FCN system detecting movement hesitations and amplitude decay in finger-tapping videos.
80.3% accuracy on 1,396 videos from 310 patients.
Hierarchical classification for temporal patterns of parkinsonian movement.
Current and past research projects.
Building scalable variable selection and ranking algorithms for noisy datasets.
AI-driven radiotherapy optimization using sensitivity analysis for treatment planning.
Robust data splitting methodology for ML based on statistical theory. Presented at JSM 2025.
Theory of AI · AI Agent · Industrial AI
Why is matrix multiplication the bottleneck in LLM inference? We formulate the rate-distortion problem for matmul and review existing weight-only quantization methods.
Introduction to Anthropic's Claude Cookbooks repository. Covers classification, summarization, and text-to-SQL capabilities.
A technical analysis of Palantir's core architecture. Why the Ontology is distinct from a Semantic Layer, the mechanics of write-back capabilities, and the trade-offs of an object-centric data model.
Get curated insights or reach out directly