SDAA429A June   2026  – October 2026 MSPM0G5187

 

  1.   1
  2.   Abstract
  3.   Trademarks
  4. 1 Introduction
  5. 2 MSPM0G5187 with TinyEngine NPU
  6. 3 Edge AI Toolchains
    1. 3.1 TI Edge AI Studio
    2. 3.2 TI Tiny ML Tensorlab
    3. 3.3 TI Neural Network Compiler
  7. 4 Rapid Prototyping Through TI Edge AI Studio
    1. 4.1 Project Setup
    2. 4.2 Data Capture
      1. 4.2.1 Hardware Setup
      2. 4.2.2 Feature Overview
    3. 4.3 Data Import
      1. 4.3.1 Classification Task
      2. 4.3.2 Regression Task
      3. 4.3.3 Forecasting Task
      4. 4.3.4 Annotation List
    4. 4.4 Data Annotation
    5. 4.5 Model Training
      1. 4.5.1 Training Parameter
      2. 4.5.2 Toolchain Dependency
    6. 4.6 Performance Reports
    7. 4.7 Model Export
  8. 5 Advanced Development Through TI Tiny ML Tensorlab
    1. 5.1 Environment Setup
    2. 5.2 Data Acquisition
      1. 5.2.1 Data Retrieval
      2. 5.2.2 Data Validation
    3. 5.3 Model Architecture Iterative Refactoring
      1. 5.3.1 Baseline Implementation
      2. 5.3.2 Refactoring With PyTorch Built-in Modules
      3. 5.3.3 Data Pipeline Optimization
      4. 5.3.4 Transition to CNN Architecture
    4. 5.4 Model Export
    5. 5.5 Quantization-Aware Training
    6. 5.6 Model Compilation
  9. 6 Edge AI Application: Digit Recognition
    1. 6.1 LeNet-5 Variant CNN Model
    2. 6.2 NPU/CPU Performance Comparison
  10. 7 Edge AI Application: Waveform Classifier
    1. 7.1 Feature Extraction
    2. 7.2 Time-Series Classification Model
    3. 7.3 Model Memory Considerations
    4. 7.4 NPU/CPU Performance Comparison
  11. 8 Summary
  12. 9 References
  13. 10Revision History

4.5.1 Training Parameter

The training parameter configuration is divided into two primary stages: preprocessing parameters and training hyperparameters.

Preprocessing Parameters

Edge AI Studio supports two preprocessing workflows to prepare data for model training:

  • Simple Window: This method utilizes raw time-series data from the dataset using a sliding window mechanism. Users can configure the window size within a range of 0.01 to 1.0
  • Feature Extraction: This workflow performs advanced data transformation. Within the Feature Extraction module, users can select between two processing modes:
    • FE_RFFT_FIXED: Uses FFT to extract frequency-domain information. Supported FFT sizes must be powers of 2, and the output size is defined as "frame size / 2". The FFT output is subsequently aggregated into a specific number of bins as defined by the "Feature Size per Frame" parameter
    • Raw: Utilizes raw time-series data without frequency transformation

The following parameters are used to configure the Feature Extraction workflow. These settings are applicable to both FFT and Raw processing modes:

Table 4-4 Feature Extraction Parameters
ParameterDescription
Frame SizeSpecifies the number of input samples per processing frame.
Feature Size per FrameDefines the number of bins (or features) generated per frame.
Frames to ConcatenateSpecifies the number of bins from preceding frames to be retrieved from the buffer. This mechanism enables the model to capture temporal variations across the frequency spectrum.
ChannelsDefines the number of sensor channels to be processed simultaneously.

For a comprehensive implementation example, please refer to Section 7.1 (Waveform Classifier Application).

Training Hyperparameters

The training hyperparameters control the optimization process and the final model characteristics:
  • Epochs: Defines the total number of complete passes through the training dataset. Generally, increasing the number of epochs allows the model parameters to converge closer to the best design, though too many epochs can lead to overfitting.
  • Learning Rate: Determines the step size during parameter updates following gradient descent in each epoch. A higher learning rate can accelerate convergence toward the global optimum but can cause oscillations or instability. Conversely, a lower learning rate provides more stable convergence but requires more epochs and can risk getting trapped in local optima.
  • Batch Size: Specifies the number of training samples processed in a single iteration. Note that larger batch sizes increase the computational load and memory pressure on the host PC.
  • Optimization (Precision Mode): Defines the target precision of the trained model:
    • High Precision: Trains a full-precision floating-point model.
    • High Speed: Utilizes Quantization-Aware Training (QAT) to produce a quantized fixed-point model optimized for high-speed inference.