# US 2026/0017529 A1 · Apparatus and method for reproducing tabular data

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2026/0017529 A1
- Document ID: US20260017529A1
- Status: Published application
- Kind: A1
- Application number: 18/918,943
- Filed: 2024-10-17
- Published: 2026-01-15
- Earliest claimed: 2024-07-11
- Assignees: Electronics and Telecommunications Research Institute
- Applicants: Electronics and Telecommunications Research Institute
- Inventors: Sung Yup LEE; Cheol Ho KIM; Byoung Hwa LEE; Jung Hoon LEE
- Web page: https://patentgrep.com/patent/US20260017529A1

## CPC

- G06N 3/092: Reinforcement learning
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

The present invention relates to an apparatus for reproducing tabular data. The apparatus for reproducing tabular data includes a generative flow networks (GFlowNets) learning network, a critic network, and a processor that performs learning through the GFlowNets learning network and the critic network, in which the processor performs learning of a policy network of GFlowNets and learning using real data, and performs learning of the critic network based on the real data and data generated by the GFlowNets.

## Claims

### Claim 1

An apparatus for reproducing tabular data, comprising:
a generative flow networks (GFlowNets) learning network;
a critic network; and
a processor that performs learning through the GFlowNets learning network and the critic network,
wherein the processor performs learning of a policy network of GFlowNets and learning using real data, and performs learning of the critic network based on the real data and data generated by the GFlowNets.

### Claim 2 (depends on 1)

The apparatus of claim 1, wherein, prior to the learning of the policy network of the GFlowNets, the processor preprocesses a continuous variable as a training data preprocessing process, and converts the preprocessed continuous variable into a categorical variable and a one-hot vector.

### Claim 3 (depends on 1)

The apparatus of claim 1, wherein the processor samples a predefined variable value (c) from a Bernoulli distribution with a specified probability in a process of learning the policy network of the GFlowNets.

### Claim 4 (depends on 3)

The apparatus of claim 3, wherein the predefined variable value (c) is a value for selecting a routine for the GFlowNets to perform learning using the real data or a routine for learning the data generated by the GFlowNets.

### Claim 5 (depends on 3)

The apparatus of claim 3, wherein, when the predefined variable value (c) is a specified first specific value, the processor samples a trajectory from an end state to a start state using a reverse policy of the GFlowNets.

### Claim 6 (depends on 3)

The apparatus of claim 3, wherein, when the predefined variable value (c) is a specified second specific value, the processor samples a trajectory from a start state to an end state using a forward policy of the GFlowNets.

### Claim 7 (depends on 3)

The apparatus of claim 3, wherein, in the process of learning the policy network of the GFlowNets, the processor updates all parameters of the GFlowNets using a trajectory balance loss.

### Claim 8 (depends on 1)

The apparatus of claim 1, wherein, in the process of performing the learning of the critic network, the processor samples a new end state using a forward policy of the GFlowNets.

### Claim 9 (depends on 8)

The apparatus of claim 8, wherein, in the process of performing the learning of the critic network, the processor updates a parameter corresponding to a reward of the critic network using an objective function of Wasserstein GAN with gradient penalty (WGAN-GP) based on training data converted into a one-hot vector and the sampled new end state.

### Claim 10 (depends on 9)

The apparatus of claim 9, wherein the critic is a value indicating how close data given as an input is to the actual data, and the closer the data generated by the GFlowNets is to the actual data, the greater the reward, and the farther the generated data is from the actual data, the smaller the reward.

### Claim 11

A method of reproducing tabular data that causes a processor to perform the processes of:
learning a policy network of generative flow networks (GFlowNets);
performing GFlowNets learning based on real data;
performing exploration and learning through the policy network learned by the GFlowNets; and
learning a critic network.

### Claim 12 (depends on 11)

The method of claim 11, further comprising, prior to the process of learning the policy network of the GFlowNets, as a training data preprocessing process, a process of preprocessing a continuous variable and converting the preprocessed continuous variable into a categorical variable and a one-hot vector.

### Claim 13 (depends on 11)

The method of claim 11, wherein, in the process of learning the policy network of the GFlowNets, a predefined variable value (c) is sampled from a Bernoulli distribution with a specified probability.

### Claim 14 (depends on 13)

The method of claim 13, wherein the predefined variable value (c) is a value for selecting a routine for the GFlowNets to perform learning using the real data or a routine for learning the data generated by the GFlowNets.

### Claim 15 (depends on 13)

The method of claim 13, wherein, when the predefined variable value (c) is a specified first specific value, a trajectory from an end state to a start state is sampled using a reverse policy of the GFlowNets.

### Claim 16 (depends on 13)

The method of claim 13, wherein, when the predefined variable value (c) is a specified second specific value, a trajectory from a start state to an end state is sampled using a forward policy of the GFlowNets.

### Claim 17 (depends on 13)

The method of claim 13, wherein, in the process of learning the policy network of the GFlowNets, all parameters of the GFlowNets are updated using trajectory balance loss.

### Claim 18 (depends on 11)

The method of claim 11, wherein the process of performing the learning of the critic network includes sampling a new end state using a forward policy of the GFlowNets.

### Claim 19 (depends on 18)

The method of claim 18, wherein, in the process of performing the learning of the critic network, a parameter corresponding to a reward of the critic network is updated using an objective function of Wasserstein GAN with gradient penalty (WGAN-GP) based on training data converted into a one-hot vector and the sampled new end state.

### Claim 20 (depends on 19)

The method of claim 19, wherein the critic is a value indicating how close data given as an input is to the actual data, and the closer the data generated by the GFlowNets is to the actual data, the greater the reward, and the farther the generated data is from the actual data, the smaller the reward.
