## Papers

Possible benchmarks, well written and worth mentioning:
- TabDDPM: Modelling Tabular Data with Diffusion Models
- Time-series Generative Adversarial Networks

CTAB-GAN: Effective Table Data Synthesizing

A Review of Tabular Data Synthesis Using GANs on an IDS Dataset

Tabular Transformers for Modeling Multivariate Time Series

CTAB-GAN: Effective Table Data Synthesizing

Modeling Tabular Data using Conditional GAN

Tabular Transformers for Modeling Multivariate Time Series

Conditional Tabular GAN-Based Two-Stage Data Generation Scheme for Short-Term Load Forecasting

TabFairGAN: Fair Tabular Data Generation with Generative
Adversarial Networks

Awesome-timeseries-spatiotemporal-lm-llm#large-language-models-and-foundation-models-llm-lm-fm-for-time-series-and-spatio-temporal-data
https://github.com/qingsongedu/awesome-timeseries-spatiotemporal-lm-llm

## Reviews:

Time-series Generative Adversarial Networks

- autoregressive models for sequence prediction
- GAN-based methods for sequence generation
- time-series representation learning

- Let $S$ be a vector space of static features
- $X$ of temporal features

$P(S,X_{1:t-1})=p(S)\prod_tp(X_t|X_{1:t-1})$


## TabDDPM: Modelling Tabular Data with Diffusion Models

Кажется, TabDDPM дает нормальный план для статьи и бенчмарки по чуть чуть вырисовываются

## TabDDPM: Modelling Tabular Data with Diffusion Models

1. Conditional generation on class or real value quantile (need to think how to avoid leakage in embedding )
2. Time embeddings with SILU?
3. Generated dataset distributions?
4. Privacy analysis - focus on privacy in the сделать более глубокий чем в TABDPPM
(Use a model based approach to find specific users)

Benchmarks:
4. Compare with SMOTE approach / adapt SMOTE for sequences (generation)
5. Compare with Time-series Generative Adversarial Networks (generation + cls)
paper.
6. CoLESS (cls only)



Кажется, TabDDPM дает нормальный план для статьи и бенчмарки по чуть чуть вырисовываются

## TabDDPM: Modelling Tabular Data with Diffusion Models

1. Conditional generation on class or real values quantiles (need to think how to avoid leakage in embedding )
2. Time embeddings with SILU?
3. Generated dataset distributions?
4. Privacy analysis - focus on privacy in the
(Use a model based approach to find specific users)

Benchmarks:
4. Compare with SMOTE approach / adapt SMOTE for sequences (generation)
5. Compare with Time-series Generative Adversarial Networks (generation + cls)
paper.
6. CoLESS (cls only)