However, the landscape of6 synthetic data research has been fragmented due to the large number of data7 modalities(e.g.,tabulardata,timeseriesdata,images,etc.) andvarioususecases8 (e.g., privacy, fairness, data augmentation, etc.). Beyond benchmarking, it also offers a single access point to a diverse range of15 cutting-edge data generators.

Visual Decathlon [53] gathers 10 diversedatasets.

Itconsists4-regular 10 different passingapproaches Table 1: Classification Method Mean MP-GNNs 0.000 10.000 RingGNN 0.000 10.000 3WLGNN 10.916 30.000

It is widely acknowledged that large language models (LLMs) encode a vast reservoirofknowledge after being trained onmass data.

Bandits, Reinforcement Learning

There have been multiple recent approaches to obtain a near-optimal policy in CMDPs in the regret-minimization or PAC-RL settings [13, 38, 9, 19, 31, 22, 36, 12, 15, 16, 11].

The quadratic computational and memory complexities of the Transformer'sattention mechanism have limited its scalability for modeling long sequences.