# US 2023/0058055 A1 · Methods and systems for integrating machine learning/analytics accelerators and relational database systems

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

- Number: US 2023/0058055 A1
- Document ID: US20230058055A1
- Status: Published application
- Kind: A1
- Application number: 18/045,940
- Filed: 2022-10-12
- Published: 2023-02-23
- Earliest claimed: 2018-03-15
- Assignees: Georgia Tech Research Corporation
- Applicants: Georgia Tech Research Corporation
- Inventors: Hadi Esmaeilzadeh; Divya Mahajan; Joon Kyung Kim
- Web page: https://patentgrep.com/patent/US20230058055A1

## CPC

- G06F 16/24569: {Query processing with adaptation to specific hardware, e.g. adapted for using GPUs or SSDs}
- G06N 20/00: Machine learning
- G06F 9/5011: {the resources being hardware resources other than CPUs, Servers and Terminals}
- G06F 9/5038: {considering the execution order of a plurality of tasks, e.g. taking priority or time dependency constraints into consideration (scheduling strategies G06F9/4881 and subgroups)}
- G06F 16/2448: {for particular applications; for extensibility, e.g. user defined types}
- G06F 16/9024: {Graphs; Linked lists (G06F16/9027 takes precedence)}
- G06N 5/02: Knowledge representation; Symbolic representation
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

A method for database management that includes receiving an algorithm from a user. Based on the algorithm, a hierarchical dataflow graph (hDFG) may be generated. The method may further include generating an architecture for a chip based on the hDFG. The architecture for a chip may retrieve a data table from a database. The data table may be associated with the architecture for a chip. Finally, the algorithm may be executed against the data table, such that an action included in the algorithm is performed.

## Claims

### Claim 1

A method comprising:
determining, based at least in part on a dataflow graph, a chronological order of one or more operations needed to perform an action against a database;
generating an architecture for a chip based on at least one of the dataflow graph and the chronological order of one or more operations, wherein the architecture for a chip includes instructions based on the chronological order of one or more operations;
retrieving, by the architecture for a chip and from the database, a data table;
associating the data table with the architecture for a chip; and
executing an algorithm, designed to perform the action against the database, against the architecture for a chip such that the action is performed against the data table.

### Claim 2 (depends on 1)

The method of claim 1, wherein the dataflow graph is a hierarchical dataflow graph (hDFG).

### Claim 3 (depends on 1)

The method of claim 1, wherein the algorithm is a machine learning algorithm.

### Claim 4 (depends on 1)

The method of claim 1 further comprising receiving user defined functions associated with the algorithm.

### Claim 5 (depends on 2)

The method of claim 2, wherein the hDFG includes nodes and edges; and
wherein each node comprises a respective mathematical operation.

### Claim 6 (depends on 2)

The method of claim 2, wherein the hDFG includes nodes and edges; and
wherein each edge comprises a respective multi-dimensional vector.

### Claim 7 (depends on 4)

The method of claim 4 further comprising determining, based on, at least in part, the user defined functions, the one or more operations needed to perform the action against the database.

### Claim 8

A method comprising:
generating, based on an algorithm including operations needed to perform an action against a database, a hierarchical dataflow graph (hDFG);
generating, based on the hDFG, an architecture for a chip;
retrieving, by the architecture for a chip and from the database, a data table;
associating the data table with the architecture for a chip; and
executing the algorithm against the architecture for a chip such that the action is performed against the data table.

### Claim 9 (depends on 8)

The method of claim 8 further comprising determining, based at least in part on the hDFG, a chronological order for the operations.

### Claim 10 (depends on 8)

The method of claim 8, wherein the algorithm is a machine learning algorithm.

### Claim 11 (depends on 8)

The method of claim 8, wherein the hDFG includes nodes representing the operations.

### Claim 12 (depends on 8)

The method of claim 8, wherein the architecture for a chip includes an access engine; and
wherein retrieving the data table is performed by the access engine.

### Claim 13 (depends on 8)

The method of claim 8, wherein the architecture for a chip includes an execution engine; and
wherein executing the algorithm is performed by the execution engine.

### Claim 14 (depends on 8)

The method of claim 8 further comprising:
receiving, from a user, the algorithm; and
determining, based at least in part on the hDFG, a chronological order for the operations;
wherein the algorithm is a machine learning algorithm;
wherein the hDFG includes nodes and edges, wherein each node represents the respective operation and a respective mathematical operation, and wherein each edge comprises a respective multi-dimensional vector;
wherein the architecture for a chip includes a first set of instructions, a second set of instructions, an access engine, and an execution engine;
wherein retrieving the data table is performed by the access engine; and
wherein executing the algorithm is performed by the execution engine.

### Claim 15 (depends on 8)

The method of claim 8 further comprising:
determining, based at least in part on the hDFG, a chronological order for the operations;
generating, based on the chronologically ordered operations, a first set of instructions and a second set of instructions;
compiling the first set of instructions and the second set of instructions into executable code;
generating, based on the first set of instructions, an access engine;
associating the access engine with the architecture for a chip;
generating, based on the second set of instructions, an execution engine; and
associating the execution engine with the architecture for a chip;
wherein retrieving the data table is performed by the access engine; and
wherein executing the algorithm is performed by the execution engine.

### Claim 16 (depends on 11)

The method of claim 11, wherein each node comprises a respective mathematical operation.

### Claim 17 (depends on 11)

The method of claim 11, wherein the hDFG further includes edges;
wherein each edge comprises a respective multi-dimensional vector.

### Claim 18 (depends on 1)

A system comprising:
one or more processors; and
memory in communication with the one or more processors and storing computer program code that, when executed by the one or more processors, is configured to cause the system to perform the method of claim 1.

### Claim 19 (depends on 8)

A system comprising:
one or more processors; and
memory in communication with the one or more processors and storing computer program code that, when executed by the one or more processors, is configured to cause the system to perform the method of claim 8.

### Claim 20 (depends on 14)

A system comprising:
one or more processors; and
memory in communication with the one or more processors and storing computer program code that, when executed by the one or more processors, is configured to cause the system to perform the method of claim 14.
