Design and Implementation of an Effective Hardware for the ML Image classifiers

2021 ELE Engineering Design Project (FM05)

---

Faculty Lab Coordinator

Farah Mohammadi

Topic Category

Microelectronics

Preamble

In this project, we aim to study the hardware implementation of the machine learning (ML) for image classification. The VHDL application of the classifier may be carried out with the Vivado HLS interpreter. There are four classification algorithm options: linear regression, Bayesian learning, multilayer perceptron neural networks, and support vector machine. When it comes to embedded system, the accuracy of any method isn't the only consideration; area, power, and timing cost of classifiers are all important considerations when choosing a cost-effective classifier. They may also take a long time to locate the correct setup due to their complexities. For the best forecasting method, the ML algorithm with high accuracy, low area, low power consumption, and reduced latency is one of the available option.

Objective

To design, develop, and implement a simple ML model using VHDL simulator to analyze the effective design with high prediction accuracy.

Partial Specifications

-Architecting a machine learning image classifier model.
- Execution of ML workloads on the proposed classifier architecture.
- Comparing the performance of all learning algorithms under study based on accuracy, area, and power outperforming

Suggested Approach

1- Developing a ML platform as tensor-flow, MATLAB to train ML classification systems.
2- Using the Vivado Design Suite platform to create an HDL execution of the learned classifier.
3- Using hardware architecture on XILINX System on Chip as a platform for implement ML models.

Group Responsibilities

1. Perform Literature review.
2. Collect and design the technical information to implement the system under consideration.
3. Implement and train ML techniques for image classification with the above objective.
4. Write a technical report and present the obtained results at the end of this project. Describing the main step of running Vivado Design Suite compiler and showing the main simulations results.

Student A Responsibilities

• Build the image classifier models: Training the classifier for different ML approaches.
• Generating Training Data, feature extraction and running the learned models.

Student B Responsibilities

- Simulate the learned models to find out the Optimal Configurations and prepare them for hardware Implementation.

Student C Responsibilities

- Hardware implementation using System Generator tool for Simulink and Xilinx ISE: Build the learned ML model, use Vivado HLS simulator to build the HDL for the proposed models.

Student D Responsibilities

• Collect simulation results from Vivado Design Suite then evaluate obtained results for Power, Performance, and Area of the integrated VHDL implementation and select appropriate platform.
• Comparing between hardware implementation based on accuracy per area ratio for all learning models.

Course Co-requisites

Programming in C++, Microprocessors, Python, MATLAB.

To ALL EDP Students

Due to COVID-19 pandemic, in the event University is not open for in-class/in-lab activities during the Winter term, your EDP topic specifications, requirements, implementations, and assessment methods will be adjusted by your FLCs at their discretion.