# US 2026/0073221 A1 · Method and device for model compression > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 2026/0073221 A1 - Document ID: US20260073221A1 - Status: Published application - Kind: A1 - Application number: 19/218,779 - Filed: 2025-05-27 - Published: 2026-03-12 - Earliest claimed: 2024-09-10 - Assignees: Hyundai Motor Company; Kia Corporation - Applicants: Hyundai Motor Company; Kia Corporation - Inventors: Keonho Lee; Moonsub Jin; Jingang Huh - Web page: https://patentgrep.com/patent/US20260073221A1 ## CPC - G06N 3/082: modifying the architecture, e.g. adding, deleting or silencing nodes or connections - G06F 16/9017: {using directory or table look-up (use of a directory or look-up table in file systems G06F16/13)} - G06N 3/0495: Quantised networks; Sparse networks; Compressed networks - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) ## Abstract In an embodiment a method includes receiving a deep learning model loaded on a target device and a computation list supported by a model converter for the target device, determining whether a computation for each layer of the deep learning model is able to be accelerated on the target device based at least in part on the computation list, acquiring a compressed model by maintaining a computation corresponding to one layer as it is, and applying a compression technique to the deep learning model when the one layer of the deep learning model is determined to include a computation unable to be accelerated on the target device, and acquiring the compressed model by changing the computation corresponding to the one layer to another computation, and then applying the compression technique to the deep learning model when the one layer is determined to include the computation able to be accelerated on the target device. ## Claims ### Claim 1 A method comprising: receiving a deep learning model loaded on a target device and a computation list supported by a model converter for the target device; determining whether a computation for each layer of the deep learning model is able to be accelerated on the target device based at least in part on the computation list; acquiring a compressed model by maintaining a computation corresponding to one layer as it is, and applying a compression technique to the deep learning model when the one layer of the deep learning model is determined to include a computation unable to be accelerated on the target device; acquiring the compressed model by changing the computation corresponding to the one layer to another computation, and then applying the compression technique to the deep learning model when the one layer is determined to include the computation able to be accelerated on the target device; and adding a model component of the compressed model to a compressed model component database. ### Claim 2 (depends on 1) The method of claim 1, wherein the compressed model component database includes a database for managing the model component of the compressed model by using information on a model name, a model size, the number of model computations, a model inference time, a model definition, and its weight. ### Claim 3 (depends on 1) The method of claim 1, wherein the compression technique includes a pruning technique for removing weights or neurons from the deep learning model or a depth compression technique for reducing the number of layers in the deep learning model. ### Claim 4 (depends on 1) The method of claim 1, further comprising: evaluating the compressed model by using data used for pre-learning and actual collected data; adding performance information to a performance lookup table database when performance of the compressed model exceeds a predefined performance indicator; and removing the model component of the compressed model from the compressed model component database when the performance of the compressed model does not exceed the predefined performance indicator. ### Claim 5 (depends on 4) The method of claim 4, wherein the performance lookup table database is a database for managing the performance information by using information on a model name, model performance, agreement with an original model, and evaluation metrics. ### Claim 6 (depends on 5) The method of claim 5, wherein the evaluation metrics are computed based at least in part on a computational approach corresponding to Equation 1 below: Evaluation ⁢ metric = { α ⁢ ⁢ A ⁢ ccuracy * ( 1 - α ) ⁢ Agreement } + IAR 2 , ( Equation ⁢ l ) wherein, Evaluation metric indicates the evaluation metrics, Accuracy indicates accuracy of data used for pre-learning, Agreement indicates prediction agreement with the actual collected data, IAR indicates an inference acceleration rate indicating an improved model inference speed, and a indicates a predetermined constant. ### Claim 7 (depends on 6) The method of claim 6, wherein the prediction agreement is computed based at least in part on a computational approach corresponding to Equation 2 below: Average ⁢ top - 1 ⁢ Agreement = 1 n ⁢ ∑ i = 1 n ⁢ 𝕀 ⁢ { arg ⁢ max ⁢ σ j ( z t , i ) = argmax ⁢ σ j ( z s , i ) } , σ i ( z ) = exp ⁡ ( z i ) ∑ j exp ⁡ ( z j ) , ( Equation ⁢ 2 ) wherein, Average Top-1 Agreement indicates the prediction agreement, n indicates the number of evaluation data, I indicates a computation that compares these two values and outputs 1 when the values are the same and o when not, argmax indicates an index having a largest value in a list data structure, j indicates the number of classes classified by a total model, zt,i indicates a logit of a pre-trained original model, zs,i indicates a logit of the compressed model, and exp indicates an exponential function. ### Claim 8 (depends on 6) The method of claim 6, wherein the inference acceleration rate is computed based at least in part on a computational approach corresponding to Equation 3 below: IAR = Original ⁢ Model ⁢ inference ⁢ time - Compressed ⁢ Model ⁢ inference ⁢ time Original ⁢ Model ⁢ inference ⁢ time , ( Equation ⁢ 3 ) wherein, IAR indicates the inference acceleration rate, Original Model inference time indicates an inference time of a model before the compression, and