# US 2026/0037812 A1 · Policy generating apparatus and method for slm

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

- Number: US 2026/0037812 A1
- Document ID: US20260037812A1
- Status: Published application
- Kind: A1
- Application number: 19/284,480
- Filed: 2025-07-29
- Published: 2026-02-05
- Earliest claimed: 2024-08-01
- Assignees: Research & Business Foundation Sungkyunkwan University
- Applicants: Research & Business Foundation Sungkyunkwan University
- Inventors: Hong Uk WOO; Won Je CHOI; Woo Kyung KIM; Min Jong YOO
- Web page: https://patentgrep.com/patent/US20260037812A1

## CPC

- G06N 3/0895: Weakly supervised learning, e.g. semi-supervised or self-supervised learning
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

It is about the policy generating apparatus and method for SLM, the policy generating method for SLM may comprise receiving an expert dataset, generating a rationale dataset based on the expert dataset and a pre-stored initial rationale set, verifying the rationale dataset through a self-verification function, learning a reasoning policy through an embodied knowledge graph based on the verified rationale dataset, learning a planning policy based on a rationale set of the learned reasoning policy and a planning policy reconstruction loss and generating an SLM policy including a final reasoning policy and a planning policy based on the learned reasoning policy and the planning policy.

## Claims

### Claim 1

A policy generating method for SLM, the method comprising:
receiving an expert dataset;
generating a rationale dataset based on the expert dataset and a pre-stored initial rationale set;
verifying the rationale dataset through a self-verification function;
learning a reasoning policy through an embodied knowledge graph based on the verified rationale dataset;
learning a planning policy based on a rationale set of the learned reasoning policy and a planning policy reconstruction loss; and
generating an SLM policy including a final reasoning policy and a planning policy based on the learned reasoning policy and the planning policy.

### Claim 2 (depends on 1)

The policy generating method for SLM of claim 1,
wherein the learning a planning policy comprises generating the rationale set of the inferred reasoning policy based on a policy of an encoder based on an encoder prompt pool, a policy of a decoder based on a decoder prompt pool, and an attention module while performing the learning of the reasoning policy.

### Claim 3 (depends on 2)

The policy generating method for SLM of claim 2,
wherein the encoder prompt pool comprises a prefix prompt and a postfix prompt.

### Claim 4 (depends on 2)

The policy generating method for SLM of claim 2,
wherein the rationale set of the inferred reasoning policy is generated through rationale reconstruction loss based on a graph extracted through a rationale and a knowledge graph retriever function.

### Claim 5 (depends on 4)

The policy generating method for SLM of claim 4,
wherein the rationale reconstruction loss is an equation
L
R
⁢
t
⁢
n
=
𝔼
(
o
,
h
,
R
)
∼
D
R
⁢
t
⁢
n
[
∑
i
=
1
m
⁢
log
⁢
Φ
R
(
r
i
|
g
)
]
(where, LRtn rationale reconstruction loss, o: observation, h: task description, R:
rationale set, g: graph extracted through knowledge graph retriever function).

### Claim 6 (depends on 1)

The policy generating method for SLM of claim 1,
wherein the embodied knowledge graph is a prompted knowledge graph in the learning a reasoning policy.

### Claim 7 (depends on 6)

The policy generating method for SLM of claim 6,
wherein the prompted knowledge graph is based on a batch sample including a positive pair, which is an embodied knowledge graph executing the same plan, and a negative pair, which is a continuous planning step.

### Claim 8 (depends on 6)

The policy generating method for SLM of claim 6,
wherein the prompted knowledge graph is based on a contrastive learning loss, and the contrastive learning loss is an equation
L
C
⁢
o
⁢
n
=
𝔼
B
C
⁢
o
⁢
n
∼
D
R
⁢
t
⁢
n
[
max
⁢
{
0
,
d
⁡
(
z
ˆ
,
z
ˆ
+
)
-
d
⁡
(
z
ˆ
,
z
ˆ
-
)
+
ϵ
}
]
(where, LCon: contrastive learning loss, BCon: batch sample, {circumflex over (z)}: embedding space, d: sum of distance metrics corresponding to elements of each rationale embedding sequence within embedding space {circumflex over (z)}eεZ, and ϵ: margin parameter).

### Claim 9 (depends on 1)

The policy generating method for SLM of claim 1,
wherein the planning policy is learned by predicting a next plan (a) based on the rationale set of the learned reasoning policy in the learning a planning policy, and learned through an equation
Φ
P
=
(
R
=
Φ
R
(
g
)
)
→
a
(where, ΦP: planning policy, R: rationale set of reasoning policy, ΦR: reasoning policy, g: graph extracted through knowledge graph retriever function, a: next plan).

### Claim 10 (depends on 1)

The policy generating method for SLM of claim 1,
wherein the planning policy reconstruction loss is an equation
L
P
⁢
l
⁢
a
⁢
n
=
𝔼
(
o
,
h
,
R
)
∼
D
R
⁢
t
⁢
n
,
R
∼
Φ
R
[
log
⁢
Φ
P
(
a
|
R
)
]
(where, LPlan: planning policy reconstruction loss, o: observation, h: task description, R: rationale set, ΦP: planning policy, a: next plan).

