Deep learning model optimization method and system based on high-level synthesis tool
A high-level synthesis and deep learning technology, applied in the field of deep learning model optimization based on high-level synthesis tools, can solve problems such as low flexibility, long development cycle, and inability to change, reducing hardware power consumption and improving throughput , the effect of shortening the delay
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0026] This embodiment proposes a deep learning model optimization method based on high-level synthesis tools. For the flow chart, please refer to figure 1 .
[0027] In the deep learning model optimization method based on high-level synthesis tools proposed in this embodiment, the following steps are included:
[0028] Step 1: Design a deep learning model based on the target function.
[0029] In this step, the input layer, convolution layer, pooling layer, output layer and other structures in the deep learning model are designed according to the target function points, and the parameters of the deep learning model are initialized.
[0030] Step 2: Obtain training samples, input the deep learning model for training, and obtain parameter weights of the deep learning model.
[0031] In this step, the training samples are obtained from the existing data set, and the test samples for verifying the deep learning model can be obtained at the same time.
[0032] Step 3: Expressin...
Embodiment 2
[0041] In this embodiment, the high-level synthesis tool-based deep learning model optimization method proposed in Embodiment 1 is applied to the optimization of the LeNet5 network.
[0042] First, the LeNet5 network model is designed according to the target function, and the existing LeNet5 network is modified accordingly. The structure diagram of the designed LeNet5 network model is as follows figure 2 shown. It includes input layer INPUT, convolutional layer C1, pooling layer S2, convolutional layer C3, pooling layer S4 and output layer OUTPUT.
[0043] Obtain 10,000 training samples from the MNIST data set and input them into the LeNet5 network model for training, and obtain the network parameter weights that complete the training. As shown in Table 1 below, it is the LeNet5 convolutional neural network parameters applicable to the MNIST data set in this embodiment.
[0044] Table 1. LeNet5 Convolutional Neural Network Parameters for MNIST Dataset
[0045]
[0046] F...
Embodiment 3
[0069] This embodiment proposes a deep learning model optimization system based on high-level synthesis tools. For the system architecture diagram, please refer to Figure 4 .
[0070] The deep learning model optimization system based on high-level synthesis tools proposed in this embodiment includes:
[0071] The deep learning model design module is used to design the deep learning model according to the target function;
[0072] A training module, configured to input training samples into the deep learning model for training to obtain parameter weights of the deep learning model;
[0073] A high-level language representation module, configured to represent the deep learning model in a high-level language according to the parameter weights of the deep learning model;
[0074] An optimization module, configured to optimize each layer of loops in the deep learning model;
[0075] High-level synthesis tools for co-simulating optimized deep learning models.
[0076] Among the...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More