Compressed Model inference time indicates the inference time of the compressed model. ### Claim 9 (depends on 4) The method of claim 4, further comprising: selecting the compressed model by using the performance lookup table database and the compressed model component database; and deploying the selected compressed model. ### Claim 10 (depends on 9) The method of claim 9, wherein selecting the compressed model comprises: performing adaptive batch normalization based at least in part on the actual collected data, performing sparse update based at least in part on a computational approach corresponding to Kullback-Leibler (KL) divergence, and updating the performance lookup table database. ### Claim 11 A device comprising: at least one processor; and a storage medium storing computer-readable instructions, wherein the instructions are executed by the at least one processor to cause the at least one processor to: receive a deep learning model loaded on a target device and a computation list supported by a model converter for the target device, determine whether a computation for each layer of the deep learning model is able to be accelerated on the target device based at least in part on the computation list, acquire a compressed model by maintaining a computation corresponding to one layer as it is, and applying a compression technique to the deep learning model when the one layer of the deep learning model is determined to include a computation unable to be accelerated on the target device, acquire the compressed model by changing the computation corresponding to the one layer to another computation, and then applying the compression technique to the deep learning model when the one layer is determined to include the computation able to be accelerated on the target device, and add a model component of the compressed model to a compressed model component database. ### Claim 12 (depends on 11) The device of claim 11, wherein the compressed model component database is a database for managing the model component of the compressed model by using information on a model name, a model size, the number of model computations, a model inference time, a model definition, and its weight. ### Claim 13 (depends on 11) The device of claim 11, wherein the compression technique includes a pruning technique for removing weights or neurons from the deep learning model or a depth compression technique for reducing the number of layers in the deep learning model. ### Claim 14 (depends on 11) The device of claim 11, wherein the instructions are executable by the at least one processor to cause the at least one processor to further: evaluate the compressed model by using data used for pre-learning and actual collected data, add performance information to a performance lookup table database when performance of the compressed model exceeds a predefined performance indicator, and remove the model component of the compressed model from the compressed model component database when the performance of the compressed model does not exceed the predefined performance indicator. ### Claim 15 (depends on 14) The device of claim 14, wherein the performance lookup table database is a database for managing the performance information by using information on a model name, model performance, agreement with an original model, and evaluation metrics. ### Claim 16 (depends on 15) The device of claim 15, wherein the evaluation metrics are computable based at least in part on a computational approach corresponding to Equation 1 below: Evaluation ⁢ metric = { α ⁢ ⁢ A ⁢ ccuracy * ( 1 - α ) ⁢ Agreement } + IAR 2 , ( Equation ⁢ l ) wherein Evaluation metric indicates the evaluation metrics, Accuracy indicates accuracy of the data used for pre-learning, Agreement indicates prediction agreement with the actual collected data, IAR indicates an inference acceleration rate indicating an improved model inference speed, and a indicates a predetermined constant. ### Claim 17 (depends on 16) The device of claim 16, wherein the prediction agreement is computable based at least in part on a computational approach corresponding to Equation 2 below: Average ⁢ top - 1 ⁢ Agreement = 1 n ⁢ ∑ i = 1 n ⁢ 𝕀 ⁢ { arg ⁢ max ⁢ σ j ( z t , i ) = argmax ⁢ σ j ( z s , i ) } , σ i ( z ) = exp ⁡ ( z i ) ∑ j exp ⁡ ( z j ) , ( Equation ⁢ 2 ) wherein, Average Top-1 Agreement indicates the prediction agreement, n indicates the number of evaluation data, I indicates a computation that compares these two values and outputs 1 when the values are the same and o when not, argmax indicates an index having a largest value in a list data structure, j indicates the number of classes classified by a total model, zt,i indicates a logit of a pre-trained original model, zs,i indicates a logit of the compressed model, and exp indicates an exponential function. ### Claim 18 (depends on 16) The device of claim 16, wherein the inference acceleration rate is computable based at least in part on a computational approach corresponding to Equation 3 below: IAR = Original ⁢ Model ⁢ inference ⁢ time - Compressed ⁢ Model ⁢ inference ⁢ time Original ⁢ Model ⁢ inference ⁢ time , ( Equation ⁢ 3 ) wherein, IAR indicates the inference acceleration rate, Original Model inference time indicates an inference time of a model before the compression, and Compressed Model inference time indicates the inference time of the compressed model. ### Claim 19 (depends on 14) The device of claim 14, wherein the instructions are executed by the at least one processor to cause the at least one processor to further: select the compressed model by using the performance lookup table database and the compressed model component database, and deploy the selected compressed model. ### Claim 20 (depends on 19) The device of claim 19, wherein the at least one processor is configured to select the compressed model by: performing adaptive batch normalization based at least in part on the actual collected data, performing sparse update based at least in part on a computational approach corresponding to Kullback-Leibler (KL) divergence, and updating the performance lookup table database.