### Claim 11

A policy generating apparatus for SLM, the apparatus comprising:
an input/output interface configured to receive an expert dataset;
a processor configured to generate an SLM policy based on the expert dataset; and
a communicator configured to transmit the generated SLM policy to a terminal,
wherein the processor is configured to generate a rationale dataset based on the expert dataset and a pre-stored initial rationale set, verify the rationale dataset through a self-verification function; learn a reasoning policy through an embodied knowledge graph based on the verified rationale dataset; learn a planning policy based on a rationale set of the learned reasoning policy and a planning policy reconstruction loss; and generate an SLM policy including a final reasoning policy and a planning policy based on the learned reasoning policy and the planning policy.

### Claim 12 (depends on 11)

The policy generating apparatus for SLM of claim 11,
wherein the processor is further configured to, when learning the planning policy, generate the rationale set of the inferred reasoning policy based on a policy of an encoder based on an encoder prompt pool, a policy of a decoder based on a decoder prompt pool, and an attention module while performing the learning of the reasoning policy, and
wherein the encoder prompt pool comprises a prefix prompt and a postfix prompt.

### Claim 13 (depends on 12)

The policy generating apparatus for SLM of claim 12,
wherein the processor is further configured to generate the rationale set of the inferred reasoning policy through rationale reconstruction loss based on a graph extracted through a rationale and a knowledge graph retriever function.

### Claim 14 (depends on 13)

The policy generating apparatus for SLM of claim 13,
wherein the rationale reconstruction loss is an equation
L
R
⁢
t
⁢
n
=
𝔼
(
o
,
h
,
R
)
∼
D
R
⁢
t
⁢
n
[
∑
i
=
1
m
⁢
log
⁢
Φ
R
(
r
i
|
g
)
]
(where, LRtn rationale reconstruction loss, o: observation, h: task description, R: rationale set, g: graph extracted through knowledge graph retriever function).

### Claim 15 (depends on 11)

The policy generating apparatus for SLM of claim 11, wherein the embodied knowledge graph is a prompted knowledge graph in the learning a reasoning policy.

### Claim 16 (depends on 15)

The policy generating apparatus for SLM of claim 15,
wherein the prompted knowledge graph is based on a batch sample including a positive pair, which is an embodied knowledge graph executing the same plan, and a negative pair, which is a continuous planning step.

### Claim 17 (depends on 15)

The policy generating apparatus for SLM of claim 15,
wherein the prompted knowledge graph is based on a contrastive learning loss, and the contrastive learning loss is an equation
L
C
⁢
o
⁢
n
=
𝔼
B
C
⁢
o
⁢
n
∼
D
R
⁢
t
⁢
n
[
max
⁢
{
0
,
d
⁡
(
z
ˆ
,
z
ˆ
+
)
-
d
⁡
(
z
ˆ
,
z
ˆ
-
)
+
ϵ
}
]
(where, LCon: contrastive learning loss, BCon: batch sample, {circumflex over (z)}: embedding space, d: sum of distance metrics corresponding to elements of each rationale embedding sequence within embedding space {circumflex over (z)}εZ, and ϵ: margin parameter).

### Claim 18 (depends on 11)

The policy generating apparatus for SLM of claim 11,
wherein the planning policy is learned by predicting a next plan (a) based on the rationale set of the learned reasoning policy in the learning a planning policy, and learned through an equation
Φ
P
=
(
R
=
Φ
R
(
g
)
)
→
a
(where, ΦP: planning policy, R: rationale set of reasoning policy, ΦR: reasoning policy, g: graph extracted through knowledge graph retriever function, a: next plan).

### Claim 19 (depends on 11)

The policy generating apparatus for an SLM of claim 11,
wherein the planning policy reconstruction loss is an equation
L
P
⁢
l
⁢
a
⁢
n
=
𝔼
(
o
,
h
,
R
)
∼
D
R
⁢
t
⁢
n
,
R
∼
Φ
R
[
log
⁢
Φ
P
(
a
|
R
)
]
(where, LPlan: planning policy reconstruction loss, o: observation, h: task description, R: rationale set, ΦP: planning policy, a: next plan).

### Claim 20

A central server for generating a policy for an SLM, the central server comprising:
an input/output interface configured to receive an expert dataset;
a processor configured to generate an SLM policy based on the expert dataset; and
a communicator configured to transmit the generated SLM policy to a terminal,
wherein the processor is configured to generate a rationale dataset based on the expert dataset and a pre-stored initial rationale set, verify the rationale dataset through a self-verification function; learn a reasoning policy through an embodied knowledge graph based on the verified rationale dataset; learn a planning policy based on a rationale set of the learned reasoning policy and a planning policy reconstruction loss; and generate an SLM policy including a final reasoning policy and a planning policy based on the learned reasoning policy and the planning policy.
