Generative AI |

Toward an Auto-Synthesizer: Stress-Aware Selection and Recommendation of Deep Tabular Generative Models

Tue, 08 Sep 2026 00:00:00 +0000

Committee: Guang Cheng (chair), Fredric Schoenberg, Michael Tsiang.

Synthony

Tue, 31 Mar 2026 00:00:00 +0000

An agent that recommends the best-fit tabular generative model (GAN / VAE / Diffusion / DP-based / Bayesian / Normalizing Flow) for a given dataset by profiling its “stress” characteristics — privacy sensitivity, class imbalance, cardinality, sample size — instead of defaulting to one architecture. Implements rule-based, LLM-based, and hybrid ranking strategies with explainable tie-breaking and benchmark-driven comparison. Published at ICLR 2026 DeLTA.

SYNTHONY: A Stress-Aware, Intent-Conditioned Agent for Deep Tabular Generative Models Selection

Tue, 31 Mar 2026 00:00:00 +0000

SYNTHONY treats synthesizer selection as a matching problem: it profiles a dataset’s “stress” — privacy sensitivity, class imbalance, cardinality, and sample size — and matches that profile against a calibrated capability registry of synthesizer families (GAN, VAE, Diffusion, DP-based, Bayesian, Normalizing Flow). It supports rule-based, LLM-based, and hybrid ranking with explainable tie-breaking and benchmark-driven comparison.

table-synthesizers

Thu, 15 Jan 2026 00:00:00 +0000

Unified 16 synthesizer architectures (CTGAN, TVAE, TabDDPM, AutoDiff, TabSyn, AIM, PATECTGAN, DPCART, Bayesian Network, ARF, NFlow, GREAT, and more) behind a single factory interface with per-model production configs. Built the end-to-end training → generation → evaluation loop, integrating with alfred-analytica for fidelity and machine-learning-efficacy scoring against real data. Deployed distributed training on NCSA DeltaAI (SLURM, H100 GPU nodes) with reproducible experiment tracking and versioned experiment outputs.

Table-Synthesizers is the generation and measurement substrate of my Master’s thesis